Simulating regional winter wheat yields using input data of different spatial resolution

(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Field Crops Research 145 (2013) 67–77

Contents lists available at SciVerse ScienceDirect

Field Crops Research

jou rn al h om epage: www.elsev ier .com/ locate / fc r

Simulating regional winter wheat yields using input data of different spatialresolution

C. Nendela,∗, R. Wielanda, W. Mirschela, X. Speckaa, C. Guddatb, K.C. Kersebauma

a Leibniz Centre for Agricultural Landscape Research, Institute of Landscape Systems Analysis, Eberswalder Straße 84, 15374 Müncheberg, Germanyb Thuringia State Office for Agriculture, Department of Plant Production and Agro-Ecology, Apoldaer Straße 4, 07774 Dornburg-Camburg, Germany

a r t i c l e i n f o

Article history:Received 27 August 2012Received in revised form 9 November 2012Accepted 22 February 2013

Keywords:MONICAAgro-ecosystem modelDynamic modellingScalingInput data

a b s t r a c t

The success of using agro-ecosystem models for the high-resolution simulation of agricultural yieldsfor larger areas is often hampered by a lack of input data. We investigated the effect of different spa-tially resolved soil and weather data used as input for the MONICA model on its ability to reproducewinter wheat yields in the Federal State of Thuringia, Germany (16,172 km2). The combination of onerepresentative soil and one weather station was insufficient to reproduce the observed mean yield of6.66 ± 0.87 t ha−1 for the federal state. Use of a 100 m × 100 m grid of soil and relief information com-bined with just one representative weather station yielded a good estimator (7.01 ± 1.47 t ha−1). The soiland relief data grid used in combination with weather information from 14 weather stations in a nearestneighbour approach produced even better results (6.60 ± 1.37 t ha−1); the same grid used with 39 addi-tional rain gauges and an interpolation algorithm that included an altitude correction of temperature dataslightly overpredicted the observed mean (7.36 ± 1.17 t ha−1). It was concluded that the apparent successof the first two high-resolution approaches over the latter was based on two effects that cancelled eachother out: the calibration of MONICA to match high-yield experimental data and the growth-defining and-limiting effect of weather data that is not representative for large parts of the region. At the county andfarm level the MONICA model failed to reproduce the 1992–2010 time series of yields, which is partlyexplained by the fact that many growth-reducing factors were not considered in the model.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

The simulation of agricultural yields has gained in importancerecently due to growing concern about food security for a worldpopulation that hit the 7 billion mark and is set to continue increas-ing (Rötter et al., 2011), with a wide range of potential maximahaving been projected based on different scenarios (UN, 2011). Inaddition, agriculture is highly vulnerable to changes in climate,as climate factors are the main drivers of open-field agriculturalproduction. The potential impact of climate change on agricultureis explored in depth in the recent debate on the various chal-lenges agriculture will have to face in a changing climate (Cassman,2007; Foley et al., 2011; Lobell et al., 2008; Porter et al., 2010;Schmidhuber and Tubiello, 2007; Vermeulen et al., 2012). The anal-ysis of yield gaps and the search for ways to ecologically intensifythe world’s staple food production systems are key areas of currentenvironmental research (Bennett et al., 2012; Lobell et al., 2009;Neumann et al., 2010). Also the increasing competition betweenbioenergy and food production for land and resources demands for

∗ Corresponding author. Tel.: +49 33432 82355; fax: +49 33432 82334.E-mail address: [email protected] (C. Nendel).

model-based yield predictions for energy cropping scenario anal-ysis (Das et al., 2012; Miguez et al., 2012; VanLoocke et al., 2012).Crop modelling as a tool to predict crop yields under assumedclimate and land use scenarios is widely accepted (White et al.,2011); the predictive power of models is currently under investi-gation, as their prediction error adds to the range of uncertaintyproduced by global scale climate models and the down-scalingprocedures required to drive small-scale crop simulation models(Déqué et al., 2007; Olesen et al., 2007; Zhang et al., 2011). The Agri-cultural Model Intercomparison and Improvement Project (AgMIP,Rosenzweig et al., 2013) and the European Commission’s knowl-edge hub MACSUR (http://www.macsur.eu) are the most recentinitiatives involving a wide range of scientists across the globetasked with quantifying the uncertainty of crop yield projectionsobtained from simulations, and find ways of reducing them. Ourstudy adds improved understanding on how the use of climatevariables for driving crop models may influence the simulationaccuracy. In this investigation real weather data is used; however,the findings can also be transferred to the use of climate scenarios.

Crop models run on a single field or crop canopy scale. This is thescale used to test how a model performs against observed data andto prove the reliability of the model’s predictions for crop or soilprocesses (De Willigen, 1991; Diekkrüger et al., 1995; Kersebaum

0378-4290/$ – see front matter © 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.fcr.2013.02.014

Author's personal copy

68 C. Nendel et al. / Field Crops Research 145 (2013) 67–77

et al., 2007; Palosuo et al., 2011; Rötter et al., 2012). This process isoften referred to as “validation”, although the modeller communityreached consensus long ago that “validity” cannot be expected fromthe result of a simulation model (Oreskes et al., 1994; Rykiel, 1996).However, this is the reason why great emphasis must be placed ondetermining the potential error produced by the model when usedto predict processes in a previously unknown – and thus unpa-rameterised – environment. Use of crop models on a large scale– up to the global scale – requires a number of assumptions thatdo not apply to the scale of model validation (Faivre et al., 2004).The restriction of the input data required to drive the model and– until recently – limited computer power often led to modellingapproaches that explained crop growth on planet Earth in a verysimplified manner. Simulating crop growth on such large areas wasmainly performed in three different ways, whereby the aim of thesimulation exercise determined which approach was consideredmost appropriate:

(1) Simple crop growth models were created and crop factors orstatistical models used to transform the biomass results intodata corresponding to typical observations for different cropsin certain areas of the world. In some cases, groups of crops withsimilar characteristics (plant functional types) were considered(e.g. Bondeau et al., 2007; Deryng et al., 2011; Lobell et al., 2008).

(2) State-of-the-art crop models were used to simulate crops atcertain sites considered representative for a larger area (catch-ment, continent, agro-ecological zone; e.g. Alexandrov et al.,2002; Elsgaard et al., 2012; Parry et al., 2004; Wolf and vanDiepen, 1995).

