Report for the ICCR-DRR project Designing index-based...
Transcript of Report for the ICCR-DRR project Designing index-based...
1
2
3
Report for the ICCR-DRR project 4
5
Designing index-based livestock insurance for 6
managing snow disaster risk in the central 7
Qinghai–Tibetan Plateau 8
9
10
State Key Laboratory of Earth Surface Process and Resource Ecology 11
Beijing Normal University 12
13
14
15
Mar 25, 2018 16
17
Please cite as: 18
Ye, T., Wu, J.D., Li, Y.J., Gao, Y. 2018. Designing index-based livestock insurance for 19
managing snow disaster risk in central Qinghai-Tibetan Plateau. Research report funded by 20
the International Center for Collaborative Research on Disaster Risk Reduction (ICCR-21
DRR). 22
23
Table of Contents 24
Table of Contents ................................................................................................................ 2 25
Executive summary .............................................................................................................. i 26
1 Project overview .......................................................................................................... 1 27
1.1 Objectives and tasks ........................................................................................................ 1 28
Objectives .................................................................................................................... 1 29
Workflow plan ............................................................................................................. 2 30
1.2 Research activities ........................................................................................................... 3 31
1.3 Outputs and deliverables ................................................................................................ 5 32
2 Understanding livestock snow disasters in the Qinghai–Tibetan Plateau ................... 7 33
2.1 Livestock snow disasters in the Qinghai–Tibetan Plateau ............................................... 7 34
2.2 Fieldwork ........................................................................................................................ 9 35
Characteristics of snow disasters ................................................................................. 9 36
Suggestions for selecting an insurance index ................................................................ 9 37
Local government preparedness for snow disasters ................................................... 10 38
Local herdsman’s preparedness for snow disasters .................................................... 10 39
2.3 Data collection .............................................................................................................. 11 40
2.4 Snow disaster loss mechanism in the study area........................................................... 12 41
Serious lack of infrastructure ..................................................................................... 12 42
Insufficient preparedness at the household level ....................................................... 12 43
High exposure due to a low pre-winter slaughter rate ................................................ 13 44
3 Livestock snow hazard analysis ................................................................................. 15 45
3.1 Review and selection of a snow hazard index ............................................................... 15 46
3.2 Hazard assessment ........................................................................................................ 19 47
Data and method ....................................................................................................... 19 48
Assessment results .................................................................................................... 21 49
4 Livestock snow disaster vulnerability analysis ........................................................... 23 50
4.1 Semi-quantitative results based on survey data ............................................................ 23 51
4.2 Quantitative results based on historical loss data ......................................................... 23 52
Factor and Data ......................................................................................................... 24 53
Methods .................................................................................................................... 26 54
Results ....................................................................................................................... 29 55
5 Design of an LSII ......................................................................................................... 34 56
5.1 Product design............................................................................................................... 34 57
Basic coverage ........................................................................................................... 34 58
Catastrophic coverage ............................................................................................... 34 59
5.2 Example of an indemnity calculation............................................................................. 35 60
Data ........................................................................................................................... 36 61
Derivation of %area snow cover ................................................................................. 36 62
Calculation of the snow disaster index ....................................................................... 37 63
Calculation of the insurance payment ........................................................................ 39 64
5.3 Premium rate making .................................................................................................... 40 65
Insurance loss risk assessment ................................................................................... 40 66
Premium rating results............................................................................................... 42 67
6 From report to policy ................................................................................................. 45 68
6.1 Involving local communities: pilot insurance programs ................................................ 45 69
Workshop/campaign design....................................................................................... 46 70
Workshop/campaign findings .................................................................................... 47 71
6.2 Involving local governments: preparation and launching of the product ...................... 52 72
7 Discussion .................................................................................................................. 55 73
7.1 Future work ................................................................................................................... 55 74
7.2 Suggestions for implementation ................................................................................... 56 75
7.3 Potential impacts .......................................................................................................... 57 76
Appendix 1 Photographs of fieldwork ............................................................................... 58 77
Appendix 2 Questionnaire of community response to livestock snow disaster ................ 60 78
Appendix 3 Information about the reviewing conference in Lhasa ................................... 71 79
80
i
81
Executive summary 82
The State Key Laboratory of Earth Surface Process and Resource Ecology, Beijing 83
Normal University (henceforth “ESPRE”), has committed to finishing a consultancy service 84
offered by International Center for Collaborative Research on Disaster Risk Reduction 85
(henceforth “ICCR-DRR”), with respect to realization of activities by ESPRE in the project 86
“Designing index-based livestock insurance for managing snow disaster risk in the central 87
Qinghai-Tibetan Plateau” (henceforth “the project”), as specified in the technical proposal. 88
According to the technical proposal, the project sought to design and develop a 89
livestock index-based insurance against snow disaster in the study area, to carry out a small 90
pilot project to test the performance of the product, and to gain experiences and lessons for 91
further sharing and replication. Specified outputs according to the technical proposal include 92
one research report, and one academic article published in an international peer-reviewed 93
SCI/SSCI-indexed journal. 94
ESPRE organized a research team to carry out the designated research activities: it 95
consisted of two associate professors and three graduate students from ESPRE, one research 96
fellow from Indonesia, and the agricultural insurance team of the largest local insurance 97
company in the study area. The following research activities were conducted: 98
(1) The team carried out two rounds of fieldwork to better understand the livestock 99
snow disaster mechanism in the study area, and to estimate the overall acceptance of local 100
herders to the index-based insurance product proposed. 101
(2) Based on the understanding gained, the team carried out a review of available snow 102
hazard indices (e.g., snow cover, snow depth, and duration of snow). 103
(3) The team carried out a vulnerability analysis based on two databases and two 104
approaches accordingly: one used empirical results based on interview information gathered 105
during fieldwork, while the other used historical data applied in generalized additive models. 106
(4) Based on the hazard and vulnerability analyses, the team suggested using the 107
duration of heavy snow cover—when the snow cover to grassland ratio exceeds certain 108
thresholds—as the livestock snow index for product’s design. Then, the insurance payment 109
scheme according to the selected snow index was designed, including the trigger, payment 110
ii
function, and deductibles. Based on the payment scheme and hazard assessment results, a 111
probabilistic risk assessment was carried out to calculate the insurance loss risks and 112
actuarially fair premium rates. 113
(5) The team worked hard to promote government stakeholder involvement to move 114
forward. Government departments, at various levels, reviewed the proposed index-based 115
product in two rounds. The product is currently waiting for its final approval; this should 116
occur once the government agrees to provide the sufficient premium subsidy. 117
Research activities carried out and reported here strictly followed those committed to in 118
the technical proposal. Two major deliverables were supplied: one summary report (this very 119
report), and an international journal paper now published 120
(https://doi.org/10.1016/j.scitotenv.2017.12.230). 121
122
1
123
124
1 Project overview 125
1.1 Objectives and tasks 126
Objectives 127
The final aim is to design and develop a livestock index-based insurance against snow 128
disaster in the study area, and to carry out a small pilot project to test the proposed product’s 129
performance, and to gain experiences and lessons for further sharing and replication. The 130
major approaches to fulfill these goals shall follow the general framework of index-based 131
insurance design, including: 132
� Comprehensive regional study into the loss mechanism of local snow disasters. This 133
study involves a literature review of snow hazard analysis, and fieldwork entailing 134
workshops and household interviews. 135
� Probabilistic risk assessment and insurance pricing. This follows the state-of-the-art risk 136
modeling approach by analyzing hazards and vulnerability and risk metrics widely 137
employed by international risk modelers, reinsurers, and brokers. 138
� Remote sensing. This technique will be intensively used for retrieving historical snow 139
cover and snow-depth data based on available satellite imagery. 140
� Pilot insurance program, with workshops and outreach campaigns. This component 141
involves more practical work by actually selecting several towns to hold workshops and 142
campaigns to hypothetically “run” the program; to help the local herdsmen to 143
understand how the index insurance could help them in handling risks; and to identify 144
potential challenges in the actual implementation of the product. 145
146
2
Workflow plan 147
148
Figure 1 Workflow of activities 149
Table 1 Activity description 150
Activity/
duration
Contents Milestones
Fieldwork � Workshop held with local herdsmen representatives, insurance company representatives, and government officials; Interview of local herdsmen
� Analysis on the loss mechanism of livestock snow disasters in the study area
Workshop memo; Interview records/questionnaires
Snow hazard
assessment
� A review of available snow hazard indices (e.g., snow cover,, depth, and its duration), upon which several livestock snow indices will be based and
derived; � Analyzing the spatial-temporal patterns of the
suggested livestock snow indices, and deriving their probabilistic risk assessment
Comparative results of potential livestock snow hazard indices according to the criteria of an insurance
index Temporal trend, spatial pattern, and return-period analysis results
Vulnerability
analysis
� Construct the historical livestock snow disaster event
catalogued dataset; Link historical snow hazard intensity to event loss;
� Applying quantitative analysis using regression or machine-learning approaches to derive the quantitative vulnerability function
Quantitative vulnerability function
Design of a
livestock snow
index insurance
� Design the insurance payment scheme according to the selected snow index, including the trigger, payment function, and deductibles;
Risk assessment results and pricing results (by pixel, and township and county boundaries)
Field work
Loss mechanism analysis
Review/ selection of
potential snow indicesVulnerability analysisHistorical
snow data
Historical
loss data
Hazard assessment Payment scheme design
Scientific plan of the product
Insurance contract
Pilot program
education underwritingLoss
estimation
Insurance
payment
Summarizing and sharing
3
(LSII)
� Insurance loss risk assessment and premium rating
Pilot program
of implementing
the designed
product
� Translate the design into a viable insurance contract
� Select 2 4 high snow risk towns in Naqu county,
central Naqu district � Hold small workshops to teach local herdsmen how
to use the index product � Use snow disaster data of the 2015, 2016, and 2017
winter seasons to generate hypothetical insurance indemnities*
� Hold multi-stakeholder workshops to summarize the viability and challenges in product implementation
LSII contract;
Workshop memo; Interview records
Viability, performance, and challenges in implementation.
* As the project is to be ended by Sep 2017, it is not viable to wait for the 2018 winter to carry out 151
real pilot project. Consequently, we have decided to use hypothetical insurance indemnities to further help 152
local herdsmen to understand the insurance project, and to identify challenges in implementation. 153
1.2 Research activities 154
As of September 30, 2017, the project team had finished all the tasks as written in the 155
technical proposal and letter of agreement. Specifically, following activities were carried out: 156
� Fieldwork for understanding the livestock snow disaster mechanism: The team 157
visited local areas in Naqu District, central Tibetan Plateau, to carry out fieldwork on 158
June 2–9. This fieldwork included a series of workshops and household interviews, 159
from which the critical mechanisms of livestock snow disaster in the study area were 160
better understood. 161
� Snow hazard assessment: The team carried out a review of the available snow hazard 162
indices (e.g., snow cover, depth, and duration). The duration of heavy snow cover (i.e., 163
when the snow cover to grassland ratio exceeds certain thresholds) has been considered 164
as the final livestock snow index for the product design. On this basis, a probabilistic 165
hazard assessment for the suggested livestock snow index was carried out. Index values 166
(duration of snow cover) by return period in years (1/5a, 1/10a, and 1/20a) have been 167
calculated and mapped. 168
� Vulnerability analysis: The team carried out a vulnerability analysis based on two 169
databases and two approaches accordingly. On the one hand, based on the data 170
collected during the household interviews, an empirical vulnerability relationship 171
(livestock mortality vs. duration of heavy snow cover) was estimated. On the other hand, 172
based on a dataset of historical livestock snow disaster mortality, generalized additive 173
4
models (GAMs) were used to quantify its relationship to snow depth, heavy snow 174
duration, and other key weather variables. GAMs revealed that the time index, snow 175
disaster duration, wind speed, and summer vegetation are critical for explaining 176
livestock mortality. The model fitting results gave adjusted-R2 values of up to 79.4%, 177
and the prediction error appeared well controlled. These quantitative vulnerability 178
relationships will serve us well for both the loss prediction and index insurance product 179
design. 180
� Design of a livestock snow index insurance (LSII). Given the loss mechanism learned 181
through fieldwork, and the quantitative vulnerability relationship derived, the insurance 182
payment scheme according to the selected snow index has been designed, which 183
includes the trigger, payment function, and deductibles. Based on these, a probabilistic 184
risk assessment was carried out to calculate the insurance loss risks and the actuarially 185
fair premium rates. 186
� Pilot program implementing the designed product. The research team visited the 187
Naqu district again, during July 24–August 4, 2017, holding 11 small campaigns in five 188
counties and 11 towns. During these campaigns, the research team further queried 189
herdsmen about their snow disaster risk management strategies (including any 190
supplementary feeding and infrastructure improvement) and their perception of the 191
existing insurance program. Based on these actions, the team explained the newly-192
designed index-based insurance program to the herdsmen, showing them how the 193
insurance works with the pilot loss adjustment using the 2007/2008, 2015/2016 winter 194
snow data. They were then asked for their perspectives, comments, and suggestions 195
about the proposed index-based product. 196
� Promoting government involvement and approval of the new product. A review 197
meeting was organized by the Property and Casualty Insurance Company of China 198
(PICC) Tibet Branch on August 17, 2017. Tibet Autonomous Region (TAR) 199
government official representatives, including people from the TAR financial office, 200
Department of Finance, Bureau of Insurance Regulation and Inspection, Department 201
Agriculture and Animal Husbandry, Department of Civil Affairs, and TAR 202
Meteorological Administration, as well as Naqu Prefecture government official 203
representatives, including the Vice-Commissioner of the prefecture government, and 204
leading persons in the corresponding departments/bureaus at the prefecture level, were 205
5
all invited to review the LSII technical plan. The local government officials appreciated 206
the LSII technical plan presented. Review comments and suggestions were recorded, 207
which were used to modify the technical plan. By late October 2017, the PICC Tibet 208
Branch had received the review comments from the document-based second round 209
review, from all departments at the TAR government level and from the Naqu 210
Prefecture government level. 211
� At the time of this report’s submission, the technical plan of the LSII product awaits 212
final approval by the TAR government, before it can be turned into an actual insurance 213
policy. Once approved by the TAR government, the PICC Tibet Branch will finalize the 214
insurance policy and submit it to the China Insurance Regulation and Inspection 215
Commission (CIRC) for its definitive approval. Then, the product can be put into the 216
marketplace. 217
1.3 Outputs and deliverables 218
According to the technical proposal, several outputs are expected. The following 219
outcomes have been achieved before submitting this report: 220
� The design of a technically sound and practical implementable index-based insurance 221
product for use in the local area (100% finished; included in this report). 222
� Experience and lessons in implementing an index-based livestock insurance program 223
in less-developed regions via pilot projects in several grazing community (towns) in 224
the region (100% finished; contained in this report). 225
� One synthesis research report: On the design of LSII for the Qinghai–Tibetan Plateau 226
(100% finished; this report). 227
� One academic research paper published in an international peer-reviewed journal: 228
“Linking livestock snow disaster mortality and environmental stressors in the 229
Qinghai-Tibetan Plateau: Quantification based on generalized additive models” has 230
been published in Science of the Total Environment 231
(https://doi.org/10.1016/j.scitotenv.2017.12.230). 232
The final outcome is under review and consideration for approval; this will require more 233
time until a final decision is reached: 234
6
� A snow disaster index-based livestock insurance policy that is officially approved: 235
The insurance plan received very positive feedback during the review conference 236
held in August 2017, and in the second round review (document-based) performed in 237
October 2017. It now awaits final approval by the TAR government. 238
239
240
7
2 Understanding livestock snow disasters in the 241
Qinghai–Tibetan Plateau 242
2.1 Livestock snow disasters in the Qinghai–Tibetan Plateau 243
The Qinghai–Tibetan Plateau (QTP) is located in Southwest China (26º00’N 39º47’N244
73º19’E 104º47’E; Figure 2). It covers an area of 2.572 × 106 km2, accounting for 26.8% of 245
the total land area of China. The region consists of 210 counties from Tibet, Qinghai, 246
Sichuan, Gansu, Xinjiang, and Yunnan provinces/autonomous regions (Zhang et al., 2002). 247
The terrain declines from the northwest to southeast, with an average elevation of 4000–248
5000 m. The climate in this region is mainly continental, dominated by great diurnal but 249
small annual variations in air temperature. Summers are characterized by higher precipitation 250
whereas the winters are dry and cold with strong winds. 251
252 Figure 2 Geographical location, elevation, and major grass vegetation types of the QTP 253
The QTP is extremely rich in grassland resources. The land currently used for farming 254
and raising livestock covers approximately 1.63 × 106 km2, of which alpine grassland covers 255
an area of 1.57 × 106 km2 (Zhao et al., 2013). The rich grassland resources thus support one of 256
the largest animal husbandry production bases in China. In 2014, the QTP housed a total of 257
38.03 million livestock, of which 10.47 million were cattle and 26.47 million were sheep 258
8
(Qinghai Provincial SB, 2015; Tibet Autonomous Region SB, 2015). As such, this region 259
supports the livelihoods of approximately 2 million pastoralists and 3 million agro-260
pastoralists (Miller, 2005). Animal husbandry production in 2014 reached 23.85 billion 261
RMB yuan*, with 9.25 billion yuan from cattle and 7.4 billion yuan from sheep and goats. 262
For centuries, raising livestock has been the most important way to survive and make a 263
living for the local herder communities in the QTP. Farming livestock is vital for obtaining 264
the daily dairy and meat consumed by local herdsmen, rather than serving as a major income 265
source as in Inner Mongolia. Because nomadic or semi-nomadic ways of grazing have 266
succeeded for hundreds of years (Wang et al., 2014) they are difficult to change in a short 267
period of time, even with considerable funds and effort from the central government of 268
China (Wang et al., 2013). 269
Climate, vegetation, and local nomadic pastoralism together, however, make the QTP 270
one of the most at-risk regions to suffer from livestock snow disasters. Seasonally, local 271
snow disasters primarily occur from October to May (Pu et al., 2007). Once the pastures are 272
snow covered, livestock have little access to food. Hence, if no supplementary feed is 273
provided to them, livestock could quickly lose weight and die from starvation. Strong winds 274
and low temperatures could accelerate the animals’ loss of body temperature and fat due to 275
the lack of robust infrastructure (i.e., roofed and warm sheds or shelters) (Wu et al., 2007). 276
Beyond the winter environmental stress, deficit summer precipitation and grassland 277
productivity could have a lag effect on livestock’s resistance to winter stress. For example, a 278
dry summer could substantially increase the risk of mortality due to insufficient nutrition and 279
the weakened body conditions of affected animals (Wang et al., 2016). 280
Snow disasters in this region are frequent. Major snow disasters have occurred 281
approximately once every ten years, while a medium disaster has occurred once every five 282
years. From 1956 to 1996, 11 snow disasters occurred in the Qinghai province that resulted 283
in the loss of 8.54 million livestock. A single catastrophic event in 1996 in the southern 284
Qinghai Province led to the loss of 1.08 million of its livestock, or approximately 40% of the 285
local herd size (Qia and Kong, 2007). In TAR, 35% of the livestock were lost during the 286
1967 snow disaster. The 1997/1998 snow disaster in Naqu, Tibet, left its herdsmen with 0.82 287
million livestock lost, with the herd size not restored for another 12 years (Wen, 2008b). 288
289
* 1 yuan = 0.151 USD as of October 24, 2017.
