Geospatial activities at ICRISAT · assessed using Google Earth ... Accuracy assessment: Error...
Transcript of Geospatial activities at ICRISAT · assessed using Google Earth ... Accuracy assessment: Error...
Geospatial activities at ICRISAT
Murali Krishna GummaHead – RS/GIS UnitSenior Scientist - Geospatial Science
Big Data in AgricultureConvention
19th – 22th September 2017CIAT HQ Major crops (2014)
01. Rainfed-sc-sorghum
02. Rainfed-sc-millets/sorghum
03. Rainfed-sc-groundnut
04. Rainfed-sc-pigeonpea
05. Rainfed-SC-maize/sorghum/millet
06. Other crops
Geospatial products for SAT
Target Research groups• Breeders• System modelers• Social scientists• Hydrologists• Planning departments
Crop type / intensity maps
Land use changes
Tracking adoption of NRM Technologies
Spatial modeling(Prioritization)
Simulated yield estimations and impact
Abiotic stresses
Impact assessment
Length of growing periods
Water productivity
GFSAD 30m project
Major goal is to produce high resolution (30-m) global cropland products, including:
• Cropland Vs Non croplands
• Irrigated Vs rainfed (including water bodies)
• Cropping intensities: single, double, continuous
• Global major crop types
• Cropland change
Overview: Mapping @ 30m Resolutions
CountryTotal
geographical area ('000 ha)
Bangladesh 14,804
Bhutan 4,365
India 345,623
Nepal 16,210
Pakistan 89,167
Sri Lanka 6,453
Iran 164,820
Afghanistan 64,750
Total 706,192
Background: Croplands of South Asia
• Geographical area: 706 Mha
• Population: 1.7 billion
• Croplands: 254 Mha
• GDP 7.7 (agriculture 19%)
• Major crops: rice, maize, wheat, barley,
pulses and Plantations (coffee, tea and etc)
CountryTotal
geographical area ('000 ha)
Total gross planted
area ('000ha)
Bangladesh 14,804 15002
Bhutan 4,365 121
India 345,623 184443
Nepal 16,210 4208
Pakistan 89,167 22817
Sri Lanka 6,453 2076
Iran 164,820 18130
Afghanistan 64,750 7770
Total 706,192 254,568
Methodology Flow-chart: Random Forest Classification
Seasonal 30m products
Season1: Jun – Oct
Season2: Nov – Feb
Season3: Feb -Apr
Image
segmentations
Landsat-8 16 daytime series data
Input Bands
(N=8)
Band1:blue
Band2:
Band3:
Band4:
Band5:
Band6:
Band7:
Random Forest Algorithms (RF)
Cropland classification (pixel based)
Ground data
• Field work during 2013-16• Validation datasets
(Thenkabail et al 2005; Gumma et al 2011, 2016)
*All samples were visualassessed using Google Earthhigh resolution imagery
Satellite image composition
Agro-ecological
zones
Cropland classification &image segmentation
Training data
Validation data
Cropland products
1. Croplands Vs non croplands
2. Irrigated Vs Rainfed croplands
Characteristics of Satellite data used for South Asia
Croplands of South Asia using Google Earth Engine (GEE) Cloud Computing
@ 30-m Resolution based on Landsat 16-day Time-Series
Region/ Country
Landsat image Series
Years of
Data
# Composites
Bands percomposite
Total # bands used
South Asia,Iran and
AfghanistanLS8
2014&
2015
Monsoon (151 – 300)
Winter (301-365,1-60)
Summer (61-150)
blue, green, red, NIR, SWIR1, temp, SWIR2 and NDVI (n= 8)
48
Croplands of South Asia using Google Earth Engine (GEE) Cloud Computing
@ 30-m Resolution based on Landsat 16-day Time-Series
• Field work during 2013-16• Validation datasets
(Thenkabail et al 2005; Gumma et al 2011, 2016)
*All samples were visualassessed using GoogleEarth high resolutionimagery
Product 1: Cropland Vs Non-cropland
Sample size = 2088
Crop lands = 1204Non croplands = 884
Croplands of South Asia using Google Earth Engine (GEE) Cloud Computing
@ 30-m Resolution based on Landsat 16-day Time-Series
Accuracy assessment: Cropland Vs Non-cropland
Land use / land cover01.