(3) State-of-the-art crop models were used to simulate crops on aspatial grid in all areas where these crops typically grow (e.g.Liu et al., 2007; Moriondo et al., 2010; Stehfest et al., 2007; Supitet al., 2012).

Now that computer power is able to produce high-resolutionsimulations of large areas using sophisticated process models, itis now possible to investigate the results generated by differentapproaches for simulating regional crop yields (Andersson et al.,2012; Folberth et al., 2012). Previous works already addressed scal-ing issues in the context of crop model application (Gimona et al.,2006; Hansen and Jones, 2000; Moen et al., 1994; Therond et al.,2011; Xiong et al., 2008), for which Ewert et al. (2011) provide atheoretical framework. A systematic investigation of the effects ofinput data aggregation on simulation results was presented by vanBussel et al. (2011) for winter wheat phenology. Several studiesdemonstrated that simulations for regional crop yields improved –within limits – with using a finer resolution of climate data (de Witet al., 2005; Easterling et al., 1998; Olesen et al., 2000; Rivingtonet al., 2006). Soil data aggregation does not seem to affect sim-ulation results much under sufficient water supply of the crop(Easterling et al., 1998; Folberth et al., 2012). However, it maywell do so when water is limiting (Baron et al., 2005; Wassenaaret al., 1999). Crop yield simulations may also be strongly affected byaggregating management information, as it was shown for irrigatedand rain-fed maize production (Folberth et al., 2012).

We chose a federal state of Germany – the Free State of Thuringiafor which a set of detailed historical yield records exists – to testhow the spatial resolution of input data would affect the result ofa spatial simulation. Our study aims to answer two main researchquestions: the first was to determine whether it is necessary to sim-ulate a region using the highest possible level of detail, includingsmall-scale soil, relief and crop distribution information, in order toproduce a regionally representative yield figure for winter wheat(Triticum aestivum L.). For the opposite case of highest possible levelof generalisation, we would end up with one carefully selected siteand one equally selected crop parameter set that would give an

estimate of the annual winter wheat yield in the region goodenough to be representative for the federal state, or at least twoor three sub-regions. The second aim was to find out whether thecrop model was able to capture the spatial and temporal variabilityof winter wheat yields at a lower scale. To do this the model resultswere evaluated also at the county level and the model was addition-ally tested against field data obtained from experimental stations.The subsequent discussion focuses on which effects of changingfactors continue across scales and which become lost in the scalingprocess.

2. Materials and methods

2.1. Simulation design

We simulated winter wheat as a single crop using the Modelfor Nitrogen and Carbon dynamics in Agro-ecosystems MONICA(Nendel et al., 2011) for the 1992–2010 period. Four different setsof input data were tested, each of which had a different informa-tion density. First, we used historical weather data from one singleweather station (Erfurt-Bindersleben, 50◦58′49′′ N, 10◦58′12′′ E,307 m), which was assumed to represent typical weather for mostof the agricultural area of Thuringia. Additionally, we used the mostabundant soil in Thuringia – a Vertic Cambisol with >60% clay – andconducted one simulation using this soil–weather station combina-tion. This simulation, referred to as the single point simulation, wasonly analysed at state level, not at county level or below. In a furtherstep, this simulation was repeated using the second most abundantsoil – a loamy sand (Dystric Cambisol) – in order to demonstratethe effect caused by the choice of soil. We also repeated this simu-lation for all combinations of the five most abundant soils with thesix Thuringian weather stations Artern, Erfurt-Bindersleben, Gera,Meinigen, Leinefelde and Schmücke (Fig. 1) to quantify the rangeof results that would arise from the choice of a non-representativesoil or weather station.

Second, we used the Erfurt-Bindersleben weather station andsimulated a 100 m × 100 m grid covering the whole 16,172 km2 ofThuringia. MONICA was executed in a parallel application on a 48core computer; more than 1.6 × 106 data points were calculateddaily. Soil data was taken from the soil map of Germany (BÜK1000, Hartwich et al., 1995b) in the scale of 1:1,000,000, whichalso provided information on average within-profile groundwaterlevels (Hartwich et al., 1995a). Altitude and slope information wasderived from a digital elevation model with a resolution of 100 m.This simulation is referred to as the single station simulation.

Third, we used the same 100 m × 100 m grid with soil, ground-water, altitude, and slope information, this time using 14 weatherstations within and outside of Thuringia. Each grid cell was assignedone nearest weather station using Thiessen polygons (Thiessen,1911). This simulation is referred to as the nearest neighbour simu-lation.

Finally, we used the 100 m × 100 m grid with soil, groundwa-ter, altitude, and slope information and 14 weather stations withinand outside of Thuringia, but added 39 rain gauges from Thuringia.For each grid cell, weather information was interpolated from theweather station data and corrected according to the grid cell’saverage altitude. This simulation is referred to as the interpolationsimulation.

For all simulations carried out on the 100 m × 100 m grid, sandysoils with <25% silt and <8% clay were excluded because winterwheat would not be produced on such soils in Thuringia. Beforeanalysing the simulated yields, the grid was filtered for agriculturalland use using CORINE 2000 land cover data (Keil et al., 2005). Yieldstatistics for each of the 17 counties were provided by the State Gov-ernment of Thuringia for the 1992–2010 period. The yield statistics

Author's personal copy

C. Nendel et al. / Field Crops Research 145 (2013) 67–77 69

Fig. 1. Location of the six agricultural experimental stations (©), 14 weather stations (�) and 39 rain gauges (�) in Thuringia, Germany.

for another six counties which mainly represent urban areas wereneglected, even though there is some agricultural production there.In the framework of scaling methods classified by Ewert et al. (2011)the latter three approaches represent data manipulation methodswhich average the outputs of multiple model runs.