9
2.2 Fieldwork 290
To analyze in-depth the triggers and causes of snow disaster occurrences as they affect 291
animal husbandry, researchers went to investigate and interview the locals at the Naqu area, 292
during July 9-16, 2017. A small workshop was held at the Lhasa Branch of the People’s 293
Property and Casualty Insurance Company of China (PICC P&C), to kick-off the fieldwork 294
operations. Then, the team visited several counties in the Naqu District, , including the 295
counties of Naqu, Nierong, and Biru. Workshops were held with local government officials 296
from the county government office, the agriculture and animal husbandry bureau, the 297
meteorological bureau, along with insurance company managers from the local PICC P&C 298
branches. The focus of these workshops was to collect first-hand information about the 299
features of local snow disasters, associated historical livestock losses, disaster prevention 300
infrastructures, and emergency coping strategies during disaster occurrences. Household 301
interviews were organized in rural areas near the central towns of Naqu County and Nierong 302
County. The major focus in each interview was to collect information about the roofed sheds, 303
pre-winter hay and fodder storage, and in-disaster coping methods used at the household 304
level. The director of the division of agricultural insurance of the PICC P&C Tibet Branch, 305
and one research fellow from Tibetan Autonomous Region Meteorological Administration, 306
also participated in this field survey. 307
Characteristics of snow disasters 308
During the interviews, the meteorological administration of TAR shared that the major 309
characteristics of snow disasters affecting Tibetan animal husbandry was “fatten in autumn, 310
emancipate in winter, and die in spring”, which indicated an obvious lag effect. The 311
frequency of a snow disaster was once every five years (1/5a) for major catastrophes, while 312
1/3a for minor disasters. The historical snow disaster which caused the most serious losses 313
occurred in 1998. The construction of a snow disaster index for livestock in TAR ought to 314
adequately consider its geographical differences, local grass height, and grassland area when 315
constructing a snow disaster index by conducting field research. 316
Suggestions for selecting an insurance index 317
For now, the meteorological administration of the autonomous region uses the MODIS 318
snow-cover spatial dataset. But since it could be contaminated by clouds, supplemental data 319
will be needed to construct a robust index. Snow-depth data should be retrieved from the 320
10
meteorological stations’ actual empirical data. But since every county has no more than one 321
meteorological station, it is not advisable to represent the entire county by station data alone. 322
Additionally, the grassland in the Naqu area is an alpine meadow type, with a vegetation 323
height that is approximately 2–3 cm. Hence a medium-level snowfall event could cover most 324
of the meadow completely. To properly categorize the cover level requires great accuracy in 325
snow depth and grass height data, which is logistically too difficult to obtain. Therefore, 326
experts from the meteorological administration advised to take the snow-cover area as the 327
key index as the proxy of snow disaster impact. 328
Local government preparedness for snow disasters 329
From our communications with the local government, researchers learned of the 330
conditions of disaster-influenced animal husbandry industry and their ability to deal with 331
snow disasters. The regional natural disasters mainly included snow disasters, droughts, and 332
windstorms. In Nierong County, 5000 to 6000 livestock died from these disasters, but snow 333
disaster was the most serious of them. Subsidized by national policy, every herdsman 334
received 500 yuan for fodder, which was sold at 1.6 RMB per 0.5 kg; so the funds enabled 335
the purchase of 155 kg of concentrated fodder. In 2015, the Nierong County government 336
prepared 460 tons of concentrated fodder, but this amount could last only a couple of days 337
for the entire county’s livestock. Once a snow disaster happened, the livestock would freeze 338
or starve to death. With respect to infrastructure, the proportion of artificial rearing and 339
captivity was too low; even when aided by the national policy that provides a subsidy, the 340
land coverage of an alpine roofed-shed construction was insufficient. Against such a 341
background, the annual death rate of local livestock due to disasters was ~3% on average, 342
yet occasionally reaching high as 10% to 20%. 343
Local herdsman’s preparedness for snow disasters 344
Researchers also learned about the local herdsmen’s disaster prevention knowledge, hay 345
storage for disaster prevention, and the vulnerability of disaster-influenced livestock. Two 346
households with distinct economic conditions were interviewed. In general, local herdsmen 347
are sufficiently aware and conscious to prepare some winter-preserved fodder. But due to the 348
distinctions of economic status among different households, the low-income family did not 349
have an alpine roofed shed, nor could it afford to prepare any winter-preserved fodder—they 350
were left with waiting for governmental subsidies. The second family was richer, had some 351
roofed sheds covering a total land area of 150 m2, of which 60 m2 consisted of thermal sheds. 352
11
Their disaster-reserve fodder in winter could feed their livestock for 12 days, and they 353
maintained a better ability to deal with disasters. According to the herdsmen’s experience, 354
due to the vulnerability of disaster-influenced livestock, thinner cattle could live at most 5 355
days whereas sheep could live up to 3 days without fodder. 356
To sum up, snow disasters in the Naqu area occurred frequently, and seriously 357
influenced the livelihood of local herdsmen. The government was not sufficiently capable to 358
deal with such disasters. The little self-preserved fodder stored was not enough; so one snow 359
disaster was likely to impoverish the herdsmen. Considering the natural differences between 360
areas in the QTP, future research should adopt the snow-cover area rather than a superficial 361
index to indicate the severity of a snow disaster. Since the alpine thermal roofed shed did not 362
have high coverage, livestock would easily die of famine or frost without receiving fodder, 363
and this situation further varied depending on the different sizes of livestock held. 364
2.3 Data collection 365
Besides the basic geographical data, the project also used historical snow coverage 366
obtained from remote-sensing data, the meteorological station’s observation data on snow 367
depth, disaster records on yearbooks, and data from interviews conducted in the Naqu area, 368
Nierong County, Biru County, and Naqu County (Table 2). 369
370 Table 2 List of data collected 371
Data type Index and description Data origin
City and county administrative area map
Naqu area administrative division (vector map)
National geographic information center
Distribution of vegetation types
Vegetation types in the study area (distribution vector diagram)
Vegetation map of China (1:1 000 000) (Chinese vegetation map editor committee of the Chinese Academy of Sciences, 2007)
Daily snow cover IMS in the northern hemisphere daily snow cover (February 1992~ now); spatial
resolution based on sensor divided into 1
km, 4 km, and 24 km
US National Snow and Ice Center (http://nsidc.org/data/g02156
Re-analyzed data from the weather forecast system (NCEP)
Lattice data of snow cover 0.312°×0.312° (1979.1.1–2011.1.1)
National Centers for Environmental Prediction (NCEP) http://rda.ucar.edu/
Station-observed snow depth
Naqu Station and Bange Station observed
daily snow depths from November 1 to
April 30, in 1984–2014
TAR climate center
Field survey Mechanism of snow disaster and vulnerability of animal husbandry
Local workshops and household interviews
Historical snow disaster loss
Tibetan plateau snow disaster historical data for 2001–2016
China meteorological disasters’ catalogue - volume in Tibet; China meteorological disaster catalogue - volume in Qinghai
12
2.4 Snow disaster loss mechanism in the study area 372
Seasonally, local snow disasters occurred from October through May of next year, 373
though in some years the earliest snow disaster could happen in September or the latest one 374
in June. Following a snow disaster, the pasture is buried beneath snow, livestock would be 375
unable to gather food or be properly fed, consequently leading to the animals’ weight loss 376
and weakened immunity. On the one hand, a snow-buried pasture reduces the livestock’s 377
available intake of food; on the other hand, more snow increases makes it more difficult for 378
livestock to forage, walk, and gather needed food, thereby increasing their energy 379
consumption. Below-average temperatures can support long-lasting snow cover above the 380
ground. Furthermore, if the daytime maximum temperatures go higher than 0°C, the surface 381
snow can melt but freeze again in the winter to form a layer of ice crust that further 382
exacerbates the grave situation (Baojireji, 2014). Together, these factors would seriously 383
challenge the physical reserves of livestock and their health conditions. If not responded 384
properly and urgently, it can lead to severe losses. 385
The natural herbage nomadic style still dominated the animal husbandry practice in 386
Naqu area. Consequently, animal husbandry has been affected by weather conditions all year 387
round. The ensuing situation of “weather-dependent foraging” production differences drives 388
the following key problems associated with snow disasters. 389
Serious lack of infrastructure 390
Namely, alpine roofed thermal sheds for livestock are lacking. Although the national 391
government proposed to subsidize the building of alpine roofed thermal sheds, these sheds 392
remained a small proportion of all shed in the Naqu area. On the one hand, the national 393
subsidy of 12 000 yuan for construction was not enough to finish the standardized 394
construction of a single shed. On the other hand, a basic tenet of the animal husbandry 395
practice was to divide the cattle and sheep; hence, for herdsmen who owned both cattle and 396
sheep, their barns must be constructed separately, leading to higher costs. Finally, it was 397
difficult to find proper construction teams that could accomplish the work locally. The lack 398
of alpine barns greatly increased the local herds’ mortality rates. 399
Insufficient preparedness at the household level 400
The experiences from eastern Inner Mongolia demonstrate that well-fed livestock are 401
barely affected by snow disasters; the disaster could be mitigated, if there were an adequate 402
13
amount of fodder preserved for the grazing livestock. In recent years, the Naqu County 403
constructed a cold-season pasture fence, mowed and set aside a certain number of forage 404
grass areas, and strengthened the prevention of and resistance to snow disasters. However, 405
the government fiscal funds to build up forage reserves for winter and spring were still 406
limited when compared with the amount required for the livestock, and this food storage 407
could only maintain emergency needs for a few days (i.e., < 10 days). The local government 408
also issued policy documents that contained winter forage reserve requirements in addition 409
to a guide for how the herdsmen could strengthen their reserves. For now, only a small 410
minority of herdsmen from higher income families with a good awareness have actually 411
cultivated and purchased high land barley to increase the potential forage reserve; 412
nevertheless, this reserve was usually insufficient for livestock to survive 20 days following 413
a snow disaster event. 414
High exposure due to a low pre-winter slaughter rate 415
To Tibetan herdsmen, cattle, sheep, and other livestock are the most basic production 416
and living materials. In historical snow catastrophes, herdsmen used to go out of their way to 417
feed the livestock, even using those grains preserved for people. Although local governments 418
have propagated hardly, the slaughter rate before winter remained at ~5%, well below the 20% 419
recommended level. By not culling the weak and old livestock in herds, any snow disaster 420
was bound to increase the risk of livestock losses. 421
Given all the conditions mentioned above, the loss mechanism of snow disaster in Naqu 422
is very close to the traditional sense in the pastoral areas suffering from snow disaster, which 423
eventually led to the snow being buried for a long period that caused the livestock to freeze 424
or starve to death. From the viewpoint of regional disaster systems, there are multiple factors 425
influencing the final disaster loss incurred by herdsmen. 426
Through the local herdsmen household surveys and interviews, we further clarified the 427
mechanisms and formation process of loss: 428
(1) When there was continuous snow cover and the area ratio of it to grassland is 429
relatively low (less than 40–60%), the herdsmen would graze their livestock in better body 430
condition, and consume only those in poor condition. Since the herdsmen’s winter forage 431
reserve was relatively limited, supplementary feeding inputs were generally done only to 432
maintain basic vital signs of livestock, or about a sheep unit 2 kg/ hay. 433
14
With a shorter duration of snow cover (under 15 days) the deteriorating conditions of 434
cattle and sheep could be limited. Once the snow melted and grass is revealed, herdsmen 435
should immediately start to graze the livestock, so cattle and sheep would be less likely to 436
die. In such cases, losses borne by the herdsmen would only be the increased hay and fodder 437
costs used for supplementary feeding due to grazing precluded by the temporary heavy snow 438
cover. 439
(2) If the proportion of land covered by snow became even higher, trans-boundary 440
grazing was not viable, and long lasting snow duration will lead to extreme cases. Local 441
herdsmen have to keep all their livestock in sheds, and all cattle and sheep will be feed by 442
stored hay and fodder, but still using an amount that only maintains their vital signs. If the 443
sheds are without a roof or not the thermal-type (i.e., heated), low temperatures will quickly 444
consume a herd’s body fat, and the herds would eventually die. Herdsmen said cattle and 445
sheep of different body condition with or without supplementary fodder could survive for 446
