Understanding The Hydrologic and Socioeconomic Impacts of Water Use and Resource Allocation Under...

1
Understanding the Hydrologic and Socioeconomic Impacts of Water Use and Resource Allocation M. Maneta 1 , P Gardner 1 , J Kimball 2, K Jencso 3,4 , B Maxwell 5 , S Ewing 5 1 Geosciences, University of Montana; 2 Numerical Terradynamic Simulation Group, U. of Montana; 3 School of Forestry, U of Montana; 4 Montana Climate Ofce; 5 LRES, Montana State UIniversity Introduction Temperature The projected increase in the frequency and intensity of droughts over agricultural regions can reduce the water available for irrigation with important ecological and economic consequences. Information on how farmers accomodate water shortage is necessary for robust and ecient agricultural and water policy analysis. New algorithms to retrieve agricultural parameters from earth-observing satellites can be used for micro-level analysis of agricultural production. This information can be used in conjunction with economic models of agricultural activity to describe how farmers will behave when faced with water shortage or policy constraints. Here we present a hydroeconomic modeling framework that assimilates remotely sensed observations of agricultural activity (acreage, yields, crop mix, and evapotranspiration) into an agroeconomic model calibrated using Positive Mathematical Programming (PMP). PMP is widely used in agricultural policy analysis to calibrate models that precisely reproduce the observed economic activity levels of a base year. An important characteristic of models calibrated using PMP is that they reect the actual behavior of farmers and their actual reaction to external factors, such as drought and risk, rather than the behavior that would be optimal from an agronomic point of view. We extend the classic PMP approach by implementing it within a data assimilation framework based on an ensemble Kalman Filter. The extension permits: Hydro-economic models: Objectives The modeling framework presented here permits the operationalization of new earth observing capabilities to predict changes in land and water use in agricultural regions, to predict the impact of such changes in the hydrologic network, and to inform agricultural and water policy. Specically, the methods can give insight into the following questions: 1. How do droughts aect crop mix and land use devoted to agriculture? 2. How does agricultural change impact water availability and other water users? 3. What choices and management practices are prioritized, and how can that inform future policy options? 4. How can new remote sensing information be integrated in models to provide insight into socioeconomic dynamics? Remote sensing of agricultural activity Example for Ravalli county, Montana Hydro-economic model of agricultural activity Positive Mathematical Programming with the enKF Satellite products and algorithms available to inform resource allocation include: - Land allocation and crop mix: USDA-NASS Cropland Data Layer (CDL) (annual for conterminous US) - Crop yield: Estimated from radiation use eciency (Doraiswamy et al, 2005; Lobell et al., 2003) - Supplemental Irrigation: ET estimates based on observed crop phenology, weather data and irrigation eciency factors Text here... Hydro-economic model at simulation time Behavior under resource constraints is simulated by substituting the identied parameters into the production function and nding the resource matrix X that solves the net revenue maximization problem. To test the model we simulate farmer behavior under known baseline constraints. Conclusions and main ndings Parameter distributions converge to a steady value in less than 5 assimilation cycles Predictive variance of the parameters is rapidly reduced from the initial values The parameter ensemble decreases but does not collapse to a single point (stability) Potential cause of bias Precipitation undercatch due to wind eects. Poor representativeness of the observation network from high grid-cell variability. Station location bias toward lower (and atter) terrain. Response to reduced access to water The heart of the agronomic model is a parametric agricultural production function q i that calculates yield of crop i given the allocated farming resources j embeded in matrix X. The model assumes that farmers will allocate their available resources with the objective of maximizing their net revenues Text here... References Lobell, D. B., Asner, G. P., Ortiz-Monasterio, J. I. and Benning, T. L.: Remote sensing of regional crop production in the Yaqui Valley, Mexico: estimates and uncertainties, Agric. Ecosyst. Environ., 94(2), 205–220 Maneta, M.P. and Howitt (2014). Stochastic calibration and data assimilation in non-stationary hydro-economic models. Water Resources Research, 50(5):3976–3993 Maneta, M. P., Torres, M. O., Wallender, W. W., Vosti, S., Howitt, R. E., Rodrigues, L., Bassoi, L. H. and Panday, S.: A spatially distributed hydroeconomic model to assess the eects of drought on land use, farm prots, and agricultural employment, Water Resour. Res., 45, W11412 Roy et al. (2010). Web-enabled Landsat Data (WELD): Landsat ETM+ Composited Mosaics of the Conterminous United States, Remote Sensing of Environment, 114: 35-49. Construct the production