2.2. The agro-ecosystem simulation model MONICA

MONICA is a dynamic, process-based agro-ecosystem simula-tion model developed from the HERMES model (Kersebaum, 1995,2007) for simulating crop growth and soil processes in CentralEurope (Nendel et al., 2011). It simulates the growth and develop-ment of annual crops and perennial grassland as well as the relatedwater, nitrogen and carbon dynamics in the soil–plant–atmospheresystem. A capacity approach describes water transport in the soilaccording to Wegehenkel (2000). Reference evapotranspiration iscalculated using the Penman-Monteith method, according to Allenet al. (1998), and crop-specific potential evapotranspiration is com-puted using crop-specific factors (Kc) during the growing season.Organic matter turn-over is calculated using algorithms from theDAISY model (Hansen et al., 1991). Crop growth follows the genericapproach first presented by the SUCROS model (van Keulen et al.,1982). [CO2] affects the crop’s maximum photosynthesis rate andstomatal resistance, which in turn influences transpiration (Nendelet al., 2009). The impact of extreme heat on crop growth and yieldformation is considered based on ideas put forward by Challinoret al. (2005) and Moriondo et al. (2010). Maintenance respirationis calculated separately for day and night periods using AGROSIMalgorithms (Mirschel and Wenkel, 2007). Root dry matter is dis-tributed over depth according to Pedersen et al. (2010), wherebythe rooting depth increases linearly with the thermal sum. Waterand N stress reduce crop growth and accelerate crop ontogenesis atspecific development stages. MONICA was calibrated to predict thegrowth of major Central European crops, also under elevated [CO2],and tested at length for its ability to predict yields at the field scalein uncalibrated situations in Germany (Nendel et al., 2011), acrossEurope (Rötter et al., 2012; Salo et al., in preparation) and globallyfor major food crops and agricultural regions (Rosenzweig et al.,2013). So far, MONICA has not been used to predict crop yields forlarger areas. Its original design was developed to achieve optimum

results at the field scale. Accordingly, its parameters were only cal-ibrated to experimental data at the field level. MONICA operatedwith one sowing date and with one genotype parameterisation forthe whole region. N fertiliser applications were modelled accord-ing to the Nmin method (Wehrmann and Scharpf, 1979) whichis implemented in MONICA. This approach takes the 0–0.9 m soilmineral N stock at 30 March into account for calculation of the best-practice N application for winter wheat and thus provides sufficientN supply to the crop in most, but not all seasons (Nendel, 2009).Irrigation was not applied.

2.3. MONICA test against field experiment data

In many simulation exercises, the simulation model is first cal-ibrated against experimental data from the region of interest – ifavailable – and then, in a second step, used to predict yields or othervariables in space or time. We postulate that MONICA has been suf-ficiently calibrated against field data from Germany (Nendel et al.,2011) and that no further calibration is required to improve themodel. However, we use the data available from different experi-mental stations across Thuringia to challenge the model again onthe scale it was initially designed to be applied to. Winter wheatdata was provided in the form of mean yields across all varietiesgrown in field experiments. This is comparable to county statisticsin which all varieties grown in the county are integrated. How-ever, contrary to county data, station data provides yields obtainedunder optimum conditions, which can easily be secured on suffi-ciently small experimental plots, but rarely under the conditionsof large-scale commercial agriculture. Experimental station datawere provided for the 1992–2010 period, with a few missing years.The experimental stations sometimes grow their variety trials ondifferent fields that exhibit different soil characteristics. For thisreason the soil information used for simulations of the experimen-tal stations may not be representative for parts of the trials in somecases (Table 1).

2.4. Input data for the simulation exercise

Soil information was taken from the 1:106 soil map of Germany(BÜK 1000, Hartwich et al., 1995a). The spatial resolution for

Author's personal copy

70 C. Nendel et al. / Field Crops Research 145 (2013) 67–77

Table 1Experimental stations in Thuringia.

Station County Altitude (m) Mean precipitation (mm a−1) Mean temperature (◦C) Substrate Texture

Burkersdorf Saale-Orla-Kreis 440 642 7.1 Withered rock Sandy loamDornburg Sömmerda 260 584 8.3 Loess Clayey siltFriemar Gotha 284 541 8.0 Loess LoamGroßenstein Altenburger Land 300 606 8.0 Loess LoamHeßberg Hildburghausen 380 773 7.4 Withered rock Clay loamKirchengel Kyffhäuser-Kreis 305 556 7.6 Loess Loam

the simulations was chosen as being the coarsest possible reso-lution that still enabled the boundaries of soil classes from thesoil map to be drawn smoothly. According to this map, eachgrid cell was assigned a soil type for which basic information isavailable in terms of a representative soil profile. MONICA useshorizon boundaries, soil texture, soil organic matter contents andbulk density. Soil types with signs of groundwater influence wereused as indicators for shallow (within profile) groundwater lev-els, where the boundary between the oxidised and reduced soilhorizon marked the mean groundwater level. MONICA uses themean groundwater level for soil water content and transport cal-culations. For Thuringia, seasonal groundwater level fluctuationswere assumed to be of minor importance. The vast majority ofgroundwater-influenced soils in Thuringia are alluvial soils andhave their groundwater level regulated by close-by rivers, whichdo not show large seasonal fluctuations in water level.

The digital elevation model for Thuringia provides informationon the mean altitude for each grid cell. An average slope within thegrid cell was calculated from this information. MONICA uses theslope information to calculate precipitation surface run-off.

Daily weather data for 1992–2010 was provided by the Ger-man Weather Service (DWD) for 14 weather stations within andoutside of Thuringia. Weather information includes minimum andmaximum air temperatures (2 m), precipitation, global radiation,mean wind speed (2 m) and air humidity on a daily basis. Addi-tionally, precipitation data from 39 DWD rain gauges was used. Forthe interpolated simulation, all weather data except precipitationwas interpolated and corrected for altitude according to Watsonand Philip (1985). The spatial interpolation algorithm is based on acombination of a regression function and an inverse distance algo-rithm. The regression estimates the response of a climate element(temperature, precipitation, etc.) X(x, y) at the location (x, y) to thealtitude h(x, y).

X(x, y) = n + m · h(x, y) (1)

The residual res(xi,yi)(x, y) = X(xi, yi) − X(x, y) of the measured X(xi,yi) at the location of each climate station (xi, yi) is used to correctthe regression using the inverse distance method

X(x, y) = n + m · h +∑

(xi,yi)(res(xi,yi)(x, y))/d(xi,yi)(x, y)∑

(xi,yi)1/(d(xi,yi)(x, y))

(2)

with the Euclidean distance d(xi,yi)(x, y) between the station i andthe location (x, y). The regression coefficients n, m were calculatedon daily basis to take into account different weather conditions.The rain gauge data was used for the interpolation of precipitationin addition to the climate station data.