different lengths of time (Table 2). 447
Table 3 Vulnerability of livestock in the Tibetan Naqu area 448
Note: Data in this table represents livestock in captivity without alpine thermal roofed sheds 449
450
Supplementary feeding condition
Livestock The longest surviving days according to the qualitative body condition before winter
Good Medium Bad
Supplementary feeding to sustain basic vital
conditions
Cattle/sheep <30 days <25 days <15 days
Without supplementary feeding
Cattle <15 days <7 days <5 days
Sheep <7 days <5 days <3 days
15
3 Livestock snow hazard analysis 451
3.1 Review and selection of a snow hazard index 452
From the harm mechanism and vulnerability of animal husbandry in the Naqu area, it is 453
known that the key element triggering the supplementary feeding condition and influencing 454
the cattle and sheep mortality rates was the snow coverage on pasture. As the snow is deeper 455
and the area covered is larger, the greater damage caused by the disaster, the necessity of 456
supplementary feeding became stronger, and the probability of death also greatly increased. 457
Therefore, three key factors contribute to building a snow disaster index: (1) the percentage 458
of grass height buried by snow (‘%Grass height buried’), (2) percentage of grass land area 459
covered by snow (‘%Grassland area covered’), and (3) the duration of snow disasters. 460
The existing literature provides important information on how snow cover impacts the 461
grazing behavior of various livestock. Local horses, sheep, and cattle have become used to 462
feeding on grass with blade heights of 20–30 cm, 10–20 cm, and < 10 cm, respectively. 463
Once the snow depth exceeds these heights, the corresponding livestock will have difficulty 464
finding food. The national standard on Snow Disaster Grades in Grazing Regions of China 465
(GB/T20482-2006) uses three indicators to represent the intensity of snow disasters. As 466
related to each grade of snow disaster, a semi-quantitative description of livestock mortality 467
is provided (Table 4). 468
Table 4 National standard from Snow Disaster Grades in Grazing Regions of China 469 (GB/T20482-2006) 470
Snow
disaster
grade
Indicators
Impact on livestock % Grass
height
covered by
snow
Duration of
snow cover
(days)
% Grassland
area covered
by snow
Small
0.30–0.40 ≥ 10
≥ 20%
Grazing by cattle affected; small and little
impact on sheep and horses, respectively;
number of deaths below 50 000 0.41–0.50 ≥ 7
Medium
0.41–0.50 ≥ 10
≥ 20%
Grazing by cattle and sheep affected; little
impact on horses; number of deaths
between 50 000 and 100 000 0.51–0.70 ≥ 7
Severe
0.51–0.70 ≥ 10
≥ 40%
Grazing by all livestock affected; large
losses claimed for cattle and sheep, with the
number of deaths between 100 000 and 200
000 0.71–0.90 ≥ 7
16
471
The local standard of the Inner Mongolia Autonomous Region employed a 472
comprehensive index. It was constructed using snow depth (cm), snow duration 473
(days), average pre-winter grass height (cm), and number of days that the daily 474
average temperature was below 0°C using a 5-day moving average, . The 475
grade of snow disaster is thus defined according to the range of the index (Table 5). 476
Table 5 Local standard for livestock meteorological disasters of the Inner Mongolia 477 Autonomous Region 478
479
Aimed at the three key factors of snow disaster index, the current China’s 480
meteorological departments and scientific research personnel have put forward some 481
beneficial proposals. The TAR Bureau of Meteorology proposed the standards for a snow 482
disaster index according to the local experiences, based on the national standards and Inner 483
Mongolia regional standards described above (Table 6). The suggested proposal basically 484
used the methods of national standards, but emphasized the differences between one-time 485
snowfall events and accumulated snowfall in determining the disaster grade. 486
487
488
489
490
SdsD
gH DD
s s
g
H DSd
H D
´=
-D
Extreme
0.71–0.90 ≥ 10
≥ 60%
Grazing by all livestock affected; large
numbers of livestock will die if not
protected, with the number of deaths
exceeding 200 000 > 0.90 ≥ 7
Snow disaster grade Snow disaster index Impacts on livestock
Small 0.31–0.50 Grazing by cattle affected; small and little impact on
sheep and horses, respectively
Medium 0.51–0.80 Grazing by cattle and sheep affected; little impact on
horses
Severe 0.81–1.30 Grazing by all livestock affected; animals lose weight;
some livestock die
Extreme ≥ 1.31 No conditions for grazing; large numbers of livestock die
17
Table 6 Suggested standards of the Naqu Bureau of Meteorology in Tibet 491
*Note The degree of snow-buried grass = the depth of snow coverage (cm)/the average height of pasture (cm), 492
to the nearest two digits after the decimal point. The degree of snow-covered area of pasture = snow-covered area 493
(ha)/useful pasture area (ha) 494
To summarize, the standards and indicators frequently mentioned in the literature 495
include snow depth, grass height, %-grass height covered by snow, %-grassland area 496
covered by snow, and the duration of snow cover. Two aspects must therefore be considered 497
in the construction of a snow-insurance index: 498
(1) Stress imposed by snow. It is important to first measure the stress imposed by snow on 499
livestock that feed on open-air grassland. Snow depth is the absolute physical measure of the 500
real-time snow hazard intensity. Snow cover thus provides an overview of how much 501
grassland has been covered by snow. Therefore, both indicators are superior to using snow 502
precipitation, since the actual snow cover that results on the ground includes the snow’s 503
redistribution by wind and landforms. 504
(2) Duration of the stress. This variable is important for measuring the cumulative stress 505
imposed on livestock during the winter season. Duration of snow stress leads to a degree of 506
starvation and loss from mortality. With small snow disasters, it is a direct measure of the 507
supplementary feeding input and, accordingly, a measure of herdsmen’s losses in terms of 508
feeding costs. With catastrophic snow disasters, it can be used to predict livestock mortality 509
and accordingly serves as a measure of herdsmen losses in terms of cattle and sheep. 510
In the research team’s earlier work, in the design of snow index insurance for the 511
Eastern Inner Mongolia Region, the snow cover (%height) had been used as the critical 512
Snow disaster
meteorological grade
index
Snow-cover condition Snow hazard harmful effect
%Grass
height
covered by
snow
Duration
of snow
cover
(days)
%Grassland
area covered
by snow
Medium disaster (one-
time snowfall)
0.51–0.70 ≥ 5 S ≥ 30% Mainly affected the feeding of
cattle and sheep, but with
diminutive effects on horses.
Medium disaster
(accumulated snowfall)
0.41–0.50 ≥ 10 S ≥ 30%
Severe disaster (one-
time snowfall)
0.71–0.90 ≥ 5 S ≥ 40% Seriously affected the feeding of
every kind of livestock, including
lowered body conditions, female
abortions, cub deaths, and if
improperly coped with, it would
cause many livestock deaths.
Severe disaster
(accumulated snowfall)
> 0.90 ≥ 10 S ≥ 40%
18
index, together with a measure of duration. However, given the actual situation in Naqu 513
Prefecture, as well as other regions in the QTP, we have decided to use the snow cover 514
(%area). The main reasons to do this are as follows: 515
(1) Local grass type and grass depth provided the possibility to simplify the index. 516
In terms of the vegetation structure of the Naqu pasture, the major grass types found in 517
the Naqu area were alpine meadow and scrub meadow. In the eastern part of Naqu, with its 518
shorter plants and better hydrothermal conditions, the grass was generally 3 to 4 cm tall, and 519
in some areas it had blade heights of 5 to 6 cm. In the middle and western areas, the grass 520
depth was even lower and shallower. At this grass depth level, a moderate snowfall could 521
cover entire plants. Meanwhile, under the precondition of a limited grass height, it requires 522
extreme precision to measure the percentage of snow-covered depths and their spatial 523
differences. Therefore, using the same approach as the eastern Inner Mongolia region is lack 524
of practical significance. 525
(2) The retrieval of snow-cover estimates by remote-sensing data is more accurate than 526
that of snow depth. 527
In the Naqu district, only four national meteorological stations are capable of observing 528
snow depth for the whole district, which covers 450 000 km2 and has 11 counties. Besides, 529
the terrain was complicated in the study area and local snowfall was common. Nevertheless, 530
the accumulated snowfall differed in places, even on the same mountain across 531
distinguishable slopes. Clearly, the low density of weather stations is far from enough to 532
provide robust snow depth data to predict the %height buried or snow duration for many 533
local areas. Increasing the number of stations to monitor snow depth is challenging because 534
it is expensive to do. Therefore, it is recommended that remote-sensing monitoring data be 535
adopted to calculate the index required for the insurance indemnity. 536
From the point of snow monitoring via remote sensing, the retrieval of snow cover has 537
better precision than that of snow depth. The existing literature shows that, in the case of 538
MODIS and its information on snow cover, the accuracy of the MODIS/Terra data was 539
related to snow’s duration and depth. If the snow cover persists continuously for three or 540
more days, mean estimation error was < 10%. When snow depth is >14 cm, the forecasting 541
accuracy can be as high as 100%. At a snow depth of 5 cm, the forecasting data accuracy is 542
75% (Pu et al., 2007). Research on the Tibetan Plateau’s snow-cover duration and snow 543
depth showed (Li et al., 2008) that snow-cover duration in the river source area of China was 544
19
frequently > 60 days, with an average snow depth of 7 cm. Therefore, MODIS can reliably 545
reflect the distributions and changes in snow cover in the river source area. As a result, a 546
wide range of snow-cover data monitoring can be retrieved from remote-sensing data, to 547
meet real-time, convenient, effective, and high-precision demands. 548
By contrast, the measuring accuracy of remote-sensing-retrieved snow-depth data was 549
relatively poorer. In our country, the Northwest Institute of Eco-Environment, Chinese 550
Academy of Sciences, have prepared, composed, and continuously upgraded The Dataset of 551
Snow Duration in China from 1978 to 2012, which is perhaps the best remote-sensing data-552
retrieved snow-depth dataset available for China. However, the retrieved results have an 553
absolute error > 5 cm (Wang Wei, 2014). For the pastures type in the Naqu area, this error is 554
enough to bury the meadow. As such, using this data is not recommended for the index that 555
calculates the basis of insurance claims. 556
To sum up, %Grass height buried is not significant and has poor precision. The degree 557
of snow-covered area and snowfall duration could, on the one hand, decrease the calculation 558
difficulty and thereby increase the transparency and friendliness of the product. On the other 559
hand, the advantages of high precision and high space-time resolution data reduce the basis 560
risk of an index insurance, which likely leads to the promotion of any insurance products. 561
Therefore, in this case, the Tibetan snow index insurance was achieved by using the %area 562
snow cover and the cumulative snow duration for determining the snow cover area extent. 563
3.2 Hazard assessment 564
Data and method 565
Gridded snow-cover data, for the period of Jan 1, 1979 to Jan 1, 2011, were obtained 566
from re-analyzed data from the weather forecast system of The National Centers for 567
Environmental Prediction (NCEP, http://rda.ucar.edu/). The data recorded the %area snow 568
cover (support: 0–100%) at each 0.312°×0.312° grid cell. 569
Given any pixels, or insurance unit (such as a township, county, city), under the premise 570
of obtaining %area snow cover, a given area ratio threshold can extract those dates that lie 571
close to the threshold for that corresponding winter season. Then, we define the days with 572
continuous snow cover that meet a certain threshold as the “snow cover” process. Once 573
the %area snow cover falls below this threshold, then one process effectively ends. On this 574
basis, the duration of each snow-cover process can be obtained from which the %area snow 575
20
cover can be derived. Then, two important indices can be built: (1) the maximum duration of 576
single snow-cover events in the snowing season, , which can serve as a good hazard 577
indicator for the livestock losses, and (2) the cumulative duration of snow disasters in the 578
snowing season, , which can serves as a good hazard indicator for losses related to the 579
costs of providing hay and fodder for supplementary feeding. 580
To facilitate the calculations, we define a %area snow-cover threshold as , and 581
then calculated for each season the single maximum duration and cumulative duration 582
in the Naqu area from 1984 to 2014. Then, the hazard assessment involves estimating 583
the probability distributions of both indices. As the value of is the annual maximum, 584
we can use the generalized extreme value distribution according to the annual maxima series 585
of the theory of extreme value. The generalized extreme value distribution of the probability 586
density function is given as follows: 587
(1) 588
Among its terms, are the location, scale, shape parameters, respectively. Based on a 589
random range, different variables can also be further divided into extreme value type I 590
(Gumbel) or type II (Frechet) or type III (Weibull) distributions. 591
For the accumulative duration , we use the non-parametric kernel density approach. 592
The underlying distribution of a random variable can be assembled from many kernel 593
densities centered at the samples : 594
(2) 595
596
In this function, is the kernel function, is the sample size, and is the window-597
width parameter. We used the Gaussian function and its corresponding 598
optimal window width, , in the fitting process, according to the “rule-of-599
thumb” on optimality. The value of takes the smaller value of the standard deviation and 600
the interquartile range divided by 1.34. 601
maxd
cumd
80%a =
maxd
cumd
maxd
( )
1 11
1; , , 1 exp 1
x xµ µµ s x x x
s s s
- - -ì üé - ù é - ùï ïæ ö æ ö
= + - +í ýç ÷ ç ÷ê ú ê úè ø è øë û ë ûï ïî þ
x xf x
, ,µ s x
cumd
x
{ }iX
( )1
,ih i
X Xf x K x R
nh h
-æ ö= Îç ÷
è øå!