Croplands
02. Non-croplands
(Other LULC)
Row
TotalCommission
error
01. Croplands 615 32 647 4.9%
02. Non-croplands (Other
LULC) 40 76 116 34.5%
Omission error 6.1% 29.6%
Producers accuracy 93.9% 70.4%
Users accuracy 95.1% 65.5%
Overall Accuracy 90.56%
Kappa 0.623
Product 2: Irirgated Vs Rainfed cropland
Sample size = 2634Irrigated croplands = 1099 Rainfed= 651Non croplands = 884
Croplands of South Asia using Google Earth Engine (GEE) Cloud Computing
@ 30-m Resolution based on Landsat 16-day Time-Series
Accuracy assessment: Irirgated Vs Rainfed cropland
Land use / land cover01. Irrigated-
croplands
02.
Rainfed-
croplands
03. other
LULC
04.
Waterb
odies
Row Total Commissi
on error
01. Irrigated-croplands 386 35 4 0 425 9.2%
02. Rainfed-croplands 56 218 4 0 278 21.6%
03. other LULC 27 20 32 0 79 59.5%
04. Waterbodies 2 1 0 2 5 60.0%
Omission error 18% 20% 20% 0%
Producers accuracy 81.95% 79.56% 80.00% 100%
Users accuracy 90.82% 78.42% 40.51% 40%
Overall Accuracy 81.07%
Kappa 0.655
Comparison : 1km, 250m, 30m and High resolution imagery
1-km Product 250-m Product 30-m Product High resolution image
Ganges river basin
Krishna river basin
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 42 32 74 56.76%
No-Crop 17 159 176 90.34%
Total 59 191 250
Producer Accuracy 71.19% 83.25% 80.40%
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 4 3 7 57.14%
No-Crop 8 235 243 96.71%
Total 12 238 250
Producer Accuracy 33.33% 98.74% 95.60%
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 111 19 130 85.38%
No-Crop 40 80 120 66.67%
Total 151 99 250
Producer Accuracy 73.51% 80.81% 76.40%
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 140 18 158 88.61%
No-Crop 24 67 91 73.63%
Total 164 85 249
Producer Accuracy 85.37% 78.82% 83.13%
Zone 1
Zone 2
Zone 3
Zone 4
Total land area of Zone 1 (TLAZ1): 159.6 MhaCropland as % of TLAZ1: 22.26 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 13.54 %
Total land area of Zone 2 (TLAZ2): 191.86 MhaCropland as % of TLAZ2: 3.41 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 2.49 %
Total land area of Zone 3 (TLAZ3): 122.43 MhaCropland as % of TLAZ3: 43.35 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 20.22 %
Total land area of Zone 1 (TLAZ1): 174.87 MhaCropland as % of TLAZ1: 57.49 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 38.30 %
Accuracy assessment: Error Matrix
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 104 17 121 85.95%
No-Crop 36 92 128 71.88%
Total 140 109 249
Producer Accuracy 74.29% 84.40% 78.71%
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 15 3 18 83.33%
No-Crop 16 216 232 93.10%
Total 31 219 250
Producer Accuracy 48.39% 98.63% 92.40%
Reference Data
Crop No-Crop Total User Accuracy
Map
Dat
a Crop 418 92 510 81.96%
No-Crop 141 849 990 85.76%
Total 559 941 1,500
Producer Accuracy 74.78% 90.22% 84.47%
Zone 5
Zone 6
All Zones Overall Error Matrix
Total land area of Zone 5 (TLAZ5): 144.32 MhaCropland as % of TLAZ5: 62.636 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 23.86 %
Total land area of Zone 6 (TLAZ6): 68.55 MhaCropland as % of TLAZ1: 4.16 %Total net cropland area of SAsia (TCASA) = 262.47 MhaCropland as % of TCASA : 1.59 %
Accuracy assessment: Error Matrix
Rice-fallows: South Asia
Legend
01. Irrigated-SC-rice in kharif-fallow in rabi-fallow in summer
06. Rainfed-SC-rice in kharif-fallow in rabi-fallow in summer
Rice-rice/othercrops
Product 5: Rice-fallows (rainfed agriculture)
StateRainfed:
rice-fallows
% of total rice-fallow
Chhattisgarh 4111731 35.2%
Madhya Pradesh 1871816 16.0%
Orissa 1793852 15.3%
Jharkhand 975780 8.3%
Maharashtra 664907 5.7%
West Bengal 605092 5.2%
Telangana 407943 3.5%
Assam 302036 2.6%
Bihar 266314 2.3%
Karnataka 235265 2.0%
Gujarat 168620 1.4%
Andhra Pradesh 95469 0.8%
11,498,823 98%
Gumma et al., (2016)Accuracy 80%
Baseline Scenario @ present
New Scenario with Improved water Productivity and less water consuming crops
Season 1
Season 1
Season 2
Season 2
Total area of croplands, season 1 (TAC) = 47,696 ha Total area of croplands, season
2 (TAC) = 33,864 ha
Total area of croplands, season 1 (TAC) = 47,696 ha