2.5. Data treatment and performance indices

The yield data obtained from official statistics show a shallowbut significant positive trend for each of the 17 counties. However,Gutzler et al. (in preparation) concluded from their analysis of yieldtrends in Germany that at least the yield trend in the Eastern statesof Germany in the 1990s was mainly driven by technology improve-ment, whereas no further positive trend can be observed after

1998. A technology trend superposing yield data cannot easily beaddressed with a biophysical process model such as MONICA. Sinceit is impossible to clearly mark the end of the technology trend, wedecided to use the data untreated. In contrast, yield data from theexperimental stations also show a significant positive trend for theperiod after 1999, showing the continuous improvement of winterwheat varieties towards higher potential yields. We did not attemptto address a temporal shift in varieties using the MONICA model.For this reason, we decided to detrend the data from the exper-imental stations with a linear model, fitted to each data set. Wepostulate that the trend elimination improves the comparabilitybetween the data and the simulation. For five out of six stations, thelinear trend of the linear model ranged between 0.54 (Kirchengel)and 2.02 (Dornburg). For Burkersdorf the trend was −0.09, probablydue to a number of missing values. This trend was also eliminatedfor consistency reasons.

The mean winter wheat yield at the stations is also consider-ably higher than official county statistics. This is due to the factthat great pains are taken to ensure that trials at the station arealways maintained in optimum conditions for crop growth. Fur-thermore, many of the high-performance varieties grown in trialsare not yet widely accepted by farmers, which may be due to theeffect of the varieties’ performance on many other characteristics,such as pest resistance and lodging tendency. In order to reducethe model performance evaluation to the year-to-year variabilityof the yield data, data from wheat variety trials were also cor-rected by the mean bias error MBE of the model prediction. Themean bias error (Addiscott and Whitmore, 1987) summarises theaverage error of model predictions and, as such, identifies any over-predictions or underpredictions generated by the model. The meanabsolute error (MAE, Shaeffer, 1980) was also used as an addi-tional performance measure. MAE measures the average magnitudeof prediction errors, however, without indicating the direction ofdeviation. MAE can be normalised by dividing the figure by themean of the observations, yielding the normalised mean absoluteerror (nMAE). Finally, modelling efficiency (ME, Nash and Sutcliffe,1970) was used. ME is another common index based on the corre-lation between observed and predicted values. It ranges from −∞to 1 and finds its optimum when ME equals 1. A value of zero indi-cates that the model is no better estimator than the observed mean.Here, ME was calculated for the simulation against the detrendedyield data (MEd) and against the detrended and mean-correctedyield data (MEdc).

3. Results

The mean observed winter wheat yield for Thuringia duringthe 1992–2010 period was 6.66 t ha−1, with a standard error of0.87 t ha−1. This observation is represented by a black line and agrey bar in Fig. 2. Using only one representative weather station andthe most abundant soil (single point simulation, boxplot A), MONICAcalculated 5.47 ± 2.37 t ha−1, underestimating the observed meanyield significantly by 1.19 t ha−1. When the second most abundantsoil was used instead (single point simulation, boxplot B), MON-ICA predicted 7.98 ± 1.18 t ha−1, significantly overestimating theobserved mean yield by 1.32 t ha−1 (Fig. 2). Other combinations of

Author's personal copy

C. Nendel et al. / Field Crops Research 145 (2013) 67–77 71

Fig. 2. Box plot of 1992–2010 winter wheat yields of the Federal State of Thuringia,Germany, as simulated using the MONICA model for a point simulation with weatherdata from a single station and the most abundant soil (A, n = 19 years), a single stationand the second most abundant soil (B, n = 19 years), all combinations of six weatherstations with the five most abundant soils (C, n = 570 = 19 years × 5 soils × 6 weatherstations) and for a high-resolution simulation across 17 counties with weatherdata from a single station (n = 323 = 19 years × 17 county means), with weather datafrom the nearest weather station (nearest neighbour, n = 323) and with interpolatedand altitude-corrected data from weather stations and rain gauges (interpolation,n = 323). The black line and the grey band represent mean and standard error,respectively, derived from official yield statistics.

soil and weather data generally yielded no better results, some ofthem rather far off, as demonstrated by the error shown in Fig. 2(single point simulation, boxplot C). Only two combinations cameclose to the observed mean, both of which included the fourth mostabundant soil, a loamy Rendzic Leptosol developed from witheredrock (data not shown).

When additional soil information was included (single stationsimulation), the yield generated was 7.01 ± 1.47 t ha−1. The nearestneighbour approach yielded 6.60 ± 1.37 t ha−1 and the interpolatedweather information led to a prediction of 7.36 ± 1.17 t ha−1 (Fig. 2).The latter three simulation results differ insignificantly from theobserved value.

The approaches based on highly resolved soil information wereadditionally analysed with regard to their performance on the nextlower spatial scale, at county level. The single station, nearest neigh-bour and interpolation simulations of the time series in each of thecounties yielded very different results. Three examples of countiesare highlighted (Fig. 6). For “Altenburger Land” – a county domi-nated by loess soils that produce the highest average wheat yields inThuringia – all three approaches were able to reproduce the countymean and–albeit far from perfectly – inter-annual dynamics. For“Sömmerda” county the observed mean was overestimated by 1.60,1.42 and 1.31 t ha−1, respectively, and all three approaches – in asimilar manner – failed to reproduce inter-annual dynamics. For“Hildburghausen” county the observed mean was overestimatedand underestimated by the three approaches. All three approachesfailed in different ways to reproduce inter-annual dynamics. Thescatterplot of simulated winter wheat yields at county level againstobserved yields – here for the interpolation approach only – showsthat year-to-year yield dynamics were not reproduced by the model(Fig. 7).