( )K × n h
( )2/21
2
uK u e
p
=
1/51.06h ns
-=
s
21
Assessment results 602
With the above estimated probability distribution, a complete spatial distribution of 603
hazard intensity at different return periods was successfully derived. The 80th, 10th, and 20th 604
percentiles of the distribution, which correspond to the 1/5a, 1/10a, and 1/20a insurance loss 605
costs, respectively, were computed for each pixel to map the insurance loss risk (Figure 3, 606
Figure 4). 607
608
609
Figure 3 Spatial distribution of the maximum duration for single snow-cover process events 610
22
611
612
Figure 4 Spatial distribution of the accumulative duration for single snow-cover process events 613
614
The hazard assessment results indicate that both and present significant east-615
to-west differences. The southeast corner of the eastern three counties (Suo, Biru, and Jiali 616
counties) have better water vapor conditions, the best pastures, and vegetation cover degree, 617
but they also feature the longest duration of snow cover, in terms of both single maximum 618
and annual accumulative duration. The remaining parts of the eastern three counties and 619
northeastern counties, including Baqing County and Nierong County, belonged to the 620
second-highest risk region. Nima County and Shenzha County in the western Naqu district 621
have the least risk. 622
623
maxd
cumd
23
4 Livestock snow disaster vulnerability analysis 624
A vulnerability analysis was carried out, mainly to derive the quantitative relationship 625
between hazard intensity and disaster loss, based on which the insurance payment scheme 626
can be designed so that the indemnity reflects the actual loss as much as possible. Based on 627
our fieldwork and data we collected, two analyses were performed: (1) a semi-quantitative 628
estimation, based on the household survey data, and (2) a quantitative estimation, based on 629
the historical livestock loss data. 630
4.1 Semi-quantitative results based on survey data 631
According to the loss mechanism of snow disaster obtained from the household survey, 632
cattle and sheep can survive a snow disaster for a different number of days, and these limits 633
differ by livestock type and their pre-winter body condition. With the information provided 634
by the local herdsmen in Table 3, a simple yet semi-quantitative function may be fitted to the 635
data, assuming there could be supplementary feeding that maintains the minimum body 636
condition should there be heavy snow cover (first row in Table 3). We further assume that 637
the share of cattle/sheep with good, medium, and bad body conditions are 30%, 40%, and 638
30%, respectively. Then, a simple and straightforward approaches use the piecewise function: 639
640
641
Or we apply a logistic function that is widely used when describing environmental stress on 642
animal species: 643
644
4.2 Quantitative results based on historical loss data 645
Based on the collected history loss data, the Generalized Additive Models (GAMs) are 646
employed to quantitatively analyze the vulnerability of snow disaster in animal husbandry. 647
( )
,max
,max
,max
,max
,max
0, 15
0.2,15 25
0.5,25 30
0.70, 30
d
dd
d
d
a
a
aa
a
<ìï
£ <ïD = í
£ <ïï ³î
( ),max
,max
,max
,max
0, 15
1, 30
1 37.3333 0.8629d
d
dd
a
a
a
a
<ìï
D = í³ï
+ ×î
24
Factor and Data 648
Our purpose is to reveal the quantitative linkage between livestock loss and snow hazard intensity 649
together with other environmental stressors, under given exposure levels. The selection of candidate 650
factors was guided by a simple conceptual framework summarized from the existing literature (Figure5), 651
together with data availability. 652
653
Figure5 Conceptual framework of factors linking to livestock mortality in snow disaster 654
The key mechanism of livestock snow disaster is forage unavailability or inaccessibility 655
due to snow cover that leads to starvation of livestock (Fernández-Giménez et al., 2012). 656
Therefore, the direct measure of snow hazard intensity generally involves snow depth, snow 657
cover, and snow disaster duration (Li, et al., 1997; Li et al., 2006; Ye et al., 2017). Snow 658
water equivalency has also been used in some regions to present hazard intensity (Tachiiri et 659
al., 2008). In addition to snow hazard intensity, during disaster and pre-season (summer) 660
environmental stress are also believed to have crucial impact on livestock loss. During a 661
snow disaster, strong winds and low temperatures are critical stressors that increases 662
livestock fat consumption to keep body warm and therefore speed starving (Wu et al., 2007). 663
Summer vegetation has important lagged effects on winter snow disaster loss, as poor 664
summer vegetation can substantially diminish livestock body reserves, making them less 665
resistant to starvation and cold (Wang et al., 2014). Last but not the least, livestock exposure, 666
and other controlling variables, i.e. time trend and elevation, were also included. Given the 667
list of potential factors/predictors, following data have been collected. 668
1) Historical loss and exposure data 669
Historical livestock snow disaster loss data were obtained from various sources. Loss 670
records during 1961-2007 were obtained from Wang et al. (2013), which were based on 671
yearbooks of meteorological disasters up to 2008. Loss records during 2008-2015 were 672
obtained from the China Meteorological Science Data Sharing Service System (CMSDS, 673
http://data.cma.gov.cn). In total, 135 snow disaster events in 41 years were included in the 674
Pre-season stress
(vegetation)Livestock
Environmental stress Exposure
Livestock Mortality (rate)
During-disaster
stress
Temperature Wind
Snow hazard
intensity
Disaster
duration
Snow
cover
Snow
depthNDVI
Precipita-
tion
Herd
size
Time
trend
Controls
elevation
25
dataset. For each snow disaster event, the dataset records its start and end dates, counties 675
affected, number of livestock lost ( ) and houses damaged. The corresponding end-of-year 676
county-level herder size data ( ) were obtained from Qinghai Provincial Statistical 677
Yearbook and Tibet Autonomous Region Statistical Yearbook. For the years that the data are 678
absent, herd sizes were interpolated using linear function of the county level time series, or 679
the county-provincial herd size relationship. Based on these data, livestock mortality rate ( ) 680
for each event was calculated as the number of livestock lost divided by the end-year herd 681
size of the previous year. 682
2) Snow hazard data 683
Historical daily snow depth data (1979–2013) were obtained from the Cold and Arid 684
Regions Science Data Center at Lanzhou (http://card.westgis.ac.cn/). The snow-depth data 685
were retrieved from passive microwave remote-sensing data with a spatial resolution of 686
25×25 km, based on Chang’s algorithm (Chang et al., 1987), and calibrated for regions in 687
China with ground-level observed snow-depth data. The accuracy varies somewhat between 688
the results retrieved from SMMR (1978–1987) and SSM/I (post-1987). The two absolute 689
errors were less than 5 cm and hold about 65% of all the data. The standard deviations were 690
6.03 and 5.61 cm for SMMR and SSM/I, respectively (Che et al., 2008). Historical daily 691
snow cover data (percentage of snow covered land/ total land area; 1980-2013) were 692
aggregated from the six-hour data provided by the National Centers for Environmental 693
Prediction (NCEP, http://rda.ucar.edu/), with a spatial resolution of 0.312° × 0.312°. For 694
each county, daily snow depth and snow cover were first calculated from pixel-based data 695
using zonal statistics, and then the maximum, minimum, and mean value of snow depth (SD; 696
m) and snow cover (SC; %) in each snow disaster period, together with duration of each 697
disaster ( ), were prepared as variables for snow hazard intensity. 698
3) During-disaster environmental stressor data 699
Wind speed and temperature have been frequently considered as the indicators of 700
during-disaster environmental stress in the literature. Historical daily weather data, including 701
daily maximum, mean and minimum temperatures and daily maximum wind speed from 702
1961 to 2013 for 106 national reference stations in this region were obtained from CMSDS 703
(http://data.cma.gov.cn). For each historical snow disaster event, mean daily maximum wind 704
speed ( ; m/s), as well as the mean daily maximum, average, and minimum temperature 705
( , , , respectively; ℃) were derived from corresponding station-observed data. 706
4) Pre-season environmental stressor data 707
L
N
LR
Dur
v
maxT
meanT
minT
26
Variables for summer rainfall and vegetation deficit were also considered as suggested 708
by the literature to reflect the potential impact of summer drought on livestock body fat. We 709
have considered using the Normalized Difference Vegetation Index (NDVI) and growing 710
season precipitation data. Due to the time span of our study, the NDVI dataset consists of 711
two parts. NDVI 15-day maximum value composite (MVC) images with the spatial 712
resolution of 8 km from 1981-2006 were obtained from the Environmental and Ecological 713
Science Data Center for West China, National Natural Science Foundation of China 714
(http://westdc.westgis.ac.cn). The dataset was prepared with the algorithm of Tucker et al. 715
(1994). Monthly MODIS MVC NDVI data with spatial resolution of 500m from 2007-2015 716
were obtained from International Scientific & Technical Data Mirror Site, Computer 717
Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn). Due to 718
differences in sensor and spatial resolution, the two NDVI datasets were not directly 719
comparable. We adapted the MODIS-NDVI series to the AVHRR-NDVI series using the 720
empirical relationships proposed by Yu (2013) and Du et al. (2014). For each NDVI pixel, 721
the annual maximum value was first identified. Then, for each county, the median of all 722
grassland pixel-based annual maximum were then selected and used as the proxy variable of 723
vegetation growth. In addition to the NDVI data, forage growing season (May to Sep in the 724
QTP; Cong et al., 2017) cumulative precipitation (P; mm) were also considered as an 725
alternative measure of vegetation growth. Anomalies of NDVI and precipitation data series 726
during the period of 1979-2015 were also calculated and included in the analysis. This is due 727
to the large spatial difference in vegetation type and NDVI in the QTP. Precipitation in the 728
QTP shows an east-to-west decreasing gradient, and regions with higher precipitation show 729
more productive vegetation growth. However, these regions also experience more frequent 730
and heavier snows in winter. Using anomalies can help remove any possible confounding 731
effects due to coincidence in the spatial patterns of vegetation and snow disaster loss. 732
In addition to the data mentioned above, digital elevation data (DEMs; 90 m × 90 m) 733
were obtained from the NASA Shuttle Radar Topographic Mission (SRTM, 734
http://srtm.csi.cgiar.org/). Elevations (ELE) for county centroids were calculated as the 735
controlling variable for topography. 736
Methods 737
Given the consensus on non-linearity of the relationship, and our goal of prediction, 738
GAMs were used to quantify the relationship among livestock mortality, snow hazard 739
27
intensity, as well as other environmental stressors. GAMs are an extension of the generalized 740
linear model. GAMs relate the expected value to the explanatory variables using a set of 741
non-parametric functions, , in which is the link function, is the 742
expected value of the response variable, are a set of explanatory variables, and are 743
unspecified smooth functions. Using the s enable GAMs to be more flexible, 744
independent of the response on the explanatory variables without specifying any parametric 745
relationship. However, the challenge remains to determine the smooth functions and their 746
smoothness. 747
Livestock mortality ( ) and mortality rate ( ) were the variables that we intended to 748
estimate and predict. Given their non-negative supports, we have used their natural logarithm 749
as the response variables. The Q-Q plots, after taking the logarithm, indicate strong evidence 750
of normality. Consequently, identical link functions were employed. 751
Three groups of predictors were considered: 1) snow hazard intensity, including snow 752
disaster duration , snow depth variables ( , , and ), and snow cover 753
variables ( , , and ), 2) during disaster environmental stressors, including 754
wind speed and temperature ( , , and ), and 3) pre-season environmental stressors 755
concerning vegetation conditions from the previous summer, including annual maximum 756
NDVI, , and growing season cumulative precipitation , and their anomalies, 757
and , respectively. 758
Time trends, elevation, and herd size were considered as the controlling variables in 759
building models. A time index variable , which uses the year of the disaster, was 760
considered to remove any collinearity related to time, i.e. socioeconomic development, 761
change of herd size, the evolvement of infrastructure, and change in climate. Elevation, , 762
was considered to control for any elevation-related effects. Additionally, herd size was also 763
controlled in the model for mortality. 764
The fitting of GAMs starts with the selection of explanatory variables. We followed the 765
procedure of preliminary analysis suggested by Anderson et al. (2016). The Pearson 766
correlation for all variables was carried out, and highly correlated predictors (with 767
correlation coefficient ) were not entered into the model simultaneously so as to 768
minimize the multi-collinearity issue (Hjort et al., 2016). Multi-collinearity diagnostics were 769
also carried out to further check this issue. The diagnostics analysis was carried out with 770
multivariate linear regression, based on which variance-inflation factor (VIF) was estimated. 771
( ) ( )1
n
j jjg f xµ a
== +å ( )g × µ
jx ( )jf ×
( )jf ×
L LR
Durmax
SDmin
SDmean
SD
maxSC
minSC
meanSC
vmaxT
meanT
minT
NDVI Pa
NDVI
aP
t
ELE
0.7r >
28
Variables having a VIF>10 will be considered highly suspicious of multi-collinearity and not 772
be considered to enter the model. In addition, single variable GAMs were estimated to reveal 773
the preliminary relationship among response variables and predictors. 774
In order to find the most promising GAMs, various combinations of response variables 775
and predictors were considered. There were two fundamental models: 1) livestock mortality 776
can be predicted by hazard intensity and other environmental stressors, when time trends, 777
elevations and herd size are controlled, 2) livestock mortality rates can be predicted by 778
hazard intensity and other environmental stressors, when time trends and elevations are 779
controlled. For each of the fundamental model, we tried to identify the “best” model by 780
screening the variables of importance. We applied the multi-model inference method 781
(Burnham and Anderson, 2002) and carried out a data dredge analysis run for all valid 782
combinations of predictors and fit a set of competing models. The importance of each 783
predictor was obtained by adding the Akaike weights to the models in which that variable is 784
present (Burnham and Anderson, 2002). The addition of the weights of each variable can be 785
a good indicator of the relative importance of the variable (Taylor and Knight, 2003). Thus, 786
for each group of variables, the variable with the highest aggregate weights were attached as 787
a priority. 788
For the final selection of GAMs, we investigated two aspects according to the 789
suggestion of Anderson et al. (2016). On the one hand, the model needed to have a high 790
predictive power and a strong goodness-of-fit. A high predictive power will help in 791
applications of predicting loss, risk assessment, and adaptation decision-making. On the 792
other hand, all the variables should be “reasonable” predictors of mortality (rate). Some of 793
the models with high values of goodness-of-fit and degree-of-freedom may have over-fitting 794
issues, or some of the response relationships may not be supported by theory or prior 795
knowledge. Whether predictors are reasonable was verified by checking response curves. In 796
addition, for each of the fundamental model, we intend to include two models: the model 797
with the minimum number of predictors (the “minimum model”), and the model with at least 798
one predictor from each of the groups (the “full model”). If a model contained pre-season 799
environmental stressors, their verities of using NDVI and precipitation variables were both 800
included. NDVI is a more direct predictor of summer vegetation than precipitation, but 801
precipitation is a ready-to-use output from climate scenarios. 802
In our analysis, GAMs were fitted using the mgcv package of R 3.3.3, and the dredge 803
analyses were carried out using the MuMIn package of R 3.3.3. For each fitting, the pseudo 804
29
adjusted-R2 and the total deviance explained were calculated as indicators of goodness-of-fit. 805
Additionally, a 10-fold cross validation (CV) was carried out to test the prediction power of 806
the underlying model. Metrics of predictive errors were also recorded, including the root 807
mean square error (RMSE), the mean absolute error (MAE) and the mean error (ME). 808
Results 809
The descriptive statistics of the variables, their correlation to the response variable, and 810
multi-collinearity diagnostic results under multi-variate linear regression are listed in. 811
After joining response variables and predictors, there were 80 observations left in our 812
sample, due to the shorter time-series of the satellite-retrieved data (primarily from 1980). 813
Correlation analysis showed that some variables considered had a significant linear 814
correlation relationship with the response factor, at least at one variable in each factor group, 815
except for the variables in the snow cover group and precipitation group. For the correlation 816
among predictors, in general, variables in each of the snow cover groups, snow depth groups, 817
and the temperature groups were highly correlated with each other. This result was also 818
supported by the VIF results. Therefore, variables in those groups were not allowed to enter 819
the model simultaneously. Correlation between Dur and each of the snow cover variables 820
and snow depth variables was moderate ( ). If two or more variables in each of 821
the group enters the model simultaneously, then extra caution will be needed to carefully 822
check the response curves to avoid multi-collinearity problem. 823
0.4 0.7r< <
30
Table 7 Descriptive statistics of the variables
Variable Definition Mean SD
Correlation coefficients
VIF Deviance explained
Dredge weights
Correlation coefficients
VIF Deviance explained
Dredge weights
Livestock loss
Livestock mortality (head) 44318 63932 — — — — — — — —
Livestock mortality rate (%) 11.21 21.76 — — — — — — — — Controlling variables
Year of the snow disaster — — -0.704** 1.68 59.90% 0.95 -0.611** 1.68 50.70% 0.99
County-level year-end herd size (head) 623633 386270 0.036 1.65 0.13% 0.63 — — — —
Elevation of the county centroid (m) 3918 556 -0.017 2.35 24.70% 0.84 0.144 2.35 32.80% 0.96
Snow disaster intensity
Duration of the snow disaster (d) 25.80 29.15 0.559** 2.42 41.00% 0.95 0.571** 2.42 47.90% 0.95
Maximum daily snow depth (cm) 6.03 4.76 0.188 10.99 5.44% 0.50 0.272* 10.99 9.06% 0.37
Minimum daily snow depth (cm) 1.03 1.86 0.143 13.65 4.26% 0.48 0.192 13.65 8.90% 0.15
Mean daily snow depth (cm) 2.73 2.95 0.186 30.38 3.89% 0.44 0.268* 30.38 12.40% 0.17
Maximum daily snow cover (%) 71.96 28.30 0.029 7.65 10.40% 0.12 0.101 7.65 14.20% 0.18
Minimum daily snow cover (%) 27.00 26.48 0.164 7.48 7.15% 0.38 0.171 7.48 4.05% 0.08
Mean daily snow cover (%) 47.91 26.28 0.150 15.10 2.25% 0.25 0.199 15.10 3.97% 0.23
During-disaster environmental stressors
Maximum daily mean wind speed (m/s) 4.41 2.06 0.185 1.72 16.00% 0.91 0.179 1.72 14.10% 0.62
Mean daily maximum temperature ( ) 3.36 5.07 -0.358** 113.36 12.80% 0.57 -0.357** 113.36 12.80% 0.49
Mean daily minimum temperature ( ) -9.16 6.49 -0.450** 177.11 20.20% 0.40 -0.400** 177.11 16.00% 0.67
Mean daily average temperature ( ) -3.70 5.62 -0.433** 510.62 18.70% 0.68 -0.399** 510.62 15.90% 0.58
Pre-season environmental stressors Annual maximum NDVI 0.46 0.19 0.100 6.22 1.00% 0.48 0.044 6.22 0.20% 0.24
NDVI anomaly (%) -1.40 19.30 -0.349** 2.19 17.30% 0.46 -0.315** 2.19 16.30% 0.59
Growing season cumulative precipitation (mm) 371.46 122.40 0.060 6.96 0.36% 0.23 0.074 6.96 0.54% 0.21
Precipitation anomaly (%) 2.39 16.76 -0.051 2.33 9.27% 0.51 -0.109 2.33 1.18% 0.25
SD: standard deviation; Correlation coefficients are the result of the Pearson correlation of a predictor with each of the response variable; VIF: Variance Inflation Factor derived from a multivariate linear regression with all predictors included in the model; Deviance explained is the result of single-variate GAM of each predictor with the response variable; Dredge weights: summation of Akaike weights of the underlying predictor in the dredge analysis. ** significant (two-tailed) at the level of 0.01; * significant (two-tailed) at the level of 0.05
lnL lnLR
LLR
t
N
ELE
Dur
maxSD
minSD
meanSD
maxSC
minSC
meanSC
v
maxT
minT
meanT
NDVI
aNDVI
P
aP
31
Single-variable GAMs indicate that time index , elevation , duration ,
maximum daily snow cover , wind speed , all temperature variables, and NDVI
anomaly all show promising capabilities (deviance explained >10%) in predicting
the response variables. Precipitation anomaly also showed potential in explaining ,
but was weak for . Dredge analysis results reported the relative importance of variables
in a multi-variate context. For mortality (lnL), evidence of including three controlling
variables, snow disaster duration, wind speed, and temperature in the model were all
convincing (the summed AIC weights >0.5). The evidence for including summer
environmental stressors was marginal (the summed AIC weights >0.4). For mortality rates
(lnLR), a very similar pattern can be found, except that none of the snow depth nor snow
cover variables had summed AIC weights >0.4.