function parameter ensemble by sampling from the smoothed mixture of Gaussian densities: Use the parameter ensemble, and decision variables ensemble to generate an ensemble of forecasted marginal revenues: Calculate Kalman gain and update the parameter ensemble: Use observed activity and physical constraints to solve the linear programming problem that yields the shadow prices of limiting inputs: subject to: Obtain modelled or observed estimates of available water B water and estimates of used water per crop Obtain satellite derived estimates of crop acreage yield for each crop k = k + 1 Obtain information on other inputs used (e.g. labor) Initial estimate of mean and covariance of parameter vector , and total land available Generate replicates of observed inputs: Generate replicates of prior guess of elasticity of subtitution parameter Members of the parameter ensemble that have been updated out of the feasible space are projected to the closest point in the boundary of the feasible space Use shadow prices and observed costs to generate the ensemble of marginal costs: [email protected] 1) The use of noisy observations of agricultural activity from satellites 2) The recursive estimation of the agroeconomic model parameters 3) A calibration that reects long- or mid- term economic behavior 4) An estimation of the uncertainty about the model predictions 5) An assessment of the value of new satellite-based information Data Assimilation Framework Ensemble Kalman Filter Agroeconomic model Agronomic model parameters get sequentially updated with the latest observations Satellite Data: - Crop Acreage - Yield - Evapotranspiration Hydrologic Data: - Water available - Streamows - Water quality - Diversion points - Well elds Additional Data: - Crop Prices - Irrigation technology - Crop Calendar Agricultural production function Net revenues Revenues Costs Price of crop i Cost of input j Calibration using Positive Mathematical Programming Subject to: Available land Available water Calibration using PMP consists in nding the production function parameters τ, σ, and β, and costs that maximize net revenues under the observed resource allocation and resource constraints. This occurs when marginal net revenues are equal to marginal costs of production. We approximate this maximization problem recursively using an ensemble Kalman lter. Water constraints Observations of resource allocation Dynamics of the parameter ensemble We assimilate seasonal retrievals on of allocated land, water. The enKF lter updates the production function parameters such that the expectation of marginal revenues are as close as possible to the expectation of marginal costs. Full catalog of Landsat and Modis images, geometrically corrected Precise geometric alignment with CDL (also in GEE) GEE premits easy atmospheric correction of Landsat suing LEDAPS Fusion algorithm based on least squre regression, very fast on GEE Water not constraining Main crops: alfalfa and wheat Example focuses on land and water Satellite-based, multi-sensor, annual reclassication of crops CONUS coverage since 2008 High spatial resolution, currently 30m Released in January of the previous year Focus on major summer crops but captures some double cropping Vertical red line is the observed allocation of resources to each crop We simulate the response of farmers to two water shortage scenarios: Scenario 1: 30% reduction in the available water (blue) Scenario 2: 50% reduction in the available water (green) Typical prediction accuracy is within 10% of observations Variance in the distribution of prediction is a direct reection of parameter uncertainty Reduced water and land for alfalfa is reallocated to wheat, less resource intensive Larger reductions in allocated water than in land for alfalfa indicates stress irrigation Vertical red line indicates baseline allocations (i.e. no change) 1) The PMP method in conjuntion with enKF is an effective method to calibrate economic models of agricultural production using noisy data 2) The enKF also permits to calculate predictive uncertainty 3) Calibration is recursive and improved as new observations become available 4) Covergence is robust and opens the door to the use of remotely sensed estimates of crop acreage, yield, and irrigation For mor information see: Howitt and Maneta (2014). Stochastic calibration and learning in non-stationary hydroeconomic models. WRR Land cost parameter Observed yield Identication of production function parameters Grey lines are the 2-98, 5-95 and 32-68 percentiles of the ensemble (darker to lighter tones). Dashed line is the ensemble average Cropland Data Layer, 2007-2014 - Ravalli county Crop acreage Cropland Data Layer (USDA NASS) Landsat-MODIS fusion for agric. applications (using Google Earth Engine) Transformation of NDVI to yield Supplemental irrigation W j,k from NDVI Landsat-MODIS fusion using Google Earth Engine High-resolution, high-frequency NDVI composite Crop phenology Simulation of baseline conditions NDVI from fused Landat/MODIS, sample scene - Weather data Fused NDVI (30 m, 8 days) MODIS NDVI 250 m spatial res. every ~8 days Landsat 5 NDVI 30 m spatial res. every ~15 days

Transcript of Understanding The Hydrologic and Socioeconomic Impacts of Water Use and Resource Allocation Under...

