gis crop growth maize zhangye

Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/264536086

Assimilatingremotesensinginformationintoacoupledhydrology-cropgrowthmodeltoestimateregionalmaizeyieldinaridregionsARTICLEinECOLOGICALMODELLINGNOVEMBER2014ImpactFactor:2.07DOI:10.1016/j.ecolmodel.2014.07.013

CITATIONS2

DOWNLOADS68

VIEWS60

6AUTHORS,INCLUDING:

YanLiLanzhouUniversity6PUBLICATIONS17CITATIONS

SEEPROFILE

QingguoZhouLanzhouUniversity59PUBLICATIONS131CITATIONS

SEEPROFILE

G.F.ZhangLanzhouUniversity6PUBLICATIONS4CITATIONS

SEEPROFILE

ChongChen2PUBLICATIONS3CITATIONS

SEEPROFILE

Availablefrom:YanLiRetrievedon:24June2015

Ecological Modelling 291 (2014) 1527

Contents lists available at ScienceDirect

Ecological Modelling

journa l h om epa ge: www.elsev ier .com/ locate /eco lmodel

Assimilating remote sensing information into a chydrol ionregions

Yan Lia, Q ng a School of Infob Cold and Arid es, Lac Lanzhou Insti ClimaProvince, Key O l Adm

a r t i c l e i n f o

Article history:Received 25 JaReceived in reAccepted 12 Ju

Keywords:EnKFCoupled hydroRemote sensinEFAST

a b s t r a c t

Regional crop yield prediction is a signicant component of national food policy making and security

1. Introdu

Crop grofood securitof factors, incropland action, and gproduction crop growt

CorrespondE-mail add

(Q. Zhou), zhochenchong87@

http://dx.doi.o0304-3800/Crnuary 2014vised form 11 July 2014ly 2014

logy-crop growth modelg information

assessments. A data assimilation method that combines crop growth models with remotely sensed datahas been proven to be the most effective method for regional yield estimates. This paper describes anassimilation method that integrates a time series of leaf area index (LAI) retrieved from ETM+ data anda coupled hydrology-crop growth model which links a crop growth model World Food Study (WOFOST)and a hydrology model HYDRUS-1D for regional maize yield estimates using the ensemble Kalman lter(EnKF). The coupled hydrology-crop growth model was calibrated and validated using eld data to ensurethat the model accurately simulated associated state variables and maize growing processes. To identifythe parameters that most affected model output, an extended Fourier amplitude sensitivity test (EFAST)was applied to the model before calibration. The calibration results indicated that the coupled hydrology-crop growth model accurately simulated maize growth processes for the local cultivation variety tested.The coefcient of variations (CVs) for LAI, total above-ground production (TAGP), dry weight of storageorgans (WSO), and evapotranspiration (ET) were 13%, 6.9%, 11% and 20%, respectively. The calibratedgrowth model was then combined with the regional ETM+ LAI data using a sequential data assimilationalgorithm (EnKF) to incorporate spatial heterogeneity in maize growth into the coupled hydrology-cropgrowth model. The theoretical LAI prole for the near future and the nal yield were obtained throughthe EnKF algorithm for 50 sample plots. The CV of the regional yield estimates for these sample plots was8.7%. Finally, the maize yield distribution for the Zhangye Oasis was obtained as a case study. In general,this research and associated model could be used to evaluate the impacts of irrigation, fertilizer and eldmanagement on crop yield at a regional scale.

Crown Copyright 2014 Published by Elsevier B.V. All rights reserved.

ction

wth and yield data are critical for informing nationaly and agricultural operation and management. A rangecluding increasingly augmented populations, reducedreage and water resources, environmental deteriora-lobal climate change, signicantly affect agriculturaland threaten food security. Therefore, accurate regionalh monitoring and yield prediction are vital for the

ing author. Tel.: +86 931 4967724.resses: liyan [email protected] (Y. Li), [email protected]@lzb.ac.cn (J. Zhou), [email protected] (G. Zhang),gmail.com (C. Chen), [email protected] (J. Wang).

sustainable development of agriculture. In the past decades, cropgrowth monitoring and crop yield predictions have moved awayfrom conventional techniques, such as agro-meteorological modelforecasting and the establishment of relationships between remotesensing spectral vegetation indexes and eld measurements, infavour of more integrated approaches (e.g. Dente et al., 2008; deWit et al., 2012; Fang et al., 2011; Liang and Qin, 2008; Ma et al.,2013; Xu et al., 2011). For instance, operational systems for regionalcrop monitoring and yield forecasting now usually rely on an inte-grated analysis of weather data, crop simulation model results andsatellite observations.

Using the biophysical principles of plant growth, crop growthmodels can provide detailed estimates of crop states, including phe-nological status, leaf area index (LAI) and yield for specic croptypes. The most commonly used crop growth model, WOFOST,

rg/10.1016/j.ecolmodel.2014.07.013own Copyright 2014 Published by Elsevier B.V. All rights reserved.ogy-crop growth model to estimate reg

ingguo Zhoua, Jian Zhoub,, Gaofeng Zhanga, Chormation Science and Engineering, Lanzhou University, Lanzhou 730000, China

Regions Environmental and Engineering Research Institute, Chinese Academy of Scienctute of Arid Meteorology, China Meteorological Administration, Key Laboratory of Arid pen Laboratory of Arid Climate Change and Disaster Reduction of China Meteorologicaoupledal maize yield in arid

Chena, Jing Wangc

nzhou 730000, Chinatic Change and Reducing Disaster of Gansuinistration, Lanzhou 730000, China

16 Y. Li et al. / Ecological Modelling 291 (2014) 1527

has been applied to evaluations of production potential, optimizedcrop management, and yield gap quantication for various crops(Van Laar et al., 1997; Bouman et al., 2001; Wolf, 2002) in additionto studies of the effects of environmental variability and climaticchange on 1997; Tsujiin WOFOSTbucket apptransmissioconditions. logic cycle et al., 2006main driverShepherd e2004), andunderstandWOFOST mneeded. Linfrom researa land surfamodel (DSSvegetation.and crop grchanges onden Hoof e(JULES) andlogical cycluxes.

Crop moat the eldrestricted bwhich resuquate accurMoulin anddifculties iof remote srecognized conditions 2008; de Wet al., 2012;been madeinformationRevill et al.,actual growcies from noVegetation rent year cgiven crop (data can beincluding, fsensitive cr

In geneand crop gilation. The(1) recalibassimilationinitializatioto re-initialting initial function besimulated p2013). Howsuch as statFor sequenthat improvcan enhanc

approach can combine many types of observations taken at dis-crete time steps, the state variables can be continually updatedand more accurately simulated. In particular, the EnKF, which isa sequence lter algorithm that combines a probabilistic approach

quenin re2007s in ous

nt dend ed Vanhis styieldy intOFOStima

teria

udy a

appangyHeihperatentim, rtes m

tharesound th

bserv

008,R) pncesgionltura

founal, bi

our fromReseogicantervand iatiopera

lative statgrowber ere

wery Anae org) and. Da

(ECd he

EC dsity crop production (Kropff et al., 1996; ten Berge et al., et al., 1998; Matthews and Stephens, 2002). However,, the soil water balance is calculated using a tippingroach within three compartments (i.e. a root zone, an zone, and a groundwater zone) under water-limitedThis approach oversimplies simulations of the hydro-in crop growth models (Eitzinger et al., 2004; Priesack). In reality, variation in available soil moisture is a

of variation in crop yield (Rodriguez-Iturbe et al., 2001;t al., 2002; Anwar et al., 2003; Patil and Sheelavantar,

accurate estimates of soil moisture are critical foring crop status and yield. To enhance the ability of theodel to simulate the hydrologic cycle, model coupling iskages between models have recently received attentionchers. For example, Casanova and Judge (2008) coupledce process (LSP) model with a widely used crop-growthAT) to estimate energy and moisture uxes in dynamic