Finally, six models were selected from the potentially promising models derived from
dredge analysis, including one minimum model, and two full model versions for each of the
response variables (Table 7). Fitting and cross-validation results indicated that all six models
showed good performance, with adjusted-R2 up to 0.794, and total deviance explained up to
85.2%. The response curves of model L-II and model LR-II are provided in Figure 6 and
Figure 7, and others are provided in Fig. S1, showing consistent patterns across the models.
Table 8 Summary of the selected promising models
ID Formula Total deviance explained
AIC GCV RMSE MAE
L-I 71.0% 261.9 1.42 1.110 0.910
L-II
81.9% 262.1 1.35 1.030 0.837
L-III 85.2% 262.4 1.28 1.077 0.876
LR-I 69.7% 277.2 1.98 1.120 0.895
LR-II 71.6% 277.6 1.99 1.139 0.898
LR-III 70.5% 288.9 2.02 1.142 0.932
t ELE Dur
maxSC t
aNDVI
aP lnL
lnLR
( ) ( ) ( ) ( )ln ~ + + +mean
L s t s Dur s v s T
( ) ( ) ( ) ( ) ( ) ( ) ( )ln ~ + + + + + +amean
L s t s Ele s N s Dur s v s s NDVIT
( ) ( ) ( ) ( ) ( ) ( ) ( )ln ~ + + + + + +amean
L s t s Ele s N s Dur s v s sT P
( ) ( ) ( ) ( )ln ~ + + +LR s t s Ele s Dur s v
( ) ( ) ( ) ( ) ( ) ( )ln ~ + + + + +amean
LR s t s Ele s Dur s v s sT NDVI
( ) ( ) ( ) ( ) ( ) ( )ln ~ + + + + +amean
LR s t s Ele s Dur s v s sT P
32
Figure 6 Response curves of the full model for livestock snow disaster mortality (Model L-II)
Figure 7 Response curves of the full model for livestock snow disaster mortality rate (Model LR-II)
0 20 40 60 80
-3-1
13
Dur
s(Dur,4.05)
3000 4000 5000
-3-1
13
Ele
s(Ele,4.13)
0.0 0.5 1.0 1.5
-3-1
13
N
s(N,3.74)
-0.4 -0.2 0.0
-3-1
13
NDVI_a
s(NDVI_a,1.78)
1985 1995 2005
-3-1
13
t
s(t,1)
-15 -5 0 5 10
-3-1
13
T_mean
s(T_mean,1)
2 4 6 8 10
-3-1
13
s(v,2.26)
0 20 40 60 80
-3-1
13
Dur
s(Dur,4.05)
3000 4000 5000
-3-1
13
Ele
s(Ele,4.13)
0.0 0.5 1.0 1.5
-3-1
13
N
s(N,3.74)
-0.4 -0.2 0.0
-3-1
13
NDVI_a
s(NDVI_a,1.78)
1985 1995 2005
-3-1
13
t
s(t,1)
-15 -5 0 5 10
-3-1
13
T_mean
s(T_mean,1)
2 4 6 8 10
-3-1
13
s(v,2.26)
0 20 40 60 80
-3-1
13
Dur
s(Dur,4.05)
3000 4000 5000
-3-1
13
Ele
s(Ele,4.13)
0.0 0.5 1.0 1.5
-3-1
13
N
s(N,3.74)
-0.4 -0.2 0.0
-3-1
13
NDVI_a
s(NDVI_a,1.78)
1985 1995 2005
-3-1
13
t
s(t,1)
-15 -5 0 5 10
-3-1
13
T_mean
s(T_mean,1)
2 4 6 8 10
-3-1
13
v
s(v,2.26)
0 20 40 60 80
-2-1
01
23
Dur
s(Dur,4.54)
3000 4000 5000
-2-1
01
23
Ele
s(Ele,1.28)
-0.4 -0.2 0.0
-2-1
01
23
NDVI_a
s(NDVI_a,1.67)
1985 1995 2005
-2-1
01
23
t
s(t,1)
-15 -5 0 5 10
-2-1
01
23
T_mean
s(T_mean,1.33)
2 4 6 8 10
-2-1
01
23
s(v,1.99)
0 20 40 60 80
-2-1
01
23
Dur
s(Dur,4.54)
3000 4000 5000
-2-1
01
23
Ele
s(Ele,1.28)
-0.4 -0.2 0.0
-2-1
01
23
NDVI_a
s(NDVI_a,1.67)
1985 1995 2005
-2-1
01
23
t
s(t,1)
-15 -5 0 5 10
-2-1
01
23
T_mean
s(T_mean,1.33)
2 4 6 8 10
-2-1
01
23
v
s(v,1.99)
33
For mortality (lnL), the minimum model included only the time index, duration, wind
speed, and average temperature. However, the performance was not as good as the full
model. Adding elevation and herd size controls, and the pre-season environmental stressors,
improved the deviance explained by more than 10%. For the full model, the version using
precipitation anomaly was markedly better than the version using the NDVI anomaly, in
terms of both goodness-of-fit and cross-validation results. Therefore, the full model with
precipitation anomaly is favored for the prediction purpose, given the availability of
variables in climate scenarios. But the full model with NDVI is favored for explanatory
purpose (Figure 6), as it is more a direct observation of summer vegetation than precipitation.
For mortality rates (lnLR), the overall goodness-of-fit and prediction power of the
models were slightly lower than those of the lnL. The minimum model included only the
time index, elevation, duration and wind speed. Adding other predictors to get the full model
version can slightly increase the goodness-of-fit, but at the cost of slightly increased cross-
validation metrics and error results. The full model using the NDVI anomaly was slightly
better than the version using the precipitation anomaly. Therefore, when predicting mortality
rates, the minimum model could be the most promising choice.
34
5 Design of an LSII
5.1 Product design
According to the surveyed loss mechanism for a local livestock snow disaster, two
types of losses can be considered for insurance coverage. If the snow disaster is not so severe
and the snow cover does not last for a long period (e.g., < 15 days), then the associated loss
will arise in the form of mainly extra costs for the hay and fodder for supplementary feeding.
However, if the snow cover is long lasting (e.g., > 15 days) and the disaster is severe or even
extreme, then a part of or even the majority of livestock could die from starvation and low
temperatures. Then the insurance payment should target the loss of livestock. Given these
facts, two types of insurance coverage can thus be designed: basic coverage providing an
indemnity for extra costs of hay and fodder, and a catastrophic coverage providing an
indemnity for cattle or sheep deaths.
Basic coverage
If more than 60% of grassland is covered by snow in a designated insurance unit (e.g.,
town or county), and the duration of continuous snow cover is no less than 5 days but still
less than 15 days, the basic coverage will be triggered.
Within an insurance period (i.e., a snowing season), the basic coverage can be triggered
multiple times. At each time, the actual duration of each snow-cover process (when snow
continuously covers more than 60% of the grassland) will be recorded (treated as a proxy for
the extra costs of hay and fodder for supplementary feeding) and summed up as the final
seasonal aggregate measure of extra costs. Together with the daily supplementary feed cost
for each sheep unit, the total indemnity can be calculated.
Insurance claims (yuan) = insured exposure (sheep unit) * aggregate duration (days) * daily
supplementary feeding cost (yuan/day/sheep unit); wherein the aggregate duration (days) is the
summed durations of all triggered snow-cover process events
Catastrophic coverage
If more than 60% of the grassland is covered by snow in a designated insurance unit
(e.g., town or county), and this persists for more than 15 successive days, the catastrophic
coverage will be triggered.
35
Within one insurance period, this catastrophe protection can be triggered just once.
Insurance claims (yuan) = insured exposure * expected mortality (%) * insured amount per unit
(yuan)
In the above equation, insured exposure is determined by the actual number of cattle
and/or sheep insured, and the expected mortality rate may be determined using the
relationship derived from the field work or GAM results. If the piecewise indemnity function
of based on field work response is used, the expected mortality can be defined as follows:
� Snow-cover ratio ≥ 60% and 15 days < duration dates ≤ 25 days is a serious snow
disaster, and the expected mortality rate of cattle and sheep is 30%;
� Snow-cover ratio ≥ 60% and 25 days < duration dates ≤ 30 days is a grave snow
disaster, and the expected mortality rate of cattle and sheep is 60%;
� Snow-cover ratio ≥ 60% and 30 days < duration dates is an extremely severe snow
disaster, and the expected mortality rate of cattle and sheep is 85%;
The amount of insurance for a single livestock unit is taken from the current agricultural
insurance in TAR, for which cattle = 4200 yuan/unit and sheep = 400 yuan/unit.
In the design of the above compensation program, the %area snow-cover threshold is
higher than 40%, as suggested by the TAR Meteorological Administration. We note the
following considerations when modifying this threshold: (1) in the case of small-scale
snowfall events, while the snow-covered meadow ratio set to 40% will show that the
meadow area is still larger than the snow-covered area in the region, whose impact on the
animal husbandry is not too serious; (2) in the original standards of the autonomous regional
meteorological administration, the snow-cover depth is considered as well as the snow-cover
ratio area; since the simplified scheme omits a snow-cover depth index, we recommend
increasing the threshold of the snow-cover area ratio; and (3) via the latter calculated case, a
snow-cover ratio set to 60% is much easier to trigger than a 80% threshold, and the
catastrophic indemnity would be too frequent, which is unrealistic.
5.2 Example of an indemnity calculation
To further illustrate the design of the snow disaster index and the operation process
under a practical situation, the following example is offered. According to the daily snowfall
36
and snow-covered data provided by the regional climate center in the Naqu area, the 2007–
2008 winter period was used in this example scenario.
Data
The example uses the National Snow and Ice Data Center data of daily snow cover
from the northern hemisphere from October 1, 2007 to May 31, 2008
(http://nsidc.org/data/g02156). The valid data spanned 240 days. It had a spatial resolution of
4 km, with a time resolution of one day, and covered the entire northern hemisphere. The
data effectively identifies surface snow cover, ice sheets, and non-snow-cover areas (Figure
8).
Note: In the above figure, the areas with snow cover are indicated in white, while the green
areas are vegetation, blue are lakes, and gray any other land types.
Figure 8 Snow-cover distribution in the Naqu area (2007–2008)
Derivation of %area snow cover
The case example here employed the spatial analysis function of ARC/GIS, and IDL, to
analyze the spatial overlay on the snow-cover raster data, on a daily basis, separately at the
county and township administrative levels in the Naqu area. It uses the zonal statistics and
zonal histogram commands to convert the snow-cover data into snow-covered ratio data on a
daily basis, for each administrative level (Figure 9, Figure 10).
37
Figure 9 Snow-covered ratio at the county administrative level in Naqu (2007–2008)
Figure 10 Snow-covered ratio at the township administrative level in Naqu (2007–2008) 2
Calculation of the snow disaster index
Using the daily data for snow-cover ratio, a snow disaster index could be calculated
according to the compensation program under the proposed snow disaster insurance scheme.
The snow-cover daily data were determined according to the specific coverage threshold.
Specifically, all the snow-covered areas reaching an 80% threshold lasting more than 15
days will trigger the compensation standards of the catastrophe guarantee, thus making the
insured eligible for compensation.