Testing the model against the detrended mean data from vari-ety trials at six different experimental stations in Thuringia (Fig. 8)revealed that the model underestimated the observed mean at fiveof the six stations by between 1.04 and 1.98 t ha−1 (Table 2, MBE).The model only overestimated the mean yields at Burkersdorf,

namely by 0.48 t ha−1 (Table 2, MBE). However, data from three rel-atively high-yielding years were missing at Burkersdorf, reducingthe overall mean for this site (Table 2, O). The MAE of the simulationranged between 1.52 and 2.37 t ha−1 (Table 2, MAE), an error whichaccounts for 17–27% of the mean (Table 2, nMAE). Consequently,the Nash-Suttcliffe modelling efficiency was negative for all sites,and thus below the acceptable range of [0;1] (Table 2, MEd). Cor-recting the observed yield at each location by the respective MBE inorder to limit the model evaluation to the year-to-year differencesin yield slightly improved the result for two of the six locations, butworsened it for two others (Table 2, MEdc).

Additional meta information on the year-to-year conditions forcereal production, obtained from annual agricultural reports, pro-vides insight into specific reasons for deviations between observedyields and model predictions. Table 3 summarises the years inwhich the wheat yield was significantly reduced at certain loca-tions. Some of the growth-reducing factors (van Ittersum andRabbinge, 1997) are not considered in MONICA, which may explainsome of the model’s overpredictions in those years.

4. Discussion

4.1. Model applicability and data availability at different spatialscales

Simulation models are commonly evaluated against observeddata to prove that the joint functioning of all the implementedprocesses in the model produces a similar behaviour to that ofthe described system (Håkanson, 1995; Nendel et al., 2011). Thisdata is usually obtained from field experiments, which alwaysharbour a larger number of uncertainties compared to labora-tory experiments, most of which are due to heterogeneities of theenvironmental conditions in the field (Kersebaum et al., 2002). Bio-physical process models are unable to capture the whole range ofpossible conditions at the field scale because the investigated sys-tem is open (Oreskes et al., 1994). However, such models can stillpredict one of the conditions, which ideally should produce a meanpredictor of any desired target variable.

Further scaling to larger spatial units causes similar problems,albeit with fewer data to drive and evaluate the model (Hansen andJones, 2000). The philosophy behind biophysical models is based onthe assumption that biological and physical processes are the sameeverywhere. Once calibrated, a model should be applicable in sim-ilar ecosystems all over the world, as long as all relevant processesare considered. Accordingly, it should not be a problem to use amodel calibrated for a specific crop on the field scale for the samecrop in a larger area, even though the purpose-oriented design ofmany models may give reason for limitations, e.g. due to the con-sideration or non-consideration of particular processes. However,even for a perfect model the data problem remains (Hoogenboom,2000). Driving variables derived from soil or weather informationare available in much lower density and quality (Rivington et al.,2006). It could be that one or another essential driving variable isnot available at all. On the evaluation side, target variables may bemeasured – or even only estimated – at a few points or may onlybe available as average figures for a larger spatial unit. In our case,winter wheat yields were averaged across a range of variety tri-als to produce a mean yield figure for an experimental station andacross a large number of estimates provided by farmers to producea mean yield figure for a county.

4.2. Evaluation of simulation results at different scales

Beginning at the largest scale, MONICA simulations were closerto the observed federal state mean when using detailed soil

Author's personal copy

72 C. Nendel et al. / Field Crops Research 145 (2013) 67–77

Table 2Performance indices for MONICA simulations of 1992–2010 winter wheat yield data from the experimental stations (average across all grown varieties), including the numberof data pairs n, the mean observed winter wheat yield O, the mean bias error of prediction MBE, the mean absolute error of prediction MAE, the normalised mean absoluteerror of prediction nMAE, the Nash–Suttcliffe modelling efficiency against detrended yield data MEd and against detrended and mean-corrected yield data MEdc .

Station n O (t ha−1) MBE (t ha−1) MAE (t ha−1) nMAE MEd MEdc

Friemar 18 8.80 −1.04 1.52 0.17 −0.23 −0.34Burkersdorf 15 7.20 0.48 1.66 0.23 −1.20 −1.52Dornburg 18 9.82 −1.98 2.02 0.21 −0.17 −1.20Großenstein 19 9.43 −1.78 2.07 0.22 −0.48 −1.26Heßberg 15 8.95 −1.74 1.94 0.22 −1.64 −0.39Kirchengel 19 8.94 −1.90 2.37 0.27 −1.31 −0.09

Table 3Growth-reducing factors that occurred during the 1992–2010 period at experimental stations Burkersdorf (B), Friemar (F), Dornburg (D), Großenstein (G), Heßberg (H) andKirchengel (K) in Thuringia. Only those years in which MONICA generated significant overpredictions were considered.

Year Observed growth-reducing factor Location Considered in MONICA

1992 Late emergence due to autumn drought Alla YesTiller abortion due to warm spring NoSecond growth due to rainfall No

1993 Late emergence due to autumn drought All YesFrost damage H NoTiller abortion due to warm spring All NoLodging F No

1994 Late emergence due to low temperatures Allc YesFrost damage D NoSilting due to heavy rain (oxygen stress) B No

1998 Damage by rodents F NoDamage by wheat bulb fly K No

1999 Ponding damage F Yes2000 Tiller abortion due to warm spring D, G, H, K No2002 Heat stress B, D, F, G, K Yes

Lodging and harvest delay due to rainfall No2003 Crop withering due to April frost All No

Summer drought YesHeat stress Yes

2007 Spring drought Allb YesHarvest delay due to rainfall, second growth No

a No data available for Friemar and Burkersdorf.b No data available for Burkersdorf.c No data available for Heßberg.

information. Although there were two combinations of a soil anda weather station that produced the federal state mean very well,these combinations were not amongst the first a-priori selection.No convincing results were produced by choosing the most com-mon or second-most common soil and the most representativeweather station according to local experts. In addition to failingto predict the observed federal state mean, they also produced amuch higher standard error than the observations. Use of a highlyresolved grid of soil and relief information improved the predictionof the federal state mean and also provided a reasonable predictionof the standard error in time and space. The interpolation approachdelivered the smoothest pattern of yields, including a clear accen-tuation of the Thuringian Forest and the Thuringian Basin northof it (Fig. 5). The reason why the federal state mean was over-predicted using the interpolation approach is due to the fact thatMONICA was calibrated to field experiments that usually producehigher yields than the average farmer’s field (Hansen and Jones,2000). In this case, the difference between simulation and obser-vation is 0.7 t ha−1 which could be regarded as a bias correctorfor regional simulations of MONICA for winter wheat. However,the temporal relationship between simulated and observed yieldis very weak (Fig. 7) and a bias correction following the example ofJagtap and Jones (2002) would not substantially improve the pre-diction. Concerning the spatial pattern, the single station approachperformed in a similar way as the interpolation approach. How-ever, dramatically low winter wheat yields were simulated for claysoils around the Thuringian Basin and south of the Thuringian For-est under the influence of weather recorded at Erfurt-Bindersleben

weather station (orange-red colours in Fig. 3). These low yieldsfrom clay soils dragged the simulated federal state mean downto what in the end seemed to be a very good result when com-pared to the observations. A similar effect was observed for thenearest neighbour approach. Here, the sharp edges of the Thiessen