Maruyama and Kuwagata (2010) linked land surfaceowth models to estimate the effects of growing season

the energy balance and water use in rice paddies. Vant al. (2011) coupled the Land Environment Simulator

a crop-growth model (SUCROS) to evaluate the hydro-e and vegetation effects on energy, water, and carbon

dels are very useful in evaluating crop growth and yield scale, but their implementation at a regional scale isy input data availability at the corresponding scale,lts in models that may not produce results with ade-acy for practical applications (Chipnashi et al., 1999;

Guerif, 1999; Mignolet et al., 2007). To overcome thenherent to these modelling approaches, the integrationensing observations and crop growth models has beenas an important approach for monitoring crop growthand estimating yield at a regional scale (Dente et al.,it et al., 2012; Fang et al., 2011; Liang and Qin, 2008; Ma

Xu et al., 2011). In recent years, extensive efforts have to integrate crop growth models with remote sensing

based on the assimilation strategy (Chen et al., 2012; 2013). Satellite data can provide a synoptic overview ofing conditions and can be used to diagnose discrepan-rmal conditions. For example, a Normalized DifferenceIndex (NDVI; Rouse and Haas, 1973) prole of the cur-ould be compared to the average historic prole for aKogan, 1998; Liu and Kogan, 2002). Moreover, satellite

used to complement crop model simulation results byor example, the impacts of re, frost, or drought duringop stages.ral, the integration of remote sensing observationsrowth models is often achieved through data assim-re are two basic strategies for data assimilation:ration/re-initialization methods and (2) sequential

algorithms, such as the EnKF. The recalibration/re-n approach commonly uses optimization algorithmsize or re-parameterize a crop growth model by adjus-conditions or input parameters to minimize the merittween remotely sensed biophysical parameters andarameters (Dente et al., 2008; Fang et al., 2011; Ma et al.,ever, this method cannot incorporate dynamic changes,e variable updates or dynamic crop growth simulations.tial assimilation algorithms, a primary assumption isements in a state variable made in the previous step

e the estimation accuracy in the next step. Because this

with setainty et al., teristicVazifeddiffereoped aWit an

In tmaize imagerthe Wcan ulscale.

2. Ma

2.1. St

Wethe Zhof the cal temand po2000 mcultivasystemwater River a

2.2. O

In 2(WATEof Sciethis reagricucan bephysicused innantlya.s.l.). hydroldata (ierties, net radair temand re

Theentire Septemdate won LAICanopstorag(TAGPgrowthsysteminstalletion ofUnivertial data assimilation, can account for sequential uncer-motely sensed observations (Curnel et al., 2011; Dorigo; Quaife et al., 2008) and nonlinear structural charac-crop growth models (de Wit and Van Diepen, 2007;t et al., 2009). Several EnKF assimilation schemes withgrees of complexity and integration have been devel-valuated during the last decade (Curnel et al., 2011; de

Diepen, 2007; Vazifedoust et al., 2009; Jin et al., 2010).udy, we present a novel method for estimating regional

that assimilates LAI values derived from remote sensingo a coupled hydrology-crop growth model (which linksT and HYDRUS models) using an EnKF. This methodtely be used to estimate crop yields at a regional

l and methods

rea

lied our coupled model to the agricultural system ine Oasis (Fig. 1), an arid region in the middle reachese River basin, northwest China. This area has a typi-te continental climate, with mean annual precipitational evaporation ranging from 60 to 280 mm and 1000 toespectively. This agricultural system, which primarilyaize and wheat, employs a highly developed irrigation

t was constructed in the last few decades. The mainrce for irrigation in this area originates from the Heihee groundwater.

ation data

the Watershed Allied Telemetry Experimental Researchrogram (Li et al., 2009), part of the Chinese Academy Action Plan for Western Development, was applied to

to study the ecological and hydrological processes ofl systems (methodological information for this programd at http://westdc.westgis.ac.cn/data). The primary bio-ochemical, meteorological and hydrological parameters

study were obtained through this program, predomi- the Yingke station (3851N, 10025 E, altitude 1519 march at this agro-ecological station focused on eco-l processes during maize growth periods and providedal: 30 min) on regional meteorology, physical soil prop-soil moisture dynamics. Meteorological data includedn, solar radiation, maximum air temperature, minimumture, precipitation, wind speed, atmospheric pressure,

humidity.us of maize was intensively monitored throughout itsth period, which lasted from April 20, 2008 through22, 2008. The sowing date, emergence date and harvestApril 20, May 6, and September 22, respectively. Datae measured once every 15 days using a LAI-2000 Plantlyzer (LI-COR Inc., Lincoln, NE, USA). The dry weight ofans (WSO), dry weight of total above-ground biomass

crop height were sampled every 15 days during cropta on latent ux were measured by an eddy covariance) (Li7500 & CSAT3, Cambell Scientic, USA) that wasre for long-term use at a height of 2.85 m. The correc-ata was produced with revised EdiRE software from theof Edinburgh (Xu et al., 2008). Nitrogen (329 kg ha1),

Y. Li et al. / Ecological Modelling 291 (2014) 1527 17

potassium were applie

2.3. Remote

Sensors leled data set al., 1980;et al., 2005;sors providat relativelyOLI sensorsthe ETM+ dthe GeospatChinese AcETM+ imagered the fumodel (Tabusing groundigital elevathe Fast Lin(FLAASH) m

2.4. The cou

In this strst presenmodel WOFused to simzones HYDRindicator fothe ratio b

spiration ra

igatium aily r

. by Fig. 1. Study region.

(220 kg ha1) and phosphorus (87 kg ha1) fertilizersd to the maize elds throughout the growth period.

sensing data

The irrmaximthe da

modelculatedonboard the Landsat satellite series provide an unparal-ource for land surface mapping and monitoring (Byrne

Cohen and Goward, 2004; Hansen et al., 2008; Healey Masek et al., 2008; Vogelmann et al., 2001). Landsat sen-e long-term, repeated, global coverage (1972present)

high spatial resolutions (30 m for the TM, ETM+ and; 80 m for the MSS sensors). In this study, we usedataset (L1T product for path 133, row 33) provided byial Data Cloud, Computer Network Information Centre,ademy of Sciences (http://www.gscloud.cn). Four L1Tes, which were captured on cloud-free days that cov-ll maize growth period, were selected for use in ourle 1). The L1T ETM+ data were geometrically correctedd control points and topographically corrected using ation model. Atmospheric correction was applied using

e-of-sight Atmospheric Analysis of Spectral Hypercubesodel in ENVI (RSI, 2001).

pled hydrology-crop growth model

udy, we used a coupled hydrology-crop growth modelted elsewhere (Li et al., 2012). The dynamic crop growthOST (Boogaard et al., 1998) and the hydrological modelulate water ow and root water uptake in unsaturatedUS-1D (Simunek et al., 2005) were linked through anr the degree of water stress, which was dened asetween actual water uptake rate and potential tran-

te. The basic coupling process is shown in Fig. 2:

1965; Montis calculate

the HYDRU

groundwatThe root waassumed thcrop transpwater uptation based o

for the degCO2 assimiWOFOST mlate the actuamong diffe

model. Tmodel, morare then us

Throughdaily time sthe coupledfrom WOFOdescription

2.5. Sensiti

A globagrowth modon and precipitation, the daily net radiation, the dailynd minimum temperatures, the daily wind speed and

elative humidity are the input terms in the HYDRUS

The potential evaporation and transpiration are cal-the Penman-Monteith combination method (Monteith,eith, 1981) in the HYDRUS model. The water uptaked according to Feddes equation (Feddes et al., 1978) in

S model. The soil water balance, soil moisture and

er depth are calculated using the HYDRUS model.ter uptake and actual transpiration on a daily basis aree same, because most root water uptake is consumed byiration. Therefore, the ratio between calculated actualke based on Feddes equation and potential transpira-n Penman-Monteith method is regarded as an indicator

ree of water stress. The potential daily total grosslation of the crop, which is calculated according to theodel, is multiplied by the water stress ratio to calcu-al daily CO2 assimilation. Then, carbohydrate allocationrent crop parts is calculated according to the WOFOST

he calculated vegetation parameters from the WOFOSTe specically rooting depth, height of the crop and LAI,ed as inputs for the HYDRUS model at the next step.

this framework, the modules were fully coupled at atep. Two types of parameter sets were required to run

hydrology-crop growth model: (1) crop parametersST and (2) soil parameters from HYDRUS. A detailed

of this coupled model can be found in Li et al. (2012).

vity analysis

l sensitivity analysis of the coupled hydrology-cropel has been implemented with the EFAST method using


Table 1Information of selected ETM+ images.

Passing date Path Raw Center location () Solar azimuth () Solar elevation () Mean cloudiness (%)

2008-5-12 133 33 38.86N, 100.62E 130.9 61.8 1.22008-5-28 133 33 38.87N, 100.62E 125.8 64.2 0.362008-7-15 133 33 38.90N, 100.63E 122.5 62.7 7.62008-8-16 133 33 38.93N, 100.63E 133.2 56.9 17.3

a SimLab Dynamic Link Library (http://simlab.jrc.ec.europa.eu/).Only model parameters were analyzed. Neither input variables northe initial conditions of the coupled model take part in the sensi-tivity analysis. The appropriate ranges for the input factors usedin the sensitivity analysis were dened based on values from a lit-erature review, experience, and research objectives as well as thedefault, minimum, and maximum values used in the WOFOST andHYDRUS databases. Uniform distributions were assigned to inputfactors when only the base value was known and when the rangewas considered nite if no explicit knowledge of the input fac-tors distribution was available (McKay, 1995). This conservativeassumption allowed for an equal probability of occurrence for theinput factors along the probability range (Munoz-Carpena et al.,2010). The Monte Carlo sample size was set to 5000 for each fac-tor. The parameters were divided into 11 groups according to theirphysical properties and functions. These parameter groups andtheir value ranges are shown in Table 2. A detailed description ofEFAST can be found in Saltelli et al. (2000).

2.6. Coupled hydrology-crop model calibration

Because the maize characteristics (e.g. drought resistance, coldresistance, plant height, leaf width, full grain of the maize vari-ety) for our study area were similar to those for the grain maize

203 variety cultivated in Europe, we used the maize dataset pro-vided by the European Community (Boons-Prins et al., 1993) toset the basic crop parameters in the model. The crop parametergroups with high rst-order and interaction indices were calibratedusing the FSEOPT optimization algorithm developed by Stol et al.(1992). The default values in the basic parameter dataset were usedfor all other parameters (MAG 203). Two parameters related totemperature accumulation, the temperature sum from emergenceto anthesis (TSUM1), and the temperature sum from anthesis tomaturity (TSUM2) were calculated by adding the daily average tem-peratures (which were above 0 C; biological zero point) from thecollected meteorological data.

The soil hydraulic parameters, including saturated water con-tent s (L3 L3), residual water content r (L3 L3), saturatedhydraulic conductivity Ks (L T1), the air entry parameter , the poresize distribution parameter n, and the pore connectivity parameterl, were obtained from previous research (Li et al., 2012).

2.7. Ensemble Kalman lter

The EnKF algorithm, proposed by Evensen (1994) and laterimproved by Burgers et al. (1998), is based on Monte Carlo orensemble generations, where the approximation of an a priori stateerror covariance matrix is forecasted by propagating an ensembleFig. 2. Illustration of the coupled hydrology-crop growth model.


Table 2The groups of parameters and the value ranges of parameters for EFAST.

Group Parameter Meaning Unit Values range

Soil hydraulparameters

(cm3 cm3) r U(0.010.1)

Emergenceratureperat

Phenologyrgencesis to

LAIase ingence

Green areaing atnction

Assimilation

Conversion oassimilates ibiomass

Maintenancerespiration

Death rates

Correction fa

Root parame

of model stthe previouble of obserfrom a distrto the obseReichle et alto land dataVan Diepen2009).

The basiing to the fo

Aa = A + Pewhere Aa anfor the staic(HYDRUS)

Parameters of HYDRUSmodel

Soil hydraulicparameters

TBASEM Lower threshold tempeTEFFMX Maximum effective tem

TSUM1 Thermal time from emeTSUM2 Thermal time from anth

RGRLAI Maximum relative increLAIEM Leaf area index at emer

SPAN Life span of leaves growSLATB Specic leaf area as a fu

stage

SLATB1 Specic leaf area as a function

stage

AMAXTB Maximum leaf CO2 assimilatidevelopment stage of the crop

AMAXTB1 Maximum leaf CO2 assimilatidevelopment stage of the crop




fnto

CVO Conversion efciency of assimorgan

CVS Conversion efciency of assimCVL Conversion efciency of assimCVR Conversion efciency of assim

RMS Relative maintenance respiratRML Relative maintenance respiratQ10 Relative change in respiration

temperature changeRMO Relative maintenance respirat

organsRMR Relative maintenance respirat

due to water stress PERDL Maximum relative death ratewater stress

ctor transpiration rate CFET correction factor transpiration

tersRRI Maximum daily increase in roRDI Initial rooting depth RDMCR maximum rooting depth

ates using updated states (ensemble members) froms time step. The EnKF algorithm generates the ensem-vations at each update time by introducing noise drawnibution with mean equal to zero and covariance equalrvational error covariance matrix (Burgers et al., 1998;., 2002a,b). The EnKF algorithm has been widely applied

assimilation systems (Reichle et al., 2002b; de Wit and, 2007; Li et al., 2007; Huang et al., 2008; Jin and Li,

c equation for the EnKF algorithm can be dened accord-llowing equation:

HT(HPeHT + Re)1(D HA), (1)d A are the analysis and forecast matrices, respectively,

te variable ensembles; D represents the observation

vector; Pe avation varia(D HA) ar

In this stable that wthrough thwas an ideinvolved, Eq

Aai = Ai + (P

where Aaian

tively, for emodelled aturbed LAI o(cm3 cm3) s U(0.250.4) a U(0.020.14) n U(0.10.6)(cm d1) Ks U(10800)

for emergence (C) U(25)ure for emergence (C) U(2030)

e to anthesis (C d) U(700900) maturity (C d) U(8001200)

LAI (ha ha1 d1) U(0.010.04)(ha ha1) U(0.010.03)