Take the example of Naqu County (Figure 11). In the winter of 2007–2008, the 80%
snow-covered area condition was met on three occasions, lasting 7, 6, and 4 days
respectively, thus all failing to satisfy the trigger condition for catastrophe guarantee.
2 At present, there is no public release of township boundary data. The data used here is for illustrative
purposes only.
38
Figure 11 Daily percentage change of snow-covered area in the Naqu County (2007–2008)
Based on this outcome, the maximum consecutive days and total number of days of
2007–2008 winter at the county and township levels in Naqu area were respectively
calculated (Figure 12, Figure 13).
Figure 12 Resulting snow disaster index at the county level in the Naqu area (2007–2008)
Figure 13 Resulting snow disaster index at the township level in the Naqu area (2007–2008)
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
2007/10/1 2007/11/1 2007/12/1 2008/1/1 2008/2/1 2008/3/1 2008/4/1 2008/5/1
Naqu County
39
Figure 12 shows that during the insurance period of winter 2007–2008 there are four
eastern counties with a high snow coverage and long snow duration in the Naqu region,
among which the Jiali County in the southeastern region was the most serious. Considering
the trigger standard used, Jiali County reached the trigger condition of a catastrophe
guarantee (the number of consequtive snow-cover days =20). This should be paid according
to the first settlement of the catastrophe guarantee and the 30% expected mortality rate.
The calculated results for the snow disaster index at the township level are similar to
those at the county level, nonetheless regional differences are evident. With respect to the
catastrophe guarantee, if the county functions as the basic insurance unit, only Jiali County
trigged the condition; however, if the township is the unit, the Nierong, Biru, Suoqing
counties (and so on) trigged the condition, whereas others did not (e.g., Jiali County).
Meanwhile, the number of consecutive days of snow cover beyond the trigger threshold for
some counties was more than 30 days, which reaches the highest level of available
compensation under the catastrophe guarantee insurance. The difference between the two
administrative levels is mainly due to differing calculations of snow-covered areas between
greater area and inner smooth area; the average regional outcome tends to hide serious snow
disasters. Therefore, in this sense, it may be more consistent with the reality of the region to
use township as the basic of unit for insurance.
Calculation of the insurance payment
Depending on the results of the snow disaster index at the spatial scales of counties and
towns, it is possible to determine the payments to either of them for case period of the
insurance. For the illustrative purpose, only the county list of insurance payment was
provided (Table 9). Technically, it is better to measure insurance payment at township level.
40
Table 9 Estimated snow disaster index insurance compensation in the Naqu Prefecture (2007–
2008)
County Catastrophic indeminity
Jiali County 20 days, using an expected mortality rate of 30%; cattle indemnity= 4200 yuan/unit*30% = 1260 yuan/unit; sheep
indemnity = 400 yuan/unit*30% = 120 yuan/unit
Shenzha County —
Bange County —
Naqu County —
Shuanghu Special
County
—
Suo County —
Biru County —
Anduo County —
Baqing County —
Nima County —
Nierong County —
5.3 Premium rate making
Insurance loss risk assessment
According to the principle of insurance pricing, the pure risk loss rate is the expected
value of the loss-cost ratio. To better link the product to practical situations, the process of
rate making often takes the average rate at an administrative unit level according to the
county and township borders, and the pure risk loss rate can then be obtained under the
conditions of catastrophe coverage. The pricing for basic coverage and catastrophic coverage
are as follows:
(1) Basic coverage
According to the payment method, under basic coverage, the insurance indemnity for each
sheep unit in a given snow season is described as follows:
(3)
where is the insurance payment of basic coverage in snowing season of year (October
1 to May 31 of the next year); is the per sheep unit per day payment, expressed in
yuan/day·sheep unit, as determined by the local feeding cost; is the duration of the ith
snow-cover process event that triggers the basic coverage (satisfied by %area
, ,, 5, 0,1,...,
L
t L i i til p d d i N
a a= × " ³ =å
tl t
Lp
,ida
80%a ³
41
snow cover); is the total number of process events that are triggered in a given winter
season.
By definition, the loss-cost ratio is the ratio of insurance loss to corresponding liability.
Therefore, the loss-cost ratio for basic coverage in year is as follows:
(4)
Applying the equation to historical snow-cover data, we can derive the loss-cost ratios
accordingly. Then, a kernel density estimator can again be applied to the lost-cost ratios to
derive its expected and return-period values (Figure 14).
Figure 14 LSII loss risk for Naqu (basic coverage)
(2) Catastrophic coverage
Under catastrophic coverage, the insurance payment for each sheep unit can be defined as
follows:
(5)
where is the catastrophic coverage of year t;
is the standard of compensation for
the catastrophic coverage, expressed as yuan/sheep unit;
is the maximum
tN
t
,
,
0
, 5, 0,1,...,243
itL i
t i
dlcr d i N
d
a
a= " ³ =
-
å
( )max
CAT
t CATl p d= ×D
CAT
tl
CATp
{ }max ,max
itd d
a=
42
duration for a single snow-cover process in year t that have triggered the threshold ;
is the vulnerability function of a serious snow disaster for livestock, which take any
form as discussed in the previous chapter.
Figure 15 LSII loss risk for Naqu (basic coverage) (catastrophe coverage)
Premium rating results
According to the principle of insurance pricing, the actuarially fair premium rate is exactly
equal to the expected loss-cost ratio. Given all distributions derived above, we can calculate
the actuarially fair premium rates for the basic and catastrophic coverage, respectively. Since
the pricing results are likely to be applied in practice, it would be better to link these rating
results to the administrative regions rather than keep them on a pixel basis. Therefore, after
getting the expected loss-cost ratios at the pixel level, we applied zonal statistics to derive
the rates at the township level. These results are presented below in
Figure 16 andFigure 17.
80%a ³
( )D ×
43
Figure 16 Resulting premium rate for the snow disaster index insurance (basic guarantee) in
the Naqu area
Figure 17 Resulting premium rate for the snow disaster index insurance (catastrophe
guarantee) in the Naqu Prefecture
44
The preceding figures reveal obvious eastern-western differences between the net risk
loss rates of catastrophe guarantee for the county unit, which are higher in the east than in
the west. For instance, Naqu region’s southwest, which includes the south of Nima County
and the west of Shenzha County, is the low risk region with a net risk loss rate of < 0.01. The
pure risk loss rate is < 0.03 in the central region and has a tendency to extend into the
medium risk region. But the southeast, which includes the townships of Zhongyu, Gequn,
Jiali in Jiayi County, the Yangxiu and Baiga townships in Biru County, and the Gamu
Township in Suo County, represents the region with highest risk.
45
6 From report to policy
6.1 Involving local communities: pilot insurance programs
To further understand how local communities may respond to snow disasters, and to
verify our vulnerability relationship as employed in the risk assessment, and to check local
herdsmen’s perspectives on the designed LSII product, more fieldwork was carried out
during July 24–August 4, 2017, in the Naqu district. During this stretch of fieldwork the
research team mainly visited western Naqu, the major animal husbandry region of the Naqu
district, which has 11 towns in five counties. The fieldwork schedule and routine are
summarized in Figure 18. We used stratified sampling approach. The sites we visited cover
every pastoral county in central and west Naqu. In each county, three or four villages were
recommended to us by representatives of the local government. In fact, the sampling done
was a compromise, in that it was impossible us to assess the full list of villages in each
county, and then randomly select sites from each. Due to some language problems, the team
had to follow the suggestions from local government representatives. In total, 11 small-scale
workshops/ campaigns were held, with 238 herdsmen interviewed, and 233 valid anonymous
questionnaires collected from them.
Figure 18 Schedule and routine of the second round of fieldwork done in Naqu
46
Workshop/campaign design
Our workshops and campaigns were organized with the help of the PICC Tibetan
headquarters and the PICC Naqu Branch. For each town visited, a local insurance agent from
the county-branch of PICC, and their contact person in the county government (usually the
official in the county department of agriculture and animal husbandry) helped to coordinate
people at the township and village levels. In each town, its herdsmen were invited to gather
for the workshop and campaign activities. The size of each workshop varied greatly by
township, from just several herdsmen up to 70 people in attendance.
At each workshop/campaign site, the local official first introduced the research team
and the goal of the research to the herdsmen’s representatives. Then the research team
coordinator made a self-introduction, and elaborated on the workshop’s purpose, detailed its
plan, and spoke of its potential practical significance. After that, personal interviews were
held. The interview process was conducted mainly by the research team members. All the
questions were designed and presented using a smart phone-based on-line survey system
(https://www.wjx.cn/). The QR plot of the survey is still available on-line (For more
information about the on-line survey, please refer to Appendix 2; in Chinese.) Since most of
the local herdsmen could neither read nor speak mandarin, the local official or the insurance
agent helped with translations.
The interviews consisted of the following parts:
(1) Household information, including the household head’s gender and age, household
size, yak/sheep owned, etc.
(2) Historical snow disaster experience, including the dates, duration, and livestock
mortality of the severest snow disaster that the household had experienced.
(3) Snow disaster preparedness and infrastructure, including the area of alpine roofed
shelters owned by each household, pre-winter storage of hay/forage and fodder,
market prices of hay/forage and fodder, etc.
(4) Snow disaster emergency management strategy, including the actions that a
household is to adopt when a snow disaster strikes, the endurance of yak/sheep
when starving, due to the inaccessibility of a forage supply during the snow disaster
given different levels of supplementary feeding inputs.
47
(5) Insurance perception and acceptance of an index-based snow disaster insurance,
including households’ awareness and perception of the existing agricultural
insurance in the study region, their perception of index-based insurance products,
and willingness-to-pay for the index-based insurance product design by this project.
Workshop/campaign findings
The descriptive statistics of the sample are listed in Table 10. The respondents were
mainly male household heads, with an average age of ~41–50 years. The households mainly
had four to five people in them. Annual household income was below 10 000 yuan,
corresponding to the middle-to-low income class, and the major source of income is selling
yak/sheep. The average number of live yak in stock is approximately 40 per household, and
every year on average three yaks are sold. The average number of live sheep in stock is
approximately 97 per household, with 12 sheep sold yearly on average.
Table 10 Descriptive statistics of the sample
(1) Historical disaster experience
Of the 236 households interviewed, 151 had experienced a snow disaster. The years 1985,
1997, and 2016 had the highest frequency of being indicated as one with a “severest snow
disaster”, and the stated livestock mortality rates for all the three years at the household level
were all above 50%. This stated experience coincides with the records kept by the
meteorological administration (Wen, 2008; China meteorological disaster catalogue: Tibet
Features Count Min Max Average Standard
deviation
Gender 236 0 1 0.83 0.38
Household (1 = yes; 0
= otherwise)
236 0 1 0.89 0.32
Age (years) 236 2 7 4.99 1.16
Household size
(persons)
236 1 6 3.67 1.60
Yak in stock 227 0 680 40.09 53.02
Sheep in stock 226 0 4500 97.19 331.22
Yak sold 227 0 60 3.59 5.05
Sheep sold 226 0 600 12.44 44.51
48
volume). The 1997–1998 snow disaster in Naqu began in September and lasted until late
June of the next year; it covered 90% of the grassland and led to the death of 0.82 million
livestock. The herd size was restored only 12 years later.
Table 11 Stated historical snow disaster experience
(2) Snow disaster preparedness and infrastructure
The average size of an alpine roofed shelter is 105 m2 per household. However, over 63%
of the respondents indicated their needs and willingness to build new shelters or to enlarge
the existing ones. A high majority of respondents (88%) did prepare forge/hay and fodder in
advance for use in the winter season. To do this, they use various approaches, including
buying forage from the market, self-planted forage, and government-allocated forage. On
average, each household spends over 3000 yuan in buying forage and fodder before every
winter season. That portion provided by the government amounts to a small share of the
forage needed by a household, which is quickly consumed once a major snow disaster strikes.
Year Household Duration Yak died Sheep died
Mortality
percentage
for yak
Mortality
percentage
for sheep
1985 34 10 67 290 72% 70%
1995 1 3 10 30 25% 15%
1996 1 30 4 15 6% 16%
1997 40 39 30 66 59% 53%
1998 3 66 31 25 62% 50%
2011 14 25 6 18 32% 41%
2013 7 8 6 18 20% 12%
2014 2 6 0 5 0 45%
2015 3 4 3 81 21% 26%
2016 36 4 2 6 7% 10%
2017 9 5 2 10 8% 11%
49
Figure 19 Sources of supplementary feeding in the winter season
(3) Snow disaster emergency management strategy
Over 70% of the respondents indicated that, once there was a snow disaster leading to
their inaccessibility to open grassland, they chose to provide supplementary feeding to their
livestock at the “half full” level, so to achieve a good balance between a decrease in body fat
and the duration that the storage could support. At the “half full” standard, a yak requires 1.4
kg of hay and 1.4 kg of fodder, while a sheep requires 1.2 kg of hay and 1.1 kg of fodder.
The supplementary feeding costs for yak and sheep are ~16 and ~13 yuan, respectively.
Table 12 Daily supplementary feeding amounts and costs
Given the mode of supplementary feeding, yaks could survive for 3.0–9.0 days on
average, while sheep could survive for 3.6–9.9 days on average. Compared to the earlier
parameter that we used in the vulnerability analysis, this estimation of survivorship is much
more conservative.
Level of
supplementary
feeding
Households Hay/yak
(500 g)
Fodder/yak
(500 g)
Hay/sheep
(500 g)
Fodder/sheep
(500 g)
Cost/yak
(yuan)
Cost/sheep
(yuan)
Full 40 3.89 4.30 2.50 3.21 22.94 11.98
Half full 165 2.87 2.85 2.44 2.26 16.10 13.28
Minimum 31 2.65 2.07 2.13 2.34 13.44 12.51
50
Figure 20 Stated livestock endurance to a low food-intake under supplementary feeing
conditions
(4) Insurance perception, experience, and WTP for snow index insurance
In Tibet, the government provides agricultural insurance for crops, livestock, housing,
and protected agriculture (or controlled environmental agriculture), but at very low liabilities.
The interviews show that local herdsmen have relatively good understanding of livestock
insurance and household property insurance, but they are not familiar with either crop
insurance or agricultural infrastructure insurance. This is closely related to the region’s
experience of purchasing insurance. Out of the 230 respondents, 166 had received an actual
livestock insurance payment, and the average indemnity per household was over 5000 yuan.
Table 13 Agricultural insurance: perceptions and experiences
Degree of
understanding
Degree of
relative
importance
Number of
households
purchased
Number of
households
with
indemnity
experience
Indemnity
/household
(yuan)
Crop insurance 2.25 2.52 11 3 1700
Livestock insurance 3.89 4.07 230 166 5176
Household property
insurance
3.58 3.81 123 15 9925
Protected agriculture
insurance
2.26 2.52 11 2 0
Note: The degree of understanding and degree of relative importance are 5-level ordinal metrics: a higher number
represents a better understanding or higher importance.
51
Due to the limited years of education, although most of the local herdsmen know the
existence of current insurance programs, they can hardly understand how it works. Because
index-based insurance is completely new to them, considerable time for its propagation and
communication is required to help local people learn how use the product to their benefit.
As for what contributed to livestock mortality during snow disasters, factors ranked in
descending order according to the tallied votes of local herdsmen are snow-cover duration,
wind speed, snow depth, and temperature. Thirty-five percent of the respondents indicated
that snow disaster duration was the most important factor for predicting the snow disaster
loss in livestock.