Page 1: Understanding The Hydrologic and Socioeconomic Impacts of Water Use and Resource Allocation Under Different Climate and Policy Scenarios

Und

erst

andi

ng th

e H

ydro

logi

c an

d So

cioe

cono

mic

Impa

cts

of W

ater

Use

and

Res

ourc

e A

lloca

tion

M. M

anet

a1 , P G

ardn

er1 , J

Kim

ball2,

K J

encs

o3,4 , B

Max

wel

l5 , S E

win

g5

1 Geo

scie

nces

, Uni

vers

ity

of M

onta

na; 2 N

umer

ical

Ter

rady

nam

ic S

imul

atio

n G

roup

, U. o

f Mon

tana

; 3 Scho

ol o

f For

estr

y, U

of M

onta

na; 4 M

onta

na C

limat

e O

ffice

; 5 LRES

, Mon

tana

Sta

te U

Iniv

ersi

ty

Intr

oduc

tion

Tem

pera

ture

The p

roje

cted

incr

ease

in t

he f

req

uency

and

in

tensi

ty o

f d

roug

hts

over

ag

ricu

ltura

l re

gio

ns

can

red

uce

the w

ate

r availa

ble

for

irri

gati

on w

ith

imp

ort

ant

eco

log

ical and

eco

nom

ic c

onse

quen

ces.

Info

rmati

on o

n h

ow

farm

ers

acc

om

od

ate

wate

r sh

ort

ag

e is

nece

ssary

for

rob

ust

an

d effi

cient

ag

ricu

ltura

l and

wate

r p

olic

y a

naly

sis.

New

alg

ori

thm

s to

retr

ieve a

gri

cult

ura

l p

ara

mete

rs f

rom

eart

h-o

bse

rvin

g

sate

llite

s ca

n b

e u

sed

for

mic

ro-l

evel analy

sis

of

ag

ricu

ltu

ral p

rod

uct

ion.

This

info

rmati

on c

an b

e u

sed

in c

onju

nct

ion w

ith

eco

nom

ic m

od

els

of

ag

ricu

ltu

ral

act

ivit

y t

o d

esc

rib

e h

ow

farm

ers

will

behave w

hen

face

d w

ith w

ate

r sh

ort

ag

e

or

polic

y c

onst

rain

ts.

Here

we p

rese

nt

a h

yd

roeco

nom

ic m

od

elin

g f

ram

ew

ork

that

ass

imila

tes

rem

ote

ly s

ense

d o

bse

rvati

ons

of

ag

ricu

ltura

l act

ivit

y (

acr

eag

e,

yie

lds,

cro

p

mix

, and

evap

otr

an

spir

ati

on)

into

an a

gro

eco

nom

ic m

od

el ca

libra

ted

usi

ng

Posi

tive M

ath

em

ati

cal Pro

gra

mm

ing

(PM

P).

PM

P is

wid

ely

use

d in a

gri

cult

ura

l

polic

y a

naly

sis

to c

alib

rate

mod

els

that

pre

cise

ly r

ep

rod

uce

th

e o

bse

rved

eco

nom

ic a

ctiv

ity levels

of

a b

ase

year.

An im

port

ant

chara

cteri

stic

of

mod

els

calib

rate

d u

sing

PM

P is

that

they r

eflect

the a

ctual b

eh

avio

r of

farm

ers

and

th

eir

act

ual re

act

ion t

o e

xte

rnal fa

ctors

, su

ch a

s d

roug

ht

and

risk

, ra

ther

than t

he b

ehavio

r th

at

would

be o

pti

mal fr

om

an a

gro

nom

ic p

oin

t

of

vie

w.

We e

xte

nd

th

e c

lass

ic P

MP a

pp

roach

by im

ple

menti

ng

it

wit

hin

a d

ata

ass

imila

tion f

ram

ew

ork

base

d o

n a

n e

nse

mb

le K

alm

an F

ilter.