35 C (d) U(3036) of development (ha kg1) U(0.0020.003)

1

of development (ha kg ) U(0.0010.002)

on rate at growth

(kg ha1 h1) U(5070)

on rate at the rst maturity

(kg ha1 h1) U(5070)

on rate at the second maturity

(kg ha1 h1) U(5070)

on rate at the third maturity

(kg ha1 h1) U(3050)

on rate at the fourth maturity

(kg ha1 h1) U(025)

ilates into storage (kg kg1) U(0.60.8)

ilates into stem (kg kg1) U(0.590.76)ilates into leaf (kg kg1) U(0.610.75)ilates into root (kg kg1) U(0.620.76)

ion rate stems (kg(CH2O) kg1 d1) U(0.0130.02)ion rate leaves (kg(CH2O) kg1 d1) U(0.0270.033)

rate per 10 C U(1.62)

ion rate storage (kg(CH2O) kg1 d1) U(0.0050.015)

ion rate roots (kg(CH2O) kg1 d1) U(0.010.016)

of leaves due to (kg kg1 d1). U(0.020.06)

rate U(0.71.2)

oting depth (cm d1) U(23)(cm) U(714)(cm) U(90.5120)

nd Re are the covariance matrices for the state and obser-bles, respectively; H is a measurement operator; and

e the innovation vectors.udy, the simulated value for LAI was the only state vari-as directly assimilated from external observations (i.e.e ETM+ dataset). Because the observation operator Hntity matrix and because no other observations were. (1) can be rewritten as the following equation:

Pe

e + Re) (Di Ai), (2)

d Ai are the analyzed and forecasted LAI states, respec-nsemble member i; Pe and Re are the covariances for thend observed LAI values, respectively; and Di is the per-bservation, which is used to update ensemble member


ion of

i (Burgers eof multiple be used. A Evensen (20

A owchthe EnKF-ahydrology-assisted assat a 30 m resowing datrecommend2007).

On the to shift thethe forecastion data fothe observamLAI2, . . ., ensemble wensemble {hydrology-time t + 1. Ifensemble wdirectly. ThThe mean oof LAI at timgrowth mo

2.8. LAI ext

A time sgrowth staThus, the creected indataset calcfeature extrmaize areas

Numerobetween sp

1987ailabcovebtai

and ng de waquirtoching oIs (T

ips w1999mospnr, d anFig. 3. Flowchart of the EnKF assimilat

t al., 1998). Note that, for the simultaneous assimilationobservations (e.g. soil moisture and LAI), Eq. (1) shoulddetailed description of this algorithm can be found in03).art for regional maize yield estimates made through

ssisted assimilation of ETM+ LAI data into the coupledcrop growth model is illustrated in Fig. 3. The EnKF-imilation was performed independently per grid unitsolution. The coupled crop model was initialized at thee. The ensemble size was set at 100 according to theations of other researchers (de Wit and Van Diepen,

rst day of simulation, white Gaussian noise was added simulated LAI and to generate the rst ensemble forted LAI {fLAI1, fLAI2, . . ., fLAIN}. If there were observa-r time t, we added a Gaussian perturbation-ensemble to

et al., the avmaps were omentssamplisatellittical rewere sremainfour SVtionshet al., and atand Taadjustetion LAI to generate an observation ensemble {mLAI1,mLAIN}. The forecasted ensemble and the observationere then assimilated to obtain the optimal estimate

LAI1, LAI2, . . ., LAIN}, which was added to the coupledcrop growth model to obtain the forecast ensemble for

an observation LAI did not exist for time t, the forecastas added to the coupled hydrology-crop growth model

is process was repeated until crop maturation occurred.f the optimal estimate ensemble was the best estimatee t, which was ultimately added to the coupled crop

del to estimate the maize yield for each grid.

raction from ETM+ images

eries of NDVI values can supply information on croptus for different crops over different growth periods.haracteristics of the maize cultivating region were

the false colour composite image created from an NDVIulated from selected ETM+ images. An object-orientedaction module (ENVI FX) in ENVI 5.0 was used to extract

from this false colour composite image.us studies have demonstrated a strong correlationectral vegetation indices (SVIs) and LAI (e.g. Peterson

3. Results

3.1. Sensiti

The EFAoutput varirst-order represents sensitivity i

Table 3Spectral vegetspectral reecrespectively thatmospheric aadopted in the

SVI

NDVI

EVI

SAVI

ARVI ETM+ LAI.

; Spanner et al., 1990a,b; Colombo et al., 2003). Withle on-site LAI measurements, LAI spatial distributionring the maize cultivation region for a selected datened based on the regression analysis of LAI measure-the spectral vegetation index (SVI) values. Regional LAIuring the crop growth period was conducted when thes scheduled to pass over the region. To meet statis-ements, 26 LAI observation samples (50% of samples)astically chosen to create the LAI spatial map, and thebservations were reserved for validation. In this study,able 3) were computed to analyze their statistical rela-ith LAI. NDVI, enhanced vegetation index (EVI) (Huete), soil adjusted vegetation index (SAVI) (Huete, 1988),herically resistant vegetation index (ARVI) (Kaufman1992) were selected as representative of intrinsic, soild atmospherically corrected indices.vity analysis

ST method calculates sensitivity indices from modelance as an indicator of importance for input factors. The(or main) sensitivity index for the ith input factor (Si)the impact on model output of this factor alone. The totalndex for the ith input factor (STi), the sum of all effects

ation indices used in LAI retrieval. blue, red, and nir represent thetance data in blue, red and infrared bands. Parameters L and regarde canopy adjustment coefcient (the SAVI term set equal to 0.5) anddjustment coefcient (the ARVI term set equal to 1). The coefcients

EVI formula are C1 = 6.0, C2 = 7.5, C3 = 1, and G = 2.5.

Formulanirrednir+redEVI = G (nirred)nir+C1redC2blue+C3SAVI = (nirred)(nir+red+L) (1 + L)ARVI = (nirrb)

(nir+rb), rb = red (blue red)


(rst and hoverall imptions with tin Fig. 4. Thrst-order hydraulic passimilatesrate. The vaeter groupsof soil hydra low total realistic irriwas therefoilation had total index.assimilationallocated amcrop. Theretant compoTable 4 presfor each parmentioned LAI had a re

Table 4Sensitivity ind

Group of par

Soil hydraulEmergence (Phenology (gLAI (group 4Green area (AssimilationConversion o

biomass (gMaintenanceDeath rates

(group 9)Correction fa

(group 10)Root parame

3.2. Parameter estimation and validation for the coupledhydrology-crop growth model

The calibration dataset collected at Yingke station (during themaize growand LAI. Tapled hydrosensitive crof the coup(ME), root-mand coefcand WSO wrespectivelyof 10 cm, 40and 0.023 cwere 13%, 6three layers5.9%, respecues were unoverestima

meae co

I ext

ausene, way 1

he bl), maop ared ble 7 d LAtwe

.7288l related L

det valu.337

spatFig. 4. Sensitivity indices of grouped parameters.

igher order) involving this input factor, represents theact on model output of this input factor through interac-he others. The estimated index pairs (Si, STi) are showne results showed that ve parameter groups had highand total indexes including the groups containing soilarameters, green areas, assimilation, the conversion of

into biomass, and the correction factor transpirationlues of the rst-order and total indexes for these param-

were more than 0.1 and 0.27, respectively. The groupaulic parameters had the highest rst-order index butindex, likely because the coupled model was run undergation conditions and the water supply for crop growthre adequate. The group of parameters for crop assim-the second highest rst-order index and the highest

This trend can be explained by the fact that the crop process determined how many carbohydrates wereong different crop parts, driving the nal yield of the

ET datadate thFig. 5.