As the loss adjustment of an index-based insurance relies heavily on the underlying
index, the local herdsmen’s concerns about the weather information is important. Our
interviews show that 73% of the respondents’ obtain weather information from the public
media. For the disaster early-warning messages (i.e., snow disaster early-warning), 97% of
the respondents will act immediately upon getting this information. They highly trust the
data reported by the meteorological administration. Therefore, trust is likely not a major
problem for the new index-based insurance product.
After carefully explaining the new product, and providing evidence of pilot loss-
adjustment results based on historical data, 96% of the respondents indicated that the product
was feasible to them, with an overall rating of 3.74 (the highest possible score is 5). An
impressive 97% of the respondents indicated their willingness to purchase the proposed
product. For yaks, the average WTP is 2.91 yuan/head (vs. a liability of 2400 yuan/head).
The WTP for yak is higher than the premium paid by herdsmen for the present yak mortality
insurance (2.52 yuan/head, after 96% of a government premium subsidy). For sheep, the
average WTP is 1.15 yuan/unit (vs. a liability of 480 yuan/sheep). This value is also higher
than the premium of the existing sheep mortality insurance (0.24 yuan/sheep, after 96% of a
government premium subsidy). Therefore, in order to implement and run the new index-
based insurance product, continuing the premium subsidy provided by the government will
be critical for any success. Fortunately, this should not be a make-or-break issue, since the
majority of agricultural insurance programs in China are government-subsidized. Given the
uniqueness of Tibet, and the urgent need to improve the livelihood of Tibetan herdsmen,
such a fund is very likely to be approved by the local, and perhaps even central, governments.
52
6.2 Involving local governments: preparation and launching of the product
To promote the approval of the designed LSII product in TAR, a review meeting was
organized by the PICC Tibet Branch on August 17, 2017. TAR government official
representatives—including people from the TAR financial office, Department of Finance,
Bureau of Insurance Regulation and Inspection, Department Agriculture and Animal
Husbandry, Department of Civil Affairs, and TAR Meteorological Administration—and
Naqu Prefecture government official representatives—including the Vice-Commissioner of
the prefecture government in addition to leading persons in the corresponding
departments/bureaus at the prefecture level—were all invited to review the LSII technical
plan together. (Formal invitation letters for this conference and a list of reviewers invited is
provided in Appendix 3, in Chinese.)
Chairing this conference was the director of the agricultural insurance division at the
PICC Tibet Branch. The project PI elaborated on the outcomes of the entire research project,
by giving an overview of the index-based insurance progress worldwide, the livestock snow
disaster mechanism in the Naqu district, the selection of the snow disaster index,
vulnerability analysis, and product design, the risk assessments and premium calculations, as
well as the local herdsmen acceptance and willingness-to-pay.
Figure 21 Overview of the conference
53
Figure 22 Project PI, Prof. Ye, Tao is presenting the technical plan of the designed insurance
product
Figure 23 Mr. Zha, Jiang (Vice-Commissioner of the Naqu Prefecture government) is
commenting on the technical plan of the product
The reviewers listened to the presentation, carefully read the reports supplied, and
brought up critical questions and comments. In general, the reviewers were quite positive
about the potential of using index-based insurance in the Tibet region, and likewise so
concerning the technical plan of the insurance product. No comments or questions were
raised concerning the general framework and the structure of the product. Questions and
comments were mostly about technical details; e.g., the threshold of snow-cover area
percentage and the minimum days of duration that triggers an insurance payment, per sheep
unit per day insurance liability, and uncertainty in the satellite-retrieved daily snow-cover
data, and so on. The research team carefully answered the questions and responded to the
comments and concerns raised by the reviewers. The reviewing team appreciated the efforts
54
made by the research team, and looked forward to a revised version of the technical plan of
the proposed product.
Revisions to the plan according to the questions and comments raised during the review
conference were carried out immediately after the conference, and finalized on September 15.
(Please refer to Appendix 3 for the complete list of questions and comments, and
corresponding response and revision records, in Chinese.)
Before this report is submitted, the PICC Tibet Branch will have begun organizing the
document-based second round of review. All the departments that were invited to the review
conference will receive a copy of the revised technical plan, with a detailed letter of response
that explains the revisions or reasons that a particular revision was not carried out. Based on
the response by the departments, a formal application for turning the LSII product technical
plan into a real LSII policy will be submitted to the TAR financial office, and to the CIRC
Tibet Bureau, for final approval.
55
7 Discussion
7.1 Future work
By the time this report is submitted, the expected outcomes and deliverables listed in
the technical proposal will have all been fulfilled, except for two pieces of work under
review. However, the ultimate goal of this project is to turn an index-based livestock snow
disaster insurance into practice, which lies beyond the outcomes written in the proposal. To
achieve such a goal, the following road map is provided:
� After the technical plan receives its approval from the TAR government, a formal
application to open a new insurance product shall be submitted to the CIRC Tibet
Bureau, and finally to the CIRC, for official approval of the insurance product. This
could take 2–4 months of review and paper work.
� After receiving the formal reply and approval of LSII from CIRC (expected in late
February, 2018, but the exact time could vary depending on the CIRC inspection
process), the PICC Tibet Branch will put the product into the marketplace.
Simultaneously, an application for the continued government premium subsidy will be
submitted to the TAR office of finance and the TAR Department of Finance.
� The signed LSII policies with local villages in the Naqu Prefecture are expected by late
April 2018. Similar to the first-year pilot program conducted, it will not cover the entire
region of Naqu, but will likely cover the counties of Naqu, Anduo, and Nierong in the
Naqu Prefecture’s central region.
� Review and revise the LSII policy after a one-year trial period, when more experience
and lessons can be learned from its actual operation (most likely to happen in summer
2019).
56
7.2 Suggestions for implementation
(1) Involvement of government agencies
Although the designed LSII aims at using a market-based solution (insurance) to
mitigate risk issues faced by local herdsmen, involvement of the public sector is necessary to
guarantee the fluid operation of LSII. There are several reasons for this.
First, as an index-based insurance product, all loss adjustments purely rely on the
underlying index. Consequently, the observation, calculation, and announcement of the
index together form the product’s keystone. The Tibet meteorological administration is
capable of receiving MODIS Aqua and Terra raw data, and with this data produce a daily
snow-cover map. Therefore, the meteorological administration will be indispensable in
running the project.
Second, China’s booming agricultural insurance market since 2007 is the result of
strong government support, particularly tens of billions in the form of a premium subsidy.
The implementation of LSII is not likely to succeed if there is no government subsidy
involved. Presently, local herdsmen can receive up to 96% of a premium subsidy from the
central, TAR, prefecture, and county governments combined. Without this subsidy, LSII can
hardly be attractive to the local herdsmen. However, if similar rates of subsidy are also
provided, LSII is likely to be preferred by local herdsmen according to our survey results.
(2) Education of local herdsmen to use the product
The literature, our interview results, and the feedbacks during the review conference all
point to the importance of educating local herdsmen to use index-based insurance product.
Due to their extremely low education level, local herdsmen in Tibet have shown difficulties
in understanding the current indemnity-based insurance programs. Rushing the LSII into the
market may lead to some confusion and misunderstanding, as the two types of insurance
products are completely different in terms of their loss-adjustment process. Particularly, the
basis risk issue whereby indemnity does not match actual losses claimed could lead to
serious issues. Therefore, much work needs to be done before the LSII is introduced to local
herdsmen, including education and propagation workshops, hypothetical loss adjustments
and indemnity campaigns, etc.
57
7.3 Potential impacts
For the Tibet region, the designed LSII in this report can help local herdsmen transfer
snow disaster risks and get much-needed funds for disaster recovery, particularly to recover
their livelihood. Once more experience and lessons are learned from Naqu Prefecture, the
product may be used in Ali Prefecture, a region with a similar snow disaster mechanism as in
Naqu, also in Tibet.
The potential impact is not limited to the Tibet region. Livestock snow disaster is not a
special case restricted to Tibet only and is quite common in Central to East Asia temperate
and alpine grasslands. Kazakhstan, Mongolia, and Inner Mongolia in China all are
susceptible to similar types of disasters. In Mongolia, an index-based livestock insurance has
been developed with assistance from the World Bank, but it uses a surveyed livestock
mortality rate reported by the bureau of statistics, which may subject to human influence or
bias. Our LSII product structure can be used in those regions as well, so long as the
parameters, thresholds, and triggers are properly adjusted to reflect the local environmental
and socio-economic conditions.
58
Appendix 1 Photographs of fieldwork
Figure 24 Workshops in Naqu District in summer 2017
Figure 25 Household interviewer (left panel) and roofed shed check (right panel) in summer
2017
Figure 26 Team members with local insurance company manager in summer 2017
59
Figure 27 Workshops at Maqu (left) and Yanshipings (right), Anduo county in summer 2018
Figure 28 Workshops at Beila (left) and Qinglong (right), Bange county in summer 2018
Figure 29 Workshops at Cuba (left) and Shenzha (right) county in summer 2018
60
Appendix 2 Questionnaire of community response to livestock snow disaster
The Vulnerability of Animal Husbandry in Snow Disaster in
Qinghai-Tibetan Plateau Questionnaire
[as translated from the Chinese verion]
Academy of Disaster Reduction and Emergency Management, Beijing Normal University
Purpose
This questionnaire is designed to understand the status of the Qinghai-Tibetan Plateau
herdsmen family during winter disasters and the characteristics of vulnerability that
results in the deaths of livestock. To provide technical assistance to reduce livestock
disaster, please fill in the questionnaire according to the actual situation and ideas.
Confidentiality
All the information that you provide in this questionnaire will be kept confidential, while
all personal information and feedback contents would be used only as a statistical
analysis of data, and the participants in the process cannot be identified as the results
will be presented on the basis of summary.
1. Gender: [Choice]*
○ Male ○ Female
2. Age: [Choice] *
○ Under 18 ○ 18~25 ○ 26~30 ○ 31~40 ○ 41~50 ○ 51~60○ Above
60
3. Please choose the city or district and province: [Choice] *
_________________________________
4. Your location: [Fill in the Blanks]
_________________________________
5. Are you the head of the household? *
○Yes.
○No, I am not the head of the household.
6. How many people are there in your family (the people you live with every day)? *
○Less than 3 people
61
○3 people
○4 people
○5 people
○6 people
○7 people and more
7. What is the average annual net income of your family? [Multiple Choice] *
○<4500 yuan
○4500~9999 yuan
○10000~14999 yuan
○15000~19999 yuan
○20000~24999 yuan
○25000~29999 yuan
○30000~34999 yuan
○35000~39999 yuan
○>40000 yuan
8. The number of livestock in your home [Matrix Text Questions] *
Cattle ________________________ unit
Sheep ________________________ unit
9. What was the number of livestock slaughter last winter? [Matrix Text Questions] *
Cattle
Slaughter ________________________ unit
Sheep
Slaughter ________________________ unit
Snow disaster is an important disaster affecting the production and life of
herdsmen in the plateau winter, which has aroused extensive concern of the
government and society. In order to reduce the impact of snow disaster on
livestock, the government intensified efforts to help herders build warm housing,
storage of forage materials, to prepare for emergency response. We would like to
know more about the impact of the snow disaster in your home.
10. Have you experienced snow disasters? [Choice] *
○No, I have not.
○Yes, I have.
62
11. Please describe the most severe snow disaster you experienced. [Matrix Text Questions]
*
Year ________________________ year
Lasting days ________________________ days
Cattle deaths ________________________ unit
Sheep deaths ________________________ unit
12. How many livestock werethere when you experienced the snow disaster in your home?
[Matrix Text Questions] *
Number of
Cattle ________________________ unit
Number of
Sheep ________________________ unit
13. Please objectively evaluate the impact of winter snow disaster on your productive life
(by severity, 1 is the lightest, 5 is the most serious). [Multiple Choice] *
○No impact ○Light impact○Average
impact
○Relative large
impact ○Serious impact
14.Is there any warm housing in your family? [Multiple Choice] *
○There is no warm barn. Warm barn is used to protect livestock from snow disaster and low
temperatures.
○There is warm barn and there is sufficient area to accommodate all livestock. Warm hosing
is used to protect livestock from snow disaster and low temperatures.
○There is warm barn, but it is too small to accommodate all livestock. Warm hosing is used to
protect livestock from snow disaster and low temperatures.
15. Warm housing can effectively help cattle and sheep against snow disaster. Do you plan
to build or expand warm housing in the near future (3 years)? [Multiple Choice] *
○Yes, I do.
○No, I do not.
63
16. Why do you choose not to create or expand? (Choose one or more) [Multiple responses]
*
□Building warm housing is not very useful to resist snow disaster.
□Lack of sufficient money.
□Lack of helpers in the household.
□Lack of sufficient area in the household.
□Other difficulties _________________
Please fill in the blanks to specify.
17. How many square meters is your warm housing area? [Fill in the Blanks] *
_________________________________
18. Will you reserve some winter fodder for the cattle and sheep before winter? [Multiple
Choice] *
○Yes, I will.
○No, I will not.
19. What is the source of winter fodder for your livestock? (Choose one or more and
specify the answers)[Multiple selections] *
□Purchase the fodder
□Harvest the fodder
□Plant the fodder
□Fetch the fodder
□Government supplies
20. Last winter, you purchased a total of ___jin fodder for livestock, concentrated (highland
barley etc.) ___ jin. [Fill in the Blanks]*
21. Last winter, you harvest a total of ___jin fodder[Fill in the Blanks]*
22. Last winter, you planted a total of ___square metersfodder,or, last winter,you planted a
total of ___jin fodder [Fill in the Blanks]*
23. Last winter, you fetched a total of ___jin fodder[Fill in the Blanks]*
64
24. Last winter, the governmentsupplied a total of ___jin fodder[Fill in the Blanks]*
25. Last winter, the price of dry fodder is ___ yuan/jin, the price of concentrated (highland
barley, etc.) is ___ yuan/jin[Fill in the Blanks]*
26. Why don’t you reserve fodder? (Choose one or more) [Multiple responses] *
□Reserving fodder does not help livestock to live a safe winter.
□I do not have a habit of stocking fodder.
□I do not have enough money.
□I want to buy, but there is no place to buy fodder.
□Other _________________
Please fill in the blanks to specify.
After a heavy snowstorm, the snow will bury the pasture, causing the livestock
fail to dig the snow to eat grass. Being kept at home, the livestock need you to fill
up fodder to keep them alive. Below, we would like to know more about the
feeding of livestock in your home and the characteristics of snow disaster.
27. How do you feed the livestock when you feed them the additional fodder? [Multiple
Choice] *
○Full complement feeding.
○A small amount of feeding to maintain the half full status of livestock.
○Just enough to keep the livestock from starving.