Fig. 3. 1992–2010 mean winter wheat yields in Thuringia, Germany, simulated ona 100 m × 100 m grid of soil and relief information using the MONICA model withweather data from Erfurt-Bindersleben meteorological station.

Author's personal copy

C. Nendel et al. / Field Crops Research 145 (2013) 67–77 73

Fig. 4. 1992–2010 mean winter wheat yields in Thuringia, Germany, simulated ona 100 m × 100 m grid of soil and relief information using the MONICA model withweather data from 14 weather stations following a nearest neighbour (Thiessenpolygon) approach.

polygons show through the spatial yield pattern, drawing a quiteunrealistic picture in the areas further away from the weather sta-tions. Fig. 4 reveals that the mountain weather station Schmücke,located 937 m above sea level, covered quite a large polygon, muchlarger than the Thuringian Forest. This means that the agriculturalarea at the foot of the mountain range was simulated with weatherdata recorded at Schmücke weather station, including the low tem-peratures typical for this altitude. The respective polygon clearlyshows significantly lower yields than the other polygons, againdragging the simulated federal state mean down and accidentlycorresponding quite well to the observed federal state mean (Fig. 4).

At county level, it becomes apparent that the model was able toreproduce the county mean for some counties, but overpredictedor underpredicted it for others. The three different approachesthat used high-resolution soil and relief information (single station,nearest neighbour and interpolation) performed quite differently,none of them showing any clear advantage over the others.More importantly, however, the model was unable to reproduceinter-annual yield dynamics in most of the counties. However,

Fig. 5. 1992–2010 mean winter wheat yields in Thuringia, Germany, simulated ona 100 m × 100 m grid of soil and relief information using the MONICA model withinterpolated and altitude-corrected temperature data from 14 weather stations andadditional precipitation data from 39 rain gauges.

Fig. 6. 1992–2010 mean winter wheat yields for three of 17 counties from the Fed-eral State of Thuringia, Germany, observed (black line) and simulated using theMONICA model on a 100 m × 100 m soil and relief information grid with weatherinformation from a single station (grey line), the nearest of 14 weather stations(light grey line) and interpolated and altitude-corrected from 14 weather stationsand 39 rain gauges (grey scattered line).

MONICA has already shown its ability to simulate winter wheatyields in uncalibrated situations, including two long-term field tri-als across different winter wheat varieties in Saxony, the federalstate that borders Thuringia to the east. Here, the yield level wasalso considerably underpredicted, but inter-annual dynamics werewell reproduced (Nendel et al., 2011).

The farm level, which comprises different variety trials at eachlocation, performs similarly. However, in addition to the inability toreproduce inter-annual yield dynamics, the model also significantlyunderpredicted the average yield level of the trials. This findingwas almost expected, as all of the trials included high-performancewinter wheat varieties to which MONICA was not calibrated. Theexperimental data used for model calibration included varietiesthat are well established on the market (Nendel et al., 2011). Thedifference between the two yield levels amounts to between 1 and2 t ha−1.

4.3. Following changing factors across scales

The overall effect of weather on crops can be seen on the field,county and federal state scale. Drought and heat stress impactsduring growth, the ontogenesis-accelerating effect of warm tem-perature, late frost as well as excessive rainfall during the harvestperiod are visible as significant yield reductions across all scales. Incontrast, mild temperatures, high radiation input and precipitationevenly distributed over time produce bumper yields, also across allscales. The effect of a rising atmospheric CO2 concentration is likelyto add to this; however, the considered time period in this study istoo short to identify any CO2 effect on wheat yields.

The spatial distribution of soils and relief in a landscape leadsto a differentiation of weather impact on yields, even though theweather affects the crop produced on a landscape in a similarmanner. Whilst loam soils with a high storage capacity for plant-available water may maintain sufficient water supplies to the cropduring shorter periods of drought, sand and clay soils may not. Hilltops dry out more quickly as soil water descends towards depres-sions, where it provides extra moisture to a thirsty crop or – incase of excessive water supply – drowns it (Manning et al., 2001;Priyashantha et al., 2007). The differentiation of crop growth byrelief is even visible at the field scale. If high-resolution soil and

Author's personal copy

74 C. Nendel et al. / Field Crops Research 145 (2013) 67–77

Fig. 7. Simulated 1992–2010 mean winter wheat yields for 17 Thuringia counties using interpolated weather data vs. observed yields from official statistics (n = 323).Highlighted A: Counties Hildburghausen (×), Sömmerda (©), Altenburger Land (+); B: Years 1992 (�), 2000 (×), 2002 (©), 2006 (�).

relief data is available, biophysical simulation models are good atcapturing these differences (Reuter and Kersebaum, 2009; Sadleret al., 2000; Wendroth et al., 2011). However, sub-surface waterflow is only considered in a few three-dimensional models (e.g.Hanson et al., 2004; Paydar and Gallant, 2008). At the field level, thedifference between soils may still be small and less important thanthe relief effect. However, if the relief amplitude remains within acertain range across the landscape, a model can successfully pre-dict mean crop growth for a specific soil, and soil patterns becomemore important.