3.3. LA

Beclate Jufrom M15 as tFig. 6(anon-crextract

TabSVIs anship be(R2 = 0nentiasimulaused toand MEand 0

Thefore, the crop assimilation process was the most impor-nent of crop development when water was not limited.ents the rst-order Si and interaction indexes (STi Si)ameter group. In addition to the ve parameter groupsabove, we also found that the group of parameters forlatively high interaction index at 0.3.

ices for each parameter group.

ameters Si STi STi Siic parameters (group 1) 0.1958 0.342 0.1462group 2) 0.0485 0.1583 0.1098roup 3) 0.0435 0.133 0.0895) 0.0432 0.3809 0.3377group 5) 0.1165 0.3696 0.2531

(group 6) 0.1674 0.6965 0.5291f assimilates intoroup 7)

0.113 0.38 0.267

respiration (group 8) 0.0541 0.2723 0.2182due to water stress 0.0212 0.1629 0.1417

ctor transpiration rate 0.1107 0.2763 0.1656

ters (group 11) 0.0569 0.2257 0.1688

relationshipdates in ourmaize grewin most pixseen in the tivated regimean LAI vawere 1.07, variation inopment or oland surfac

3.4. Numer

Becauseof the Zhanhad no obvassumed toences in control andregional scaLAI distributhis scale, ath period in 2008) included values for TAGP, WSO,ble 5 shows the estimated parameters for the cou-logy-crop growth model, including 15 of the mostop and soil hydraulic parameters. The performanceled model (Table 6) was evaluated using mean errorean-square error (RMSE), coefcient of variation (CV)

ient of determination (R2). The RMSEs for LAI, TAGP,ere 0.346 cm2 cm2, 944.8 kg ha1, and 539.8 kg ha1,. The RMSEs for soil moisture in three layers (at depths

cm and 100 cm) were 0.017 cm3 cm3, 0.02 cm3 cm3,m3 cm3, respectively. The CVs for LAI, TAGP, and WSO.9%, and 11%, respectively. The CVs for soil moisture in

(at depths of 10, 40 and 100 cm) were 6.8%, 6.9%, andtively. The ME values indicated that soil moisture val-derestimated while those for LAI, TAGP, and WSO wereted. The R2 value was higher than 0.8 for all parameters.sured by an eddy covariance system were used to vali-upled model. The results of this analysis are shown in

raction for the study region

the NDVI values for maize tend to be low in May throughe created a false colour composite image using images2 as the red band, May 28 as the green band and Julyue band. In this false colour composite image, shown inize appears blue, other crops appear cyan or yellow andeas appear black. The information of maize cultivatingy ENVI FX is shown in Fig. 6(b).presents the different statistical relationships betweenI. The results indicated that the exponential relation-

en NDVI and LAI had a larger determination coefcient) compared to other statistical relationships. This expo-tionship (Fig. 7a) was validated by comparing theAI with the observed LAI (i.e. those values that were notermine the empirical relationship) (Fig. 7b). The RMSEes for the estimated and observed LAI values were 0.51, respectively.ial LAI maps were extracted based of the exponential

between NDVI and LAI for the maize crops at selected study area. LAI maps (Fig. 8) for May demonstrated that

during the early growth period and that the LAI valueels was lower than 1.5. An apparent increase could beLAI maps (Fig. 8) for July and August over the entire cul-on, and the LAI value in most pixels exceeded 3.0. Thelues for maps for May 12, May 28, July 15, and August 161.29, 3.39, and 3.25, respectively. Temporal and spatial

LAI is generally caused by different rates of crop devel-ther factors resulting from the spatial heterogeneity in

e characteristics.

ical experiments of LAI assimilation

the same maize variety was planted throughout muchgye Oasis and because soil characteristics in this regionious changes, the calibrated parameter set could be

be invariant in this region. However, due to differ-eld management techniques (e.g. fertilization, weed

pest control), maize tends to grow unevenly at thele. As a comprehensive representation of crop status,tion was used to reect differences in crop growth atnd the EnKF algorithm integrated differences in maize


Table 5Key parameters of the coupled hydrology-crop growth model.

Parameter Unit Lower limit Upper limit Default value Calibrated value

SPAN kg ha1 17.00 50.00 33.00 39.05LAIEM ha ha1 0.03385 0.06287 0.04836 0.04413RGRLAI ha ha1 d1 0.0206 0.0382 0.0294 0.0281SLATB (DVS = 0) ha kg1 0.0018 0.0034 0.0026 0.0023SLATB1 (DVS = 0.78) ha kg1 0.0008 0.0016 0.0012 0.0009AMAXTB (DVS = 0) kg ha1 h1 49.00 91.00 70.00 51.50AMAXTB1 (DVS = 1.25) kg ha1 h1 49.00 91.00 70.00 51.19AMAXTB2 (DVS = 1.50) kg ha1 h1 44.10 81.90 63.00 49.20AMAXTB3 (DVS = 1.75) kg ha1 h1 34.30 63.70 49.00 58.90AMAXTB4 (DVS = 2.00) kg ha1 h1 14.70 27.30 21.00 21.06CVL (kg kg1) 0.6 0.76 0.68 0.656CVO (kg kg1) 0.45 0.85 0.671 0.71CVR (kg kg1) 0.65 0.76 0.69 0.67CVS (kg kg1) 0.63 0.76 0.65 0.64

CFET 0.8 1.2 1.0 1.02

r (cm3 cm3) 0.001 0.2 0.05(10 cm)0.05 (40 cm)0.05 (100 cm)

s (cm3 cm3) 0.1 1.0 0.41 (10 cm)0.41 (40 cm)0.41 (100 cm)

0.01 0.2 0.08 (10 cm)0.087 (40 cm)0.11 (100 cm)

n 0.1 1 0.13 (10 cm)0.115 (40 cm)0.10 (100 cm)

Ks (cm d1) 10 1000 569 (10 cm)344.85 (40 cm)91.62 (100 cm)

Table 6Performance of the coupled hydrology-crop growth model.

Soil moisture (cm3 cm3) LAI (cm2 cm2) TAGP (kg ha1) WSO (kg ha1)

Depth 10 cm 40 cm 100 cm

R2 0.813 0.804 0. 917 0.964 0.981 0.985RMSE 0.017 0.02 0.023 0.346 944.8 539.8ME 0.003 0.007 0.016 0.012 933.1 415.7CV 6.8% 6.9% 5.9% 13% 6.9% 11%

Fig. 5. Comparison of estimated evapotranspiration and observed evapotranspiration.


growth rateobtain the results of tand observtation of crobtained insample ploalso differeFig. 6. Extraction of the maize cultivating region fro

s into the coupled hydrology-crop growth model tooptimal LAI prole for every maize-planting grid. Thehis analysis were evaluated based on the simulateded yields from fty sample plots. A visual represen-op growth and its spatial heterogeneities were also

our LAI assimilation (Fig. 9). We found that everyt had a different growth curve for LAI (Fig. 9a), whichd from the LAI evolution curve without assimilation. In

general, theupdating stries obtaineplot are shple yield wintegrationtions providyield.m NDVI data set.

EnKF algorithm can inuence the LAI simulation by therategy. According to the different simulation trajecto-d in the sample plots, the nal yields for each sampleown in Fig. 9b. The RMSE, ME, and CV for the sam-ere 747.2, 546.17 kg ha1, and 8.7%, respectively. The

of the crop growth model and remote sensing observa-ed a strong approach for estimating the regional crop


Fig. 7. The statistical model between the observed LAI and ETM+ NDVI (a) and model evaluation (b).