The e

xte

nsi

on

perm

its:

Hyd

ro-e

cono

mic

mod

els:

Obj

ecti

ves

The m

od

elin

g f

ram

ew

ork

pre

sente

d h

ere

perm

its

the o

pera

tionaliz

ati

on o

f

new

eart

h o

bse

rvin

g c

ap

ab

iliti

es

to p

red

ict

chang

es

in land

and

wate

r use

in

ag

ricu

ltura

l re

gio

ns,

to p

red

ict

the im

pact

of

such

ch

ang

es

in t

he h

yd

rolo

gic

netw

ork

, an

d t

o info

rm a

gri

cult

ura

l and

wate

r p

olic

y.

Sp

ecifica

lly,

the m

eth

od

s ca

n g

ive insi

gh

t in

to t

he f

ollo

win

g q

uest

ions:

1.

How

do d

rou

gh

ts aff

ect

cro

p m

ix a

nd

lan

d u

se d

evote

d t

o

ag

ricu

ltu

re?

2.

How

does a

gri

cu

ltu

ral ch

an

ge im

pact

wate

r availab

ilit

y a

nd

oth

er

wate

r u

sers

?

3.

Wh

at

ch

oic

es a

nd

man

ag

em

en

t p

racti

ces a

re p

riori

tized

, an

d

how

can

th

at

info

rm f

utu

re p

olicy o

pti

on

s?

4.

How

can

new

rem

ote

sen

sin

g in

form

ati

on

be in

teg

rate

d in

mod

els

to p

rovid

e in

sig

ht

into

socio

econ

om

ic d

yn

am

ics?

Rem

ote

sens

ing

of a

gric

ultu

ral a

ctiv

ity

Exam

ple

for

Ravalli cou

nty

, M

on

tan

a

Hyd

ro-e

cono

mic

mod

el o

f agr

icul

tura

l act

ivit

y

Posi

tive

Mat

hem

atic

al P

rogr

amm

ing

wit

h th

e en

KF

Sate

llite

pro

duct

s and a

lgori

thm

s availa

ble

to info

rm r

eso

urc

e a

lloca

tion incl

ude:

- L

and a

lloca

tion a

nd c

rop m

ix:

USD

A-N

ASS C

ropla

nd D

ata

Layer

(CD

L) (

annu

al fo

r co

nte

rmin

ous

US)

- C

rop y

ield

: E

stim

ate

d f

rom

radia

tion u

se effi

ciency

(D

ora

isw

am

y e

t al, 2

00

5;

Lobell

et

al., 2

00

3)

- S

upple

men

tal Ir

rigati

on:

ET e

stim

ate

s base

d o

n o

bse

rved c

rop p

henolo

gy, w

eath

er

data

an

d irr

igati

on

effi

ciency

fact

ors

Text

here

...

Hyd

ro-e

cono

mic

mod

el a

t sim

ulat

ion

tim

e

Beha

vior

und

er re

sour

ce c

onst

rain

ts is

sim

ulat

ed b

y su

bstit

utin

g th

e id

entifi

ed

para

met

ers

into

the

prod

uctio

n fu

nctio

n an

d fin

ding

the

reso

urce

mat

rix X

that

sol

ves

the

net r

even

ue m

axim

izat

ion

prob

lem

. To

test

the

mod

el w

e si

mul

ate

farm

er b

ehav

ior

unde

r kno

wn

base

line

cons

train

ts.

Conc

lusi

ons

and

mai

n fin

ding

s

Para

mete

r d

istr

ibu

tions

con

verg

e t

o a

ste

ad

y v

alu

e in less

th

an 5

ass

imila

tion c

ycl

es

Pre

dic

tive v

ari

ance

of

the p

ara

mete

rs is

rap

idly

red

uce

d f

rom

the in

itia

l valu

es

The p

ara

mete

rense

mb

le d

ecr

ease

s b

ut

does

not

colla

pse

to a

sin

gle

poin

t (s

tab

ility

)

Pote

nti

al cau

se o

f b

ias

Pre

cip

itati

on u

nd

erc

atc

h d

ue t

o w

ind

eff

ect

s.

Poor

rep

rese

nta

tiveness

of

the o

bse

rvati

on n

etw

ork

fro

m h

igh g

rid

-cell

vari

ab

ility

.