At county level, weather may be still rather homogeneous whilstsoils may not. However, macro-relief may then become the mostimportant factor (mountain ranges, valleys, long-range altitudegradients). With the example of Thuringia, some counties includeparts of the Thuringian forest, a mountain ridge peaking at 983 mabove sea level. For submontane areas, the lower average tem-perature at higher altitudes partly superimposes the differencesin crop growth on different soils within a county. The presence ofa macro-relief may also significantly affect rainfall patterns. Effectsof smaller relief structures (undulating moraine landscape, river

terraces, etc.) become less important, also because data availabilitybecomes insufficient. Even a 100 m × 100 m grid of relief informa-tion as used in our study is too coarse to capture the true slopeangles of the relief properly. Solar angle (Reuter et al., 2005) andrainfall surface run-off calculations (Nippgen et al., 2011; Rihaniet al., 2010; Saue and Kadaja, 2009) become erroneous.

For higher spatial levels – national or continental – Rivingtonet al. (2006) demonstrated the uncertainty introduced to a simula-tion by using weather data via the nearest neighbour approach overlarger distances. They found that meteorological stations could beeasily substituted by others within a few 10 km range, but largeerrors were introduced by using meteorological stations from over100 km distance.

4.4. Growth-reducing factors and processes not considered inMONICA

The analysis of the meta data provided with annual yield figuresfor Thuringia revealed that there were many reasons why winterwheat was not produced at the maximum level of its physiologi-

Fig. 8. Difference between MONICA-simulated 1992–2010 winter wheat yields and detrended, mean-corrected observed yield data averaged across all grown varieties atdifferent experimental stations in Thuringia, Germany. Missing data (�).

Author's personal copy

C. Nendel et al. / Field Crops Research 145 (2013) 67–77 75

cal potential. The first group of factors relate to climate: extremeheat and frost damaged the crop and reduced its ability to producegrain. Accelerated ontogenesis due to warm temperatures short-ened the grain filling period and led to lower 1000 grain weights.Drought and the associated limitation to take up soil mineral nitro-gen reduced photosynthesis or halted it altogether. These processesare considered in MONICA and should be captured by the model.

A second group of factors also relate to climate-influenced plantphysiological processes. The only difference between this groupand the first is that these processes are not considered in MON-ICA. This group includes tiller initiation of winter wheat in relationto the soil mineral N supply in autumn, tiller abortion due tospring drought or warmth and tiller regrowth during a rainy har-vest period and the production of green grains that interminglewith matured grain and may clog the elevators of older combineharvesters; no mention is made of the subsequent problems withgrain storage if a part of the grain yield is still green. The modelconcept of MONICA is source-driven, whilst consideration of theseprocesses would require the additional modelling of sink physi-ology. Ponding and subsequent oxygen deficiency of the crop isconsidered in MONICA. However, the spatial distribution of pondsin a field, their extension and their infiltration rate is extremelyvariable and difficult to predict.

A third group of factors, also related to climate, is not consid-ered in MONICA because the weather data required to drive therespective algorithms is not available in sufficient quality and tem-poral resolution to deliver acceptable simulations. This includesstem lodging, for which the speed and duration of wind gusts andrain intensity are decisive factors, and crop damage from hail, theappearance of which is often so local that it cannot be measured orpredicted well in scenarios.

Finally, a fourth group of factors includes technological reasonsfor failing to gain a perfect yield. Amongst others, it includes fieldinaccessibility due to rain-soaked soils, machine failure or unavail-ability and the failure of individual farmers to protect their crops orto perform fertiliser management. So far, there has been insufficientdata to feed a crop simulation model to enable it to simulate theseeffects. However, whole-farm modelling approaches and agent-based modelling may be of use in this respect (e.g. Rebaudo andDangles, 2011).

5. Conclusions

It is apparent from the present simulation study that the goodresult obtained for the reproduction of federal state-level winterwheat yields using single station and nearest neighbour simulationis based on compensation of errors that operate at the lower scaleand affect yield estimates in different directions: on the one hand,the model overestimates yields at the farm level due to the cal-ibration with data from optimum-conditioned field experimentsthat inadequately represent average agricultural production. Onthe other hand, the weather information provided with inade-quate spatial distribution leads to considerable underestimationof yields in some areas. When using the single station and near-est neighbour approach, these two effects cancel each other out.Weather information is distributed much more accurately in spacewhen the interpolation approach is used. Hence, using this approachthe overprediction of yields due to inadequate calibration of themodel at the federal state level prevails. We learned from thisexercise that in regions with agricultural production across a signif-icant altitude range the use of Thiessen polygons for extrapolationof weather data may include climate–soil combinations that arenot representative for the production area. Same applies for theuse of one representative climate station, for which the area of

representativeness may be even smaller than the range identifiedby Rivington et al. (2006).

The simulation result for the experimental stations and countiesin Thuringia reveal a number of growth-reducing factors for whichnone of the currently available crop models can account for. Har-vest problems related to technology are not yet considered in cropmodels and lodging of cereals after heavy storms, hail damageand the failure of farmers to conduct crop protection managementpose a distinct challenge to modellers. Some of these growth-reducing factors may be addressed in the future. Model approachesfor cereal lodging have already been presented (Berry et al., 2003,2006; Cui and Shen, 2011; Martinez-Vazquez and Sterling, 2011)and it also seems feasible to be able to develop a model approachfor field accessibility. It will be interesting to see how combiningremote sensing with improved bio-physical and stochastic simu-lation models (de Wit et al., 2012; de Wit and van Diepen, 2007;Hu and Mo, 2011; Nearing et al., 2012) will contribute to furtherdevelopment of model-based regional yield estimation (Moen et al.,1994; Moulin et al., 1998; Savary et al., 2006).

Acknowledgement

The authors gratefully acknowledge the provision of climatedata by the German Weather Service.

References

Addiscott, T.M., Whitmore, A.P., 1987. Computer-simulation of changes in soil min-eral nitrogen and crop nitrogen during autumn, winter and spring. J. Agric. Sci.109, 141–157.

Alexandrov, V., Eitzinger, J., Cajic, V., Oberforster, M., 2002. Potential impact ofclimate change on selected agricultural crops in north-eastern Austria. Glob.Change Biol. 8, 372–389.

Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration. Guide-lines for computing crop water requirements. FAO Irrigation and Drainage Paper56, Roma.

Andersson, J.C.M., Zehnder, A.J.B., Wehrli, B., Yang, H., 2012. Improved SWAT modelperformance with time-dynamic Voronoi tessellation of climatic input data inSouthern Africa. J. Am. Water Resour. Assoc. 48, 480–493.