Table 7Statistical relationships between SVIs and LAI.

NDVI R2 EVI R2

y = 1.05e1.41x 0.7288 y = 2.03e1.12x 0.4887y = 4.21x 0.05 0.7024 y = 3.45x + 1.86 0.5061y = 2.77ln(x) + 3.97 0.6576 y = 1.33ln(x) + 4.54 0.5075y = 3.96x2 1.44x + 1.89 0.722 y = 3.16x2 + 6.05x + 1.38 0.5107y = 4.06x0.94 0.7029 y = 4.85x0.43 0.5005

SAVI R2 ARVI R2

y = 2.21e0.31x 0.1987 y = 1.27e1.41x 0.6821y = 0.99x + 2.1 0.2174 y = 4.35x + 0.41 0.7054y = 0.48ln(x) + 3.19 0.0805 y = 2.6ln(x) + 4.41 0.6806y = 3.02x2 4.69 + 4.42 0.6059 y = 2.34x2 + 1.4x + 1.3 0.7091y = 3.12x0.15 0.0703 y = 4.64x0.85 0.6643

3.5. Regional yield estimation after assimilating extracted ETM+LAI data

The EnKF algorithm was applied to each pixel to calibrate themodel trajectory and to estimate the maize yield in the studyregion. All LAI maps were integrated into the coupled hydrology-crop growth model, and we obtained a spatially distributed yieldfor the study region at a spatial resolution similar to that of ETM+data. The spatial distribution of the regional maize yield is pre-sented in Fig. 10. In typical regions, maize yields generally do notexceed 8600.00 kg ha1. Throughout most of the Zhangye Oasis,maize yields ranged from 6500.00 to 8000.00 kg ha1, which occu-pied 89.9% of the maize planting region. A total of 17.7% of theplanted areas experienced yields that ranged from 7000.00 to7500.00 kg

Fig. 8. LAI distribution maps for maize cultivaha1, and 60.8% of the planted areas had yields fromting region.


Fig. 9. Simulated LAI on 50 sample plots after assimilation (a) and comparison of observed yield with the simulated yield (after assimilation of LAI) on 50 sample plots (b).

7500.00 to 8000.00 kg ha1. The minimum, maximum and meanyields were 6502, 8590 and 7596 kg ha1, respectively. In regionswhere the yield was below the mean value, the crops could havesuffered from water stress, plant disease or insect pests. In con-trast with the coupled hydrology-crop growth model alone, theintegrated approach was able to obtain the spatial distribution andregional variation in crop yield.

4. Discussi

The objesupply the managemenpurpose, thETM+ LAI dhydrology-cand regiona

evaluated using the yield observations collected from sample plots.The results showed that the assimilation of remote sensing datainto the coupled hydrology-crop growth model provided a rela-tively precise, region-wide spatial distribution for maize yield inthe Zhangye Oasis. This method has great potential for regionalcrop yield estimates.

In this study, only the maize yield was estimated for study. Thecolouer, wpecie

for istrib

planl regied oainedon and conclusions

ctive of this study was to develop a method that couldimpacts of irrigation, fertilizer application, and eldt on variations in maize yield at a regional scale. For this

e EnKF algorithm was developed to integrate regionalata with a high spatial resolution into a fully coupledrop growth model to obtain the spatial distributionl variation in maize yield. The assimilation results were

regiona false Howevcrop scriticalcrop dof cropationa

Basbe obtFig. 10. Spatial distribution of the maize yields af maize cultivation region was effectively extracted fromr composite image created from an ETM+ NDVI data set.hen applied to a wider scale such as a watershed, thes are miscellaneous. Accurate crop classications arethe correct determination of temporal and spatial LAIutions. Our future studies will focus on the extractionting structures, which is an essential step toward oper-onal yield predictions.n in situ LAI observations, LAI distribution maps can

from the statistical relationship between SVI and LAI.ter assimilation.


However, due to the lower temporal resolution of the ETM+ data,the temporal mismatch between the ETM+ data and ground LAIobservations will induce errors in the extraction of the LAI map. Animproved retrieval algorithm, such as the canopy radiative trans-fer model, wgrowth motemporal analso be usedsensing (De

Only LAtranspiratiomeasuremeilation algomodel and on yield forassimilation4D-VAR, cobe combine

Acknowled

This worof China, gr(grant numfor the CenSSSTC (Sinogrant numbronmental providing d

References

Anwar, M.R., Mwater dea cool-tem

Boogaard, H.L.WOFOST 7and WOFONetherlan

Boons-Prins, ECrop specCommunitCentre, JRC

Bouman, B.A.MLaar, H.H.,versity.

Burgers, G., vaKalman l

Byrne, G.F., Crapal compo10, 17518

Casanova, J.J., vegetationW07415, h

Chen, S.N., Zhaensemble Agrometeo

Chipnashi, A.Cyields in a5766.

Cohen, W.B., Gsensing. B

Colombo, R., Bin differenEnviron. 8

Curnel, Y., de of remoteliment. Agr

Dente, L., Sataderived fryield. Rem

de Wit, A., VanKalman l146, 3856

de Wit, A., Duveillerb, G., Defournyc, P., 2012. Estimating regional winter wheatyield with WOFOST through the assimilation of green area index retrieved fromMODIS observations. Agric. For. Meteorol. 164, 3952.

Dorigo, W.A., Zurita-Milla, R., de Wit, A.J.W., Brazile, J., Singh, R., Schaepman, M.E.,2007. A review on reective remote sensing and data assimilation techniques

nhanc1651, G., 1el usi0143, G., 20lemenr, J., of CEng gr246.

L., Lian indexJ. RemR.A., K

Yiel30.

M.C.,ethodrest c251S.P., Cd Cap-ote Se.L., Li,

on sys900..R., 19309..R., Julgorittp://

13.pdang,

g the ece an

i, X., 2e lay. Sci. C, Y.J.,

-MOD.N., 19. Adv. .J., Te

licatioishersang,

007. Dprosp, X., Lihang,g, W.,tal reJD01

inzelb a coin th://dx.d, Qin, Jiang, SApplicT., KoA/AVH117uangOFO

ell. 58uang

ter whFOST597a, A.

to estnce a930.

.G., Hurican . Envis, R.Bevelo77.ill also be integrated into the coupled hydrology-cropdel; and multi-source remote sensing data at a multi-d multi-spatial resolution, such as microwave data, will

to avoid the disadvantages inherent in optical remotente et al., 2008).I was assimilated for yield estimates. When evapo-n or/and soil moisture can be derived from eldnts or remotely sensing data, a multi-objective assim-rithm based on the coupled hydrology-crop growththe EnKF could be applied to simulate different impactsmation in the future. Furthermore, the use of alternative

algorithms, such as the ensemble Kalman smoother oruld be explored, and different kinds of approaches couldd to more accurately determine regional crop yields.

gements

k was supported by NSFC (National Science Foundationant number: 91125023), Gansu Sci. & Tech. Programber: 1204WCGA013), the Fundamental Research Fundstral Universities (grant number: lzujbky-2013-44) and

Swiss Science and Technology Cooperation Program,er: IP14-092010). Gratitude is expressed to the Envi-and Ecological Science Data Center in West China forata.