Sta

tion loca

tion b

ias

tow

ard

low

er

(an

d fl

att

er)

terr

ain

.

Resp

on

se t

o r

ed

uced

access t

o w

ate

r

The h

eart

of

the a

gro

nom

ic m

odel is

a

para

metr

ic a

gri

cult

ura

l pro

duct

ion f

unct

ion q

i

that

calc

ula

tes

yie

ld o

f cr

op i g

iven t

he

allo

cate

d f

arm

ing r

eso

urc

es j em

beded in

matr

ix X

.

The m

odel ass

um

es

that

farm

ers

will

allo

cate

their

availa

ble

reso

urc

es

wit

h t

he o

bje

ctiv

e o

f

maxim

izin

g t

heir

net

revenues

Text h

ere...

Refe

renc

esLo

bell,

D. B

., A

sner,

G. P.

, O

rtiz

-Monast

eri

o, J. I. and B

enn

ing, T.

L.:

Rem

ote

sensi

ng o

f re

gio

nal cr

op p

roduct

ion

in t

he Y

aqui Valle

y, M

exic

o:

est

imate

s an

d u

nce

rtain

ties,

Agri

c. E

cosy

st. En

vir

on., 9

4(2

), 2

05

–22

0

Maneta

, M

.P. an

d H

ow

itt

(20

14

). S

toch

ast

ic c

alib

rati

on a

nd d

ata

ass

imila

tion in n

on

-sta

tionary

hydro

-eco

nom

ic

models

. W

ate

r R

eso

urc

es

Rese

arc

h,

50

(5):

39

76

–39

93

Maneta

, M

. P.

, To

rres,

M.

O., W

alle

nder,

W. W

., V

ost

i, S

., H

ow

itt,

R. E., R

odri

gues,

L., B

ass

oi, L

. H

. and P

anday,

S.:

A

spati

ally

dis

trib

ute

d h

ydro

eco

nom

ic m

odel to

ass

ess

the eff

ect

s of

dro

ught

on

land u

se, fa

rm p

rofits

, and

agri

cult

ura

l em

plo

ym

ent,

Wate

r R

eso

ur.

Res.

, 4

5, W

11

41

2

Roy e

t al. (

20

10

). W

eb-e

nable

d L

an

dsa

t D

ata

(W

ELD

): L

andsa

t ETM

+ C

om

posi

ted M

osa

ics

of

the C

on

term

inous

Unit

ed S

tate

s, R

em

ote

Sensi

ng o

f Envir

onm

ent,

11

4:

35

-49

.

Con

stru

ct t

he p

rodu

ctio

n fu

ncti

on p

aram

eter

ens

embl

e by

sam

plin

g fr

om t

he s

moo

thed

mix

ture

of

Gau

ssia

n de

nsit

ies:

Use

the

par

amet

er e

nsem

ble,

and

dec

isio

n va

riab

les

ense

mbl

e to

gen

erat

ean

ens

embl

e of

for

ecas

ted

mar

gina

l re

venu

es:

Cal

cula

te K

alm

an g

ain

and

upda

te t

he p

aram

eter

ens

embl

e:

Use

obse

rved a

cti

vit

y a

nd p

hysi

cal const

rain

ts t

o s

olv

e t

he lin

ear

pro

gra

mm

ing p

roble

m t

hat

yie

lds

the s

hadow

pri

ces

of

lim

itin

g inputs

:

subj

ect

to:

Obt

ain

mod

elle

d or

obs

erve

d es

tim

ates

of

avai

labl

ew

ater

Bw

ater

and

esti

mat

es o

f us

ed w

ater

per

cro

pO

btai

n sa

tell

ite

deri

ved

esti

mat

es o

f cr

op a

crea

geyi

eld

for

each

cro

p

k =

k +

1

Obt

ain

info

rmat

ion

on o

ther

inp

uts

used

(e

.g. l

abor

)

Init

ial

esti

mat

e of

mea

n an

dco

vari

ance

of

para

met

er v

ecto

r

, and

tot

al l

and

avai

labl

e

Gen

erat

e re

plic

ates

of

obse

rved

inp

uts:

Gen

erat

e re

plic

ates

of

prio

r gu

ess

of e

last

icit

y of

sub

titu

tion

par

amet

er

Mem

bers

of

the

para

met

er e

nsem

ble

that

hav

e be

enup

date

d ou

t of

the

fea

sibl

e sp

ace

are

proj

ecte

dto

the

clo

sest

poi

nt i

n th

e bo

unda

ry o

f th

e fe

asib

le s

pace

Use

sha

dow

pri

ces

and

obse

rved

cos

ts t

o ge

nera

te t

he e

nsem

ble

of m

argi

nal

cost

s:

mar

co.m

anet

a@um

onta

na.e

du

1) T

he u

se o

f noi

sy o

bser

vatio

ns o

f agr

icul

tura

l act

ivity

from

sat

ellit

es2)

The

recu

rsiv

e es

timat

ion

of th

e ag

roec

onom

ic m

odel

par

amet

ers

3) A

cal

ibra

tion

that

refle

cts

long

- or m

id- t

erm

eco

nom

ic b

ehav

ior

4) A

n es

timat

ion

ofth

e un

cert

aint

y ab

outt

he m

odel

pre

dict

ions

5) A

n as

sess

men

t of t

he v

alue

of n

ew s

atel

lite-

base

d in

form

atio

n

Data

Ass

imila

tion F

ram

ew

ork

Ense

mb

le K

alm

an F

ilter

Ag

roeco

nom

ic m

odel

Ag

ronom

ic m

od

el p

ara

mete

rsg

et

seq

uenti

ally

up

date

dw

ith t

he late

st o

bse

rvati

ons

Sate

llite

Data

:-

Cro

p A

creag

e-

Yield

- Evap

otr

ansp

irati

on

Hydro

logic

Data

:-

Wate

r availa

ble

- S

treamflow

s-

Wate

r q

ualit

y-

Div

ers

ion p

oin

ts-

Wellfield

s

Ad

dit

ional D

ata

:-

Cro

p P

rice

s-

Irri

gati

on t

ech

nolo

gy

- C

rop

Cale

ndar

Ag

ricu

ltu

ral p

rod

ucti

on

fu

ncti

on

Net

reven

ues

Reven

ues

Costs

Pric

e of

cro

p i

Cost

of i

nput

j

Calib

rati

on

usin

g P

osit

ive M

ath

em

ati

cal P

rog

ram

min

g

Su

bje

ct

to:

Avai

labl

e la

nd

Avai

labl

e w

ater

Calib

rati

on u

sing

PM

P c

on

sist

s in

find

ing

the p

rod

uct

ion f

unct

ion p

ara

mete

rs

τ, σ,

and

β,

and

cost

s th

at

maxim

ize n

et

revenues

und

er

the o

bse

rved

reso

urc

e

allo

cati

on a

nd

reso

urc

e c

onst

rain

ts.

This

occ

urs

when m

arg

inalnet

revenues

are

eq

ual

to m

arg

inal co

sts

of

pro

duct

ion.

We a

pp

roxim

ate

this

maxim

izati

on p

rob

lem

recu

rsiv

ely

usi

ng

an e

nse

mb

le K

alm

an fi

lter.

Wate

r con

str

ain

ts

Ob

serv

ati

on

s o

f re

sou

rce a

llocati

on

Dyn

am

ics o

f th

e p

ara

mete

r en

sem

ble

We a

ssim

ilate

seaso

nal re

trie

vals

on

of

allo

cate

d land

, w

ate

r. T

he e

nK

F filt

er

up

date

s th

e p

rod

uct

ion f

unct

ion p

ara

mete

rs s

uch

that

the e

xp

ect

ati

on o

f

marg

inal re

venues

are

as

close

as

poss

ible

to t

he e

xp

ect

ati

on o

f m

arg

inal co

sts.

Full

cata

log

of

Land

sat

and

Mod

is im

ag

es,

geom

etr

ically

corr

ect

ed

Pre

cise

geom

etr

ic a

lignm

en

t w

ith

CD

L (a

lso in G

EE)