Total area of croplands, season 2 (TAC) = 37,231 ha
Croplands (%)
01. Rice (71)
02. Pulses (20)
03. Corn (3)
04. Onions (0)
05. Cropland fallow (6)
06. Other LULC
Telangana_dist
Croplands (%)
01. Rice (68)
02. Pulses (0)
03. Corn (3)
04. Onions (0)
05. Cropland fallow (29)
06. Other LULC
Telangana_dist
Croplands (%)
01. Rice (46)
02. Pulses (36)
03. Corn (14)
04. Onions (0)
05. Cropland fallow (4)
06. Other LULC
Telangana_dist
Croplands (%)
01. Rice (19)
02. Pulses (17)
03. Corn (28)
04. Onions (15)
05. Cropland fallow (22)
06. Other LULC
Telangana_dist
Crop production and water use – Kadam command area
Table A. Kaddam water use in: (A) baseline scenario, and (B) New scenario of improved water productivity and re-allocation of crops
Crop type
Percent of total
cropland area
in season 1A,C
(%)
Percent of total
cropland area in
season 2A,C
(%)
Water used for
producing 1 kg of
grainE,F
(liters)
Yield per
hectares in
(kg/hectare)
Total water used
by all crops in 2
season (liters)
Rice 71 68 3400 2500 483579280000
Pulses 20 0 1608 1320 20247524352
corn 3 3 1222 6500 19434932400
onions 0 0 345 19000 0
Cropland fallow 6 29 100 0 0
Total Area of croplands (hectares) 47696 33864 532,262 billion liters
Other land cover area (hectares) 8910 22742 Current water use 523 billion litersTotal area (croplands + non-croplands) (hectares) 56606 56606
Rice 46 19 2600 2400 1.81048E+11
Pulses 36 17 1608 1320 49879799165
corn 14 28 1222 6500 1.35842E+11
onions 0 15 345 19000 36607380750
Cropland fallow 4 22 100 0 0
Total Area of croplands (hectares) 47696 37231 403,377 billion liters
Other land use land cover area (hectares) 8910 19375 New reduced water use 403 billion litersTotal area (croplands + non-croplands) (hectares) 56606 56606
Reduced water use in new scenario Water Savings 120 billion liters
A. Baseline scenario: with business as usual crops at present during season 1 and\or 2
B. New scenario: with improved water productivity, re-allocation of less water consuming water-smart, economically-smart crops
25% water savings and 52% rice equivalent yield gain
Crop land extent map: Africa
Products can be accessed through our map web portal:https://web.croplands.org/app/map
(Jun et al. 2017)
Cropland area comparison with Survey-based stat
The total net cropland area of Africa was estimated as 313 Mha. Cropland areas were computed for each of the 55 African Countries and compared with the UN FAO statistics; this explained 65% of variability for all 55 Countries
Season wise Crop extent maps: Myanmar
Season
Months of the season
Overall accuracy of 7 classes (%)
Kappa (no units)
Producer's accuracy for fallow cropland classes (%)
Users'saccuracy for fallow cropland classes (%)
1June-October
90 0.85 57 80
2November-February
85 0.77 98 82
3 March-May 87 0.7 92 92
Accuracy assessment
After Syncing the data it will display the message
Tap on Toggle button and select map plotting to see the points plotted on google map.
Ground data: Year 2016-17 (Eastern India)
• Ideal Signature total number of points – 353
• Validation total number of points – 1463
• Total number of kilometers travelled - 7083
Conclusions
▪ Random forest algorithm was used to create Crop extent for South Asia using Google Earth Engine on Landsat 30m data at good accuracy;
▪ Irrigation vs. rainfed (Product 2) was produced based on irrigation information (secondary sources)
▪ 7 Journal articles; 4 book chapters; 6 papers in review(from last CSI meeting to till now)
Further work▪ Crop extent maps in South Asia and African smallholders regions
▪ Ground data collection for crop type mapping
▪ Publications
Data access / web portals
Web sites and Data portals: http://croplands.org (30-m global croplands visualization tool)http://geography.wr.usgs.gov/science/croplands/index.html (GFSAD30 web portal and dissemination)http://geography.wr.usgs.gov/science/croplands/products.html#LPDAAC (dissemination on LP DAAC)http://geography.wr.usgs.gov/science/croplands/products.html (global croplands on Google Earth Engine)croplands.org (crowdsourcing global croplands data)
• ICRISAThttp://maps.icrisat.org/rs/maps/index.html