cKenzie, B.A., Hill, G.D., 2003. Water-use efciency and the effect ofcits on crop growth and yield of kabuli chickpea (Cicer arietinum L.) inperate subhumid climate. J. Agric. Sci. 141, 285301., Van Diepen, C.A., Rtter, R.P., Cabrera, J.C.M.A., Van Laar, H.H., 1998..1: Users Guide for the WOFOST 7.1 Crop Growth Simulation ModelST Control Center 1.5, Techn. Doc. 52. Alterra, WUR, Wageningen, Theds, pp. 144..R., de Koning, G.H.J., Van Diepen, C.A., Penning de Vries, F.W.T., 1993.ic simulation parameters for yield forecasting across the Europeany, Simulation Reports CABO-TT 32, CABO-DLO, DLO Winand Staring, Wageningen.., Kropff, M.J., Tuong, T.P., Wopereis, M.C.S., Ten Berge, H.F.M., Van

2001. ORYZA2000: Modeling Lowland Rice. IRRI/Wageningen Uni-

n Leeuwen, P.J., Evensen, G., 1998. Analysis scheme in the ensembleter. Mon. Wea. Rev. 126, 17191724.pper, P.F., Mayo, K.K., 1980. Monitoring land-cover change by princi-nent analysis of multi-temporal Landsat data. Remote Sens. Environ.4.

Judge, J., 2008. Estimation of energy and moisture uxes for dynamic using coupled SVAT and crop-growth models. Water Resour. Res. 44,ttp://dx.doi.org/10.1029/2007WR006503.o, Y.X., Shen, S.H., 2012. Applicability of PyWOFOST model based onKalman lter in simulating maize yield in Northeast China. Chin. J.r. 33 (2), 245253.., Ripley, E.A., Lawford, R.G., 1999. Large-scale simulation of wheat

semi-arid environment using crop-growth model. Agric. Syst. 59,

oward, S.N., 2004. Landsats role in ecological applications of remoteioscience 54, 535545.ellingeri, D., Fasolini, D., Marino, C.M., 2003. Retrieval of leaf area indext vegetation types using high resolution satellite data. Remote Sens.6 (1), 120131.Wit, A.J.W., Duveiller, G., Defourny, P., 2011. Potential performancesy-sensed LAI assimilation in WOFOST model based on an OSS exper-ic. For. Meteorol. 151, 18431855.lino, G., Mattia, F., Rinaldi, M., 2008. Assimilation of leaf area indexom ASAR and MERIS data into CERES-Wheat model to map wheatote Sens. Environ. 112 (4), 13951407.

Diepen, K., 2007. Crop model data assimilation with the ensembleter for improving regional crop yield forecasts. Agric. For. Meteorol..

for e(2),

Evensenmod99, 1

Evensenimp

Eitzingeson duri223

Fang, H.tionInt.

Feddes, Croppp. 9

Hansen,A mof fo2495

Healey, seleRem

Huang, Cilati888

Huete, A295

Huete, A13) Aat hmod

Jin, H., Wusinscien

Jin, R., Lactivdata

KaufmanEOS

Kogan, Flites

Kropff, MAppPubl

Li, X., HuC., 2and

Li, X., LiZ., ZWanimen2008

Li, Y., Kwithsis, http

Liang, S.In: Land

Liu, W.NOA1161

Ma, G., Hthe WMod

Ma, H., HwinWO58, 7

Maruyamels bala919

Masek, JAmeSens

Matthewin Dpp. 2ed agroecosystem modeling. Int. J. Appl. Earth Observ. Geoinform. 993.994. Sequential data assimilation with a nonlinear quasi-geostrophicng Monte Carlo methods to forecast error statistics. J. Geophys. Res.10162.03. The ensemble Kalman lter: theoretical formulation and practicaltation. Ocean Dyn. 53 (4), 343367.Trnka, M., Hsch, J., Zalud, Z., Dubrovsky , M., 2004. Compari-RES, WOFOST and SWAP models in simulating soil water contentowing season under different soil conditions. Ecol. Modell. 171,

g, S.L., Hoogenboom, G., 2011. Integration of MODIS LAI and vegeta- products with the CSM-CERES-Maize model for corn yield estimation.ote Sens. 32 (4), 10391065.owalik, P.J., Zaradny, H., 1978. Simulation of Field Water Use and

d, Simulation Monograph. Pudoc, Wageningen, The Netherlands,

Roy, D.P., Lindquist, E., Adusei, B., Justice, C.O., Altstatt, A., 2008. for integrating MODIS and Landsat data for systematic monitoringover and change in the Congo Basin. Remote Sens. Environ. 112,3.ohen, W.B., Yang, Z.Q., Krankina, O.N., 2005. Comparison of Tas-based Landsat data structures for use in forest disturbance detection.ns. Environ. 97, 301310.

X., Lu, L., 2008. Experiments of one-dimensional soil moisture assim-tem based on ensemble Kalman lter. Remote Sens. Environ. 112,

88. A soil adjusted vegetation index (SAVI). Remote Sens. Environ. 25,

stice, C.O., van Leeuwen, W., 1999. MODIS Vegetation Index (MODthm Theoretical Basis Document (ATBD-MOD-13) version 3, availableeospso.gsfc.nasa.gov/ftp ATBD/REVIEW/MODIS/ATBD MOD 13/atbdfJ., Xiao, Z., Fu, Z., 2010. Leaf area index estimation from MODIS datansemble Kalman smoother method. In: 2010 IEEE International Geo-d Remote Sensing Symposium, pp. 260263.009. Improve the estimation of hydrothermal state variables in theer of frozen ground by assimilating in situ observations and SSM/Ihina Ser. D: Earth Sci. 52 (11), 17321745.Tanr, D., 1992. Atmospherically resistant vegetation index (ARVI) forIS. IEEE Trans. Geosci. Remote Sens. 30, 261270.98. Global drought and ood-watch from NOAA polar-orbitting satel-Space Res. 21 (3), 477480.ng, P.S., Aggarwal, P.K., Bouman, B., Bouma, J., Van Laar, H.H., 1996.ns of Systems Approaches at the Field Level, vol. 2. Kluwer Academic, The Netherlands, pp. 465.C., Che, T., Jin, R., Wang, S., Wang, J., Gao, F., Zhang, S., Qiu, C., Wang,evelopment of a Chinese land data assimilation system: its progressects. Prog. Nat. Sci. 17 (8), 881892., Z., Ma, M., Wang, J., Xiao, Q., Liu, Q., Che, T., Chen, E., Yan, G., Hu,

L., Chu, R., Su, P., Liu, Q., Liu, S., Wang, J., Niu, Z., Chen, Y., Jin, R., Ran, Y., Xin, X., Ren, H., 2009. Heihe watershed allied telemetry exper-search. J. Geophys. Res. 114 (D22103), http://dx.doi.org/10.1029/

1590ach, W., Zhou, J., Cheng, G., Li, X., 2012. Modelling irrigated maizembination of coupled-model simulation and uncertainty analy-e northwest of China. Hydrol. Earth Syst. Sci. 16, 14651480,oi.org/10.5194/hess-16-1465-2012.