28. Under the condition of “full complement feeding”, the daily feeding of each cattle
(jin)[Matrix Text Questions] *
Dry
Fodder ________________________ Jin (as 500g)
Concentra
ted
Fodder
________________________ Jin
29. Under the condition of “a small amount of feeding to maintain the half full status of
livestock”, the daily feeding of each cattle (jin)[Matrix Text Questions] *
65
Dry
Fodder ________________________ Jin
Concentra
ted
Fodder
________________________ Jin
30. Under the condition of “just enough to keep the livestock from starving”, the daily
feeding of each cattle (jin)[Matrix Text Questions] *
Hay ________________________ Jin
Concentrated
Fodder ________________________ Jin
31. Under the condition of “full complement feeding”, the daily feeding of each sheep
(jin)[Matrix Text Questions] *
Hay ________________________ Jin
Concentrated
Fodder ________________________ Jin
32. Under the condition of “a small amount of feeding to maintain the half full status of
livestock”, the daily feeding of each sheep (jin)[Matrix Text Questions] *
Hay ________________________ Jin
Concentrated
Fodder ________________________ Jin
33. Under the condition of “just enough to keep the livestock from starving”, the daily
feeding of each sheep (jin)[Matrix Text Questions] *
Hay ________________________ Jin
Concentrated
Fodder ________________________ Jin
66
34. Cattlein the above feeding mode, how many days can the adult cattleof different fat
survive (good fat > poor fat)? [Matrix Text Questions] *
Good Fat ________________________ Days
Poor Fat ________________________ Days
35. Sheep in the above feeding mode, how many days can the adult sheepof different fat
survive (good fat > poor fat)? [Matrix Text Questions] *
Good Fat ________________________ Days
Poor Fat ________________________ Days
We would like to know more about your purchase of agriculturalinsurance and
agricultural insurance compensation in recent year.
36. How much do you understand the following agriculturalinsurance products? (Choose
one or more)[Matrix Multiple Choice] *
Do not
understand at
all
Not
too
well
Average Relatively
well
Very
well
Planting (highland barley, wheat,
potatoes, etc.) ○ ○ ○ ○ ○
Aquaculture (Tibetan cattle,
Tibetan sheep) ○ ○ ○ ○ ○
Property insurance (agricultural
herdsmen own housing,
agricultural motor vehicles)
○ ○ ○ ○ ○
Facilities agriculture (greenhouse
vegetables, greenhouse frames) ○ ○ ○ ○ ○
37. What do you think of the importance of the above agricultural insurance? [Matrix
Multiple Choice] *
Not
at all
Not too
important Average
Relatively
important
Very
important
67
Planting (highland barley,
wheat, potatoes, etc.) ○ ○ ○ ○ ○
Aquaculture (Tibetan cattle,
Tibetan sheep) ○ ○ ○ ○ ○
Property insurance
(agricultural herdsmen own
housing, agricultural motor
vehicles)
○ ○ ○ ○ ○
Facilities agriculture
(greenhouse vegetables,
greenhouse frames)
○ ○ ○ ○ ○
38. As far as you know, does your household have purchased the following insurance?
(Choose one or more) [Multiple responses] *
□Planting (highland barley, wheat, potatoes, etc.)
□Aquaculture (Tibetan cattle, Tibetan sheep)
□Property insurance (agricultural herdsmen own housing, agricultural motor vehicles)
□Facilities agriculture (greenhouse vegetables, greenhouse frames)
□None
39. Have you or your household ever been indemnified for some kind of event? [Multiple
Choice] *
○No
○Yes
40. Please providethe latest indemnity year. [Multiple Choice] *
○2017
○2016
○2015
○2014
○2013
○2012
○2011
○2010
41. Cause of payment (Choose one or more) [Multiple responses] *
68
□Planting (highland barley, wheat, potatoes, etc.)
□Aquaculture (Tibetan cattle, Tibetan sheep)
□Property insurance (agricultural herdsmen own housing, agricultural motor vehicles)
□Facilities agriculture (greenhouse vegetables, greenhouse frames)
42. What is the total compensation of planting insurance (yuan)? [Matrix Text Questions] *
Planting (highland
barley, wheat,
potatoes, etc.)
________________________
43. What is the total compensation of aquaculture insurance (yuan)? [Matrix Text
Questions] *
Aquaculture
(Tibetan cattle,
Tibetan sheep)
________________________
44. What is the total compensation of property insurance (yuan)? [Matrix Text Questions] *
Property insurance
(agricultural
herdsmen own
housing, agricultural
motor vehicles)
________________________
45. What is the total compensation of facilities agriculture insurance (yuan)? [Matrix Text
Questions] *
Facilities agriculture
(greenhouse
vegetables,
greenhouse frames)
________________________
69
In response to the loss of cattle and sheep caused by local snow disasters, we want to
design a new type of insurance products. Here is the following several questions that we
would like to know.
46. What do you think is the most important factor affecting the death of livestock when
the snow disaster occurs? [Multiple Choice] *
○The duration of the snow
○The depth of the snow
○Temperature
○Wind speed
□Other _________________
Please fill other reasons to specify.
47. Do you pay attention to meteorological information of National Meteorological
Department (such as weather forecast)? [Multiple Choice] *
○No, I do not.
○Yes, I do.
48. Do you prepare for disaster after receiving information from the state and government?
[Multiple Choice] *
○Yes, I do.
○No, I do not. _________________
The reason is
49. What do you think of the accuracy of the followingweather information provided by the
National Meteorological Department? [Matrix Multiple Choice] *
Very
inaccurate Inaccurate Average Accurate Very accurate
Temperature ○ ○ ○ ○ ○
Precipitation ○ ○ ○ ○ ○
Wind Speed ○ ○ ○ ○ ○
50. Do you think it would be possible if we use meteorological data from the National
Meteorological Department to predict the number of deaths in your household? [Multiple
Choice] *
○No
70
○Yes
51. If applicable, please rate the feasibility of this method (1->5 increased degree of
feasibility) [Multiple Choice] *
Very
unlikely ○1 ○2 ○3 ○4 ○5 Very likely
52. Do you know anything aboutindex insurance products? [Multiple Choice] *
○No, I do not.
○Yes, I do.
The current insurance product is to be compensated by the actual investigation and
damage of the insurance personnel. We intend to design a new insurance product, using
meteorological data issued by the Meteorological Bureau to predict the actual disaster
losses, and not to do a field survey of the damage. Its advantage is to simplify the
insurance investigation and the loss of the process to lower the premium, while the
disadvantage is that depending on the data published by the Meteorological section may
cause deviations from the actual loss.
53. If this insurance product is provided, would you like to buy this product to prevent the
snow disaster loss in winter? [Multiple Choice] *
○Yes, I do.
○No, I do not. _________________
Please specify your reasons.
54. How much would you like to pay for this insurance product (yuan)?
Every unit of cattle_______ yuan Every unit of sheep _______yuan[Fill in the Blanks]*
The QR code for online-questionnaire:
71
Appendix 3 Information about the reviewing conference in Lhasa
72
73
“The Snow Disaster Weather Index Insurance product design scheme of animal
husbandry in Naqu, Tibet”
Reviewer comments and response
“The Snow Disaster Weather Index Insurance product design scheme of animal husbandry
in Naqu, Tibet” held an expert consultation, on the design of related programs to consult the
relevant comments on August 17, 2017 at the Lhasa Tibet Hotel. The questions and
recommendations raised by the consultants at the conference are as follows:
Chen, Manjiang (Deputy director of the Office of Finance of the government of the
autonomous region):
Questions/comments:
1) In the snow disaster index insurance’s basic guarantee payment shceme, one of the
triggering conditions is: snow covered grassland area is more than 60%. Is it
realistic? In the report, the Meteorological Office provided the area ratio is 30% as
the difference is one-fold.
2) Is there any official basis for the number of days for indemnity in the snow disaster
index insurance?
3) The indemnity for each sheep unit is measured by 2 yuan/day. Is it consistent with
local government standards?
4) If the insurance product is promoted in Naqu, can the premium be included in the
country’s subsidy range?
Suggestions:
� Index insurance is based on pre-set parameters to trigger, so the setting of trigger
parameters is the core of insurance products, as the triggering conditions must
reflect the objective law of local agricultural insurance, scientific and reasonable
combination to realize the reduction of the government burden, the decrease of the
loss of farmers, and the profit of insurance companies.
� The index insurance is determined by the same index, facing a big base difference
risk. In the actual promotion, the local government may face the contradiction
caused by the base difference risk, so we should strengthen the popularization and
guide work of relevant knowledge.
74
� Local governments and herdsmen have limited knowledge of index insurance, so
the relationship between the new insurance product and the existing insurance
product should be handled properly.
Shu, Hua (Director of the Property Insurance Supervision Office of the Provincial
Bureau of Security Supervision):
Questions/comments:
5) When the triggering condition of the snow disaster compensation is snow-covered
grassland area more than 60%, the compensation is 2 yuan/sheep unit every day.
When the recommended area ratio is less than 60%, such as 40% or 50%, the
compensation is paid proportionally.
6) The duration of the drought and snow disaster insurance coincidein May. Whether a
region in May will occur in the case of snow disaster and drought disaster? Is the
probability of occurrence big?
Xi, Qiong (Director of Disaster Response Command Office, Agriculture and Animal
Husbandry Department, Autonomous Region):
Questions/comments:
7) 24 consecutive hours of snow is called a snow disaster, which generally occurs in
November to March the next year. At the end of March, in April and May, due to
the rise of temperature, it will not cause a snow disaster. It is suggested that the
insurance period of snow disaster in the report be changed to November to March of
the following year.
Suggestions:
� In recent years, disasters have reduced, as the annual disaster statistics of the crop
disaster area is less than 150,000 mu, livestock died less than 150,000, of which 70%
were calves. At present, large animals have insurance, while cubs, boars, piglets,
and Tibetan pigs have no insurance.
� Increase the publicity of index insurance.
� The policy type insurance of planting is limited to highland barley, wheat, corn, rice
food crops, rapeseed, potato, greenhouse vegetables. Aquaculture insurance is
limited to yak, ox, sheep, goat, sow.
75
Yang, Bin (Director of the Tibet Meteorological Administration, Public Meteorology
Service Center)
Questions/comments:
8) Because of the risk of base difference of index insurance, can the compensation of
the East and West region be differentiated?
9) Increase the ground observation data in sensitiveareas (more disaster-prone regions).
Fu, Mingxing (Director of Naqu Agriculture and Animal Husbandry Bureau)
Questions/comments:
10) If it is cloudy for a week or half a month, it is difficult for the satellite to discern the
snow cover. What to do then?
11) One of the triggering conditions for snow disaster is: 5 days≤ the duration ≤15 days.
Because Naqu county is relatively cold, can it be changed to 3 days ≤the duration
≤15 days?
12) The cost of daily feeding is too low. At present, in accordance with each sheep unit
4 jin/day for feeding, the market price of fodder is 1.4 yuan/jin, so the cost should
be 5 to 6 yuan/day.
Zha, Yu (Chief of Agriculture, Naqu Finance Bureau)
Questions/comments:
13) The proportion of premiums should be solicited by the government and herdsmen.
Bian Ba Ci Ren (Party member/Dicipline Team leader, County Meteorological Bureau)
14) The remote sensing data may be erroneous. It is recommended that some key areas
be supplemented by manual monitoring.
15) The Meteorological Bureau has no rural border data, only county-level border day.
Can the research group of the Normal University provide relevant township
boundary data?
Lang Jie, Yu Zhen (Director of Finance Department of the Autonomous Region)
76
Questions/comments:
16) It is suggested thatthe index insurance adopt the ppp (public-private property) mode,
as the government and the insurance company jointly protect the mode, the pure
business model will have a lot of problems when it carries out.
17) The Inner Mongolia Autonomous Region has launched a pilot work on the relevant
index products. What is the payment rate in the past few years?
De Qing, Zhuo Ga (Chief Meteorological Service of the Regional Meteorological
Bureau Service Center)
Questions/comments:
18) From the meteorological point of view, the triggering conditions of snow disaster,
the ≥60% of the area of snow-covered grassland is too high. Because pasture grass
is low, once it snows, the pasture will be covered. It is recommended that the
number of days to be lowered changed to 3-5 days.
The project team studied and modified the technical details of the snow disaster index
insurance product scheme on the basis of careful understanding of experts’ questions,
opinions and suggestions. The main points of modification are as follow:
1. Questions/Comments on the triggering conditions of the snow disaster index insurance
products: discussion on the area ratio of snow-covered grassland and the duration of
snow disaster
As suggested by the reviewers, further analyses were carried out on the trigger conditions
of snow disaster. The triggering duration is changed to ≥ 3 days; and the cover area of snow
ratio is separately considered as in ≥ 30% and 40%. After careful calculation, it is found that
the threshold of 30% may be too low, leading to high probability of triggering and indemnity
eastern counties. After carefully considering the trade-offs, we set the trigger in duration as
≥ 3 days and snow-covered ratio as ≥ 40% as the final plan snow disaster index insurance’s
triggering conditions.
2. Questions/Comments discussion on the unit indemnity per sheep unit and the market
price of fodder material
Original daily average indemnity of 2 yuan/sheep was the result of communication with
local governments in the July 2016 survey, but this level refers to the level of supplementary
77
feeding under the condition of maintaining basic vital signs. If we want to reach the half full
above the supplementary feeding level, it is really necessary to achieve the 5 yuan/sheep
expenditure level. Therefore, in the revised version, the local daily supplementary feeding
cost is changed to 2 to 5 yuan/sheep. The example of calculation uses 2 yuan/sheep every
day and 5 yuan/sheep every day to calculate the size of the premium revenue and premium
subsidy expenditure. As the level of the unit insured will have a greater impact on the final
premium revenue, it is suggested that the final results should be decided on the basis of the
affordability of local herdsmen and the amount of financial subsidy by the government of the
autonomous region.
3. Questions/Comments discussion on basis risk and data accuracy
There are obvious differences in the east and west parts of Naqu county, and the revised
report fully agrees with the expert’s recommendation to make differential payments by
township, or at least the counties to determine premium rates.
Using MODIS products to observe a snow cover can indeed be affected by clouds. This
problem can be modified by a certain data processing algorithm. For example, the snow
dataset (http://westdc.westgis.ac.cn/data/94a8858b-3ace-488d-9233-75c021a964f0) of the
Qinghai-Tibet Plateau area prepared by the Environmental and Engineering Research
Institute of the cold zone in the Chinese Academy of Science. It uses a third-order-spline
interpolation method to interpolate the cloud-contaminated data. Such kind of algorithms can
be included in the operational system supporting the index insurance products being
constructed by the Regional Meteorological Administration.
With regard to the use of township boundaries, the township boundaries used in this report
are non-national standards, which can only be used as scientific research and examples. In
the process of building a business operational system, it is suggested to coordinate with the
autonomous region government. The administrative boundary map of the county with legal
effect should be issued by the SMC or Civil Affairs Bureau of Autonomous Region to
support the calculation of snow index.
4. Questions/Comments on premium subsidy and PPP model
As to the proportion of subsidy, the government and the insurance company liaison are not
the technical problems of the insurance product. On the basis of the revised report, it is
suggested that the rate of measurement and the size of premium are discussed separately by
relevant departments of the autonomous region.
78
5. Questions/Comments on the indemnity rate of similar products in Inner Mongolia
Autonomous Region
Inner Mongolia Autonomous Region runs a short time on the snow disaster insurance
index of sheep. In 2015, Wulagai district of Dongwuzhumuqin Qi in Xilinguole Meng,
carried out a pilot program the insurance. Wulagai Pasture and Hesigewula Pasture’s
premiums each had a premium revenue of 1 million, and the total of 2.4 million yuan has
been paid for the snowstorm. In 2016, Abaga Qi in Xilinguole Meng was paid 1.0995
million yuan due to the snow disaster. Due to only two years of data, it is not enough to
make a general consideration of the indemnity.