Baron, C., Sultan, B., Balme, M., Sarr, B., Traore, S., Lebel, T., Janicot, S., Dingkuhn, M.,2005. From GCM grid cell to agricultural plot: scale issues affecting modellingof climate impact. Philos. Trans. R. Soc. B: Biol. Sci. 360, 2095–2108.

Bennett, A.J., Bending, G.D., Chandler, D., Hilton, S., Mills, P., 2012. Meeting thedemand for crop production: the challenge of yield decline in crops grown inshort rotations. Biol. Rev. 87, 52–71.

Berry, P.M., Sterling, M., Baker, C.J., Spink, J., Sparkes, D.L., 2003. A calibrated modelof wheat lodging compared with field measurements. Agric. For. Meteorol. 119,167–180.

Berry, P.M., Sterling, M., Mooney, S.J., 2006. Development of a model of lodging forbarley. J. Agron. Crop. Sci. 192, 151–158.

Bondeau, A., Smith, P.C., Zaehle, S., Schaphoff, S., Lucht, W., Cramer, W., Gerten, D.,Lotze-Campen, H., Müller, C., Reichstein, M., Smith, B., 2007. Modelling the role ofagriculture for the 20th century global terrestrial carbon balance. Glob. ChangeBiol. 13, 679–706.

Cassman, K.G., 2007. Climate change, biofuels, and global food security. Environ. Res.Lett. 2, 1–3, 011002.

Challinor, A.J., Wheeler, T.R., Craufurd, P.Q., Slingo, J.M., 2005. Simulation of theimpact of high temperature stress on annual crop yields. Agric. For. Meteorol.135, 180–189.

Cui, H.L., Shen, H.S., 2011. Modeling and simulation of buckling and postbuckling ofplant stems under combined loading conditions. Int. J. Appl. Mech. 3, 119–130.

Das, S., Priess, J.A., Schweitzer, C., 2012. Modelling regional scale biofuel scenarios –a case study for India. Glob. Change Biol. Bioenergy 4, 176–192.

De Willigen, P., 1991. Nitrogen turnover in the soil-crop system – comparison of 14simulation models. Fert. Res. 27, 141–149.

de Wit, A.J.W., Boogaard, H.L., van Diepen, C.A., 2005. Spatial resolution of pre-cipitation and radiation: the effect on regional crop yield forecasts. Agric. For.Meteorol. 135, 156–168.

de Wit, A.J.W, Duveiller, G., Defourny, P., 2012. Estimating regional winter wheatyields with WOFOST through the assimilation of green area index retrieved fromMODIS observations. Agric. For. Meteorol. 164, 39–52.

de Wit, A.M., van Diepen, C.A., 2007. Crop model data assimilation with the EnsembleKalman filter for improving regional crop yield forecasts. Agric. For. Meteorol.146, 38–56.

Déqué, M., Rowell, D.P., Luthi, D., Giorgi, F., Christensen, J.H., Rockel, B., Jacob, D.,Kjellstrom, E., de Castro, M., van den Hurk, B., 2007. An intercomparison ofregional climate simulations for Europe: assessing uncertainties in model pro-jections. Clim. Change 81, 53–70.

Author's personal copy

76 C. Nendel et al. / Field Crops Research 145 (2013) 67–77

Deryng, D., Sacks, W.J., Barford, C.C., Ramankutty, N., 2011. Simulating the effectsof climate and agricultural management practices on global crop yield. GlobBiochem. Cycles 25, GB2006.

Diekkrüger, B., Söndgerath, D., Kersebaum, K.C., McVoy, C.W., 1995. Validity of agroe-cosystem models – a comparison of results of different models applied to thesame data set. Ecol. Model. 81, 3–29.

Easterling, W.E., Weiss, A., Hays, C.J., Mearns, L.O., 1998. Spatial scales of climateinformation for simulating wheat and maize productivity: the case of the USGreat Plains. Agric. For. Meteorol. 90, 51–63.

Elsgaard, L., Borgesen, C.D., Olesen, J.E., Siebert, S., Ewert, F., Peltonen-Sainio, P., Rot-ter, R.P., Skjelvag, A.O., 2012. Shifts in comparative advantages for maize, oat andwheat cropping under climate change in Europe. Food Addit. Contam. A: Chem.Anal. Control Expo. Risk Assess. 29, 1514–1526.

Ewert, F., van Ittersum, M.K., Heckelei, T., Therond, O., Bezlepkina, I., Andersen, E.,2011. Scale changes and model linking methods for integrated assessment ofagri-environmental systems. Agric. Ecosyst. Environ. 142, 6–17.

Faivre, R., Leenhardt, D., Voltz, M., Benoit, M., Papy, F., Dedieu, G., Wallach, D., 2004.Spatialising crop models. Agronomie 24, 205–217.

Folberth, C., Yang, H., Wang, X.Y., Abbaspour, K.C., 2012. Impact of input data res-olution and extent of harvested areas on crop yield estimates in large-scaleagricultural modeling for maize in the USA. Ecol. Model. 235, 8–18.

Foley, J.A., Ramankutty, N., Brauman, K.A., Cassidy, E.S., Gerber, J.S., Johnston, M.,Mueller, N.D., O’Connell, C., Ray, D.K., West, P.C., Balzer, C., Bennett, E.M., Car-penter, S.R., Hill, J., Monfreda, C., Polasky, S., Rockstrom, J., Sheehan, J., Siebert,S., Tilman, D., Zaks, D.P., 2011. Solutions for a cultivated planet. Nature 478,337–342.

Gimona, A., Birnie, R.V., Sibbald, A.R., 2006. Scaling up of a mechanistic dynamicmodel in a GIS environment to model temperate grassland production at theregional scale. Grass Forage Sci. 61, 315–331.

Gutzler, C., Nendel, C., Mirschel, W. Yield development in Germany between 1980and 2010: a trend analysis for winter wheat, winter rye, rapeseed and silagemaize, in preparation.

Håkanson, L., 1995. Optimal size of predictive models. Ecol. Model. 78, 195–204.Hansen, J.W., Jones, J.W., 2000. Scaling-up crop models for climate variability appli-

cations. Agric. Syst. 65, 43–72.Hansen, S., Jensen, H.E., Nielsen, N.E., Svendsen, H., 1991. Simulation of nitrogen

dynamics and biomass