., 2008. Data assimilation methods for land surface variable estimation.. (Ed.), Advances in Land Remote Sensing: System, Modeling, Inversionation. Springer, New York, pp. 319339.gan, F., 2002. Monitoring Brazilian soybean production usingRR based vegetation condition indices. Int. J. Remote Sens. 23 (6),

9., J., Wu, W., Fan, J., Zou, J., Wu, S., 2012. Assimilation of MODIS-LAI intoST model for forecasting regional winter wheat yield. Math. Comput., 634643., J., Zhu, D., Liu, J., Su, W., Zheng, C., Fan, J., 2013. Estimating regionaleat yield by assimilation of time series of HJ-1 CCD NDVI into

ACRM model with Ensemble Kalman Filter. Math. Comput. Modell.70., Kuwagata, T., 2010. Coupling land surface and crop growth mod-imate the effects of changes in the growing season on energynd water use of rice paddies. Agric. For. Meteorol. 150 (7),

ang, C.Q., Wolfe, R., Cohen, W., Hall, F., Kutler, J., Nelson, P., 2008. Northforest disturbance mapped from a decadal Landsat record. Remoteron. 112, 29142926.., Stephens, W., 2002. Crop-soil Simulation Models, Applicationsping Countries. CABI, Publishing, Wallingford, United Kingdom,


McKay, M.D., 1995. Evaluating Prediction Uncertainty. NUREG/CR-6311. US NuclearRegulatory Commission and Los Alamos National Laboratory, Los Alamos, NM.

Mignolet, C., Schott, C., Benot, M., 2007. Spatial dynamics of farming practices in theSeine basin: methods for agronomic approaches on a regional scale. Sci. TotalEnviron. 375, 1332.

Monteith, J.L.,1965. Evaporation and environment. In: Proceedings of the 19th Sym-posium of the Society for Experimental Biology. Cambridge University Press,New York, pp. 205233.

Monteith, J.L., 1981. Evaporation and surface temperature. Quart. J. Royal Soc. 107,127.

Moulin, S., Guerif, M., 1999. Impacts of model parameter uncertainties on cropreectance estimates: a regional case study on wheat. Int. J. Remote Sens. 20(1), 213218.

Munoz-Carpena, R., Fox, G.A., Sabbagh, G.J., 2010. Parameter importance and uncer-tainty in predicting runoff pesticide reduction with lter strips. J. Environ. Qual.39 (1), 112.

Patil, S.L., Sheelavantar, M.N., 2004. Effect of cultural practices on soil properties,moisture conservation and grain yield of winter sorghum in semi-arid tropicsof India. Hydrol. Process. 64, 4967.

Peterson, D.L., Spanner, M.A., Running, S.W., Teuber, K.B., 1987. Relationship of the-matic mapper simulator data to leaf area index of temperate coniferous forests.Remote Sens. Environ. 22 (3), 323341.

Priesack, E., Gayler, S., Hartmann, H.P., 2006. The impact of crop growth sub-modelchoice on simulated water and nitrogen balances. In: Modelling Water andNutrient Dynamics in Soilcrop Systems. Springer, Germany, pp. 183195.

Quaife, T., Lewis, P., Kauwe, M.D., Williams, M., Law, B.E., Disney, M., Bowyer, P.,2008. Assimilating canopy reectance data into an ecosystem model with anensemble Kalman lter. Remote Sens. Environ. 112, 13471364.

Reichle, R.H., McLaughlin, D.B., Entekhabi, D., 2002a. Hydrologic data assimilationwith the ensemble Kalman lter. Mon. Weather Rev. 130 (1), 103114.

Reichle, R.H., Walker, J.P., Koster, R.D., Houser, P.R., 2002b. Extended versusensemble Kalman ltering for land data assimilation. J. Hydrometeorol. 3 (6),728740.

Revill, A., Sus, O., Barrett, B., Williams, M., 2013. Carbon cycling of Europeancroplands: a framework for the assimilation of optical and microwave Earthobservation data. Remote Sens. Environ. 137, 8493.

Rodriguez-Iturbe, I., Porporato, A., Laio, F., Ridol, L., 2001. Intensive or extensiveuse of soil moisture: plant strategies to cope with stochastic water availability.Geophys. Res. Lett. 28, 44954497.

Rouse, J.W., Haas, R.H.,1973. Monitoring Vegetation Systems in the Great Plain withERTS. In: Third ERTS Symposium. NASA, Washington, DC, pp. 309317.

RSI, 2001. ENVI Users Guide. September 2001 edition. Research Systems.Saltelli, A., Chan, K., Scott, E.M., 2000. Sensitivity Analysis. Wiley Series in Probability

and Statistics. Wiley, Chichester, New York.

Shepherd, A., McGinn, S.M., Wyseure, G.C.L., 2002. Simulation of the effect of watershortage on the yields of winter wheat in north-east England. Ecol. Modell. 147,4152.

Simunek, J., Van Genuchten, M.Th., Sejna, M., 2005. The HYDRUS-1D SoftwarePackage for Simulating the Movement of Water, Heat, and Multiple Solutes inVariability Saturated Media, Version 3. 0. Department of Environmental SciencesUniversity of California Riverside, Riverside, California, USA, pp. 270.

Spanner, M.A., Pierce, L.L., Peterson, D.J., Running, S.W., 1990a. Remote sensing oftemperate coniferous forest leaf area index, the inuence of canopy closure,understory vegetation and background reectance. Int. J. Remote Sens. 11 (l),95111.

Spanner, M.A., Pierce, L.L., Running, S.W., Peterson, D.L., 1990b. The seasonality ofAVHRR data of temperate coniferous: relationship with leaf area index. RemoteSens. Environ. 33, 97112.

Stol, W., Rouse, D.I., Kraalingen, D.W.G., Klepper, O., 1992. FSEOPT a FORTRAN Pro-gram for Calibration and Uncertainty Analysis of Simulation Models. SimulationReport CABO-TT, No. 24. CABO-DLO and Agricultural University, Wageningen,pp. 24.

ten Berge, H.F.M., Aggarwal, P.K., Kropff, M.J. (Eds.), 1997. Applications of Rice Mod-elling. Elsevier, The Netherlands, p. 166.

Tsuji, G.Y., Hoogenboom, G., Thornton, P.K., 1998. Understanding Options forAgricultural Production. Kluwer Academic Publishers, The Netherlands, pp.399409.

Van den Hoof, C., Hanert, E., Vidale, P.L., 2011. Simulating dynamic crop growthwith an adapted land surface modelJULES-SUCROS: model development andvalidation. Agric. For. Meteorol. 151 (2), 137153.

Van Laar, H.H., Goudriaan, J., Van Keulen, H., 1997. SUCROS 97: simulation of cropgrowth for potential and water-limited production situations, as applied tospring wheat. Quantitative Approaches in Systems Analysis, vol. 14. AB-DLO,Wageningen, The Netherlands.

Vazifedoust, M., van Dam, J.C., Bastiaanssen, W., Feddes, R.A., 2009. Assimilation ofsatellite data into agrohydrological models to improve crop yield forecasts. Int.J. Remote Sens. 30, 25232545.

Vogelmann, J.E., Howard, S.M., Yang, L.M., Larson, C.R., Wylie, B.K., Van Driel, N.,2001. Completion of the 1990 National Land Cover Data set for the conterminousUnited States from Landsat Thematic Mapper data and Ancillary data sources.Photogramm. Eng. Remote Sens. 67, 650662.

Wolf, J., 2002. Comparison of two potato simulation models under climatic change.I. Model calibration and sensitivity analysis. Climate Res. 21, 173186.

Xu, W., Jiang, H., Huang, J., 2011. Regional crop yield assessment by combination of acrop growth model and phenology information derived from MODIS. Sens. Lett.9, 981989.

Xu, Z., Liu, S., Gong, L., Wang, J., Li, X., 2008. A study on the data processing andquality assessment of the eddy covariance system. Adv. Earth Sci. 23, 357370.

gis crop growth maize zhangye

Documents