GEE p

rem

its

easy

atm

osp

heri

c co

rrect

ion o

f La

nd

sat

suin

g L

ED

APS

Fusi

on a

lgori

thm

base

d o

n least

sq

ure

reg

ress

ion,

very

fast

on G

EE

Wate

r n

ot

const

rain

ing

Main

cro

ps:

alfalfa a

nd

wheat

Exam

ple

focu

ses

on lan

d a

nd

wate

r

Sate

llite

-base

d,

mult

i-se

nso

r, a

nnual re

class

ifica

tion o

f cr

op

s

CO

NU

Sco

vera

ge s

ince

20

08

Hig

h s

pati

al re

solu

tion,

curr

entl

y 3

0m

R

ele

ase

d in Jan

uary

of

the p

revio

us

year

Focu

s on

majo

r su

mm

er

crops

but

captu

res

som

e d

ouble

cro

ppin

g

Vert

ical

red

line

is th

e ob

serv

ed a

lloca

tion

of re

sour

ces

to e

ach

crop

We

sim

ulat

e th

e re

spon

se o

f far

mer

s to

two

wat

er s

hort

age

scen

ario

s:Sc

enar

io 1

: 30%

redu

ctio

n in

the

avai

labl

e w

ater

(blu

e)Sc

enar

io 2

: 50%

redu

ctio

n in

the

avai

labl

e w

ater

(gre

en)

Typ

icalp

red

icti

on a

ccura

cy is

wit

hin

10

%of

ob

serv

ati

ons

Vari

an

ce in t

he d

istr

ibuti

on o

f p

red

icti

on is

a d

irect

reflect

ion o

f p

ara

mete

r unce

rtain

ty

Red

uce

d w

ate

r an

d land

for

alfalfa is

reallo

cate

d t

o w

heat,

less

reso

urc

e inte

nsi

ve

Larg

er

red

uct

ions

in a

lloca

ted

wate

r th

an

in lan

d f

or

alfalfa

ind

icate

s st

ress

irr

igati

on

Vert

ical

red

line

indi

cate

s ba

selin

e al

loca

tions

(i.e

. no

chan

ge)

1) T

he P

MP

met

hod

in c

onju

ntio

n w

ith e

nKF

isan

eff

ectiv

e m

etho

d to

cal

ibra

te

econ

omic

mod

els

of a

gric

ultu

ral p

rodu

ctio

n us

ing

nois

y da

ta2)

The

enK

F al

so p

erm

its to

cal

cula

te p

redi

ctiv

e un

cert

aint

y3)

Cal

ibra

tion

is re

curs

ive

and

impr

oved

as

new

obs

erva

tions

bec

ome

avai

labl

e4)

Cov

erge

nce

is ro

bust

and

ope

ns th

e do

or to

the

use

of re

mot

ely

sens

ed e

stim

ates

of

crop

acr

eage

, yie

ld, a

nd ir

rigat

ion

For m

or in

form

atio

n se

e: H

owitt

and

Man

eta

(201

4). S

toch

astic

cal

ibra

tion

and

lear

ning

in n

on-s

tatio

nary

hyd

roec

onom

ic m

odel

s. W

RR

Land

cos

tpa

ram

eter

Obs

erve

d yi

eld

Iden

tific

atio

n of

pro

duct

ion

func

tion

par

amet

ers

Gre

y lin

es

are

the 2

-98

, 5

-95

an

d 3

2-6

8 p

erc

enti

les

of

the e

nse

mb

le (

dark

er

to

lighte

r to

nes)

. D

ash

ed

lin

e is

the e

nse

mb

le a

vera

ge

Cro

pla

nd

Data

Layer,

2007-2

014 -

Ravalli cou

nty

Crop acreage

Cro

pla

nd

Data

Layer

(US

DA

NA

SS

)

Lan

dsat-

MO

DIS

fu

sio

n f

or

ag

ric.

ap

plicati

on

s (

usin

g G

oog

le E

art

h E

ng

ine)

Tran

sfo

rmati

on

of

ND

VI

to y

ield

Su

pp

lem

en

tal ir

rig

ati

on

Wj,

k f

rom

ND

VI

Land

sat-

MO

DIS

fusi

on u

sing

Goog

le E

art

h E

ng

ine

Hig

h-r

esolu

tion

, h

igh

-fre

qu

en

cy N

DV

I com

posit

eC

rop

ph

en

olo

gy

Sim

ula

tion

of

baselin

e c

on

dit

ion

sN

DV

I fr

om

fu

sed

Lan

dat/

MO

DIS

, sam

ple

scen

e

- W

eath

er

data

Fuse

d N

DV

I

(30

m,

8 d

ays)

MO

DIS

ND

VI

25

0 m

spati

al re

s.

every

~8

days

Landsa

t 5

ND

VI

30

m s

pati

al re

s.

every

~1

5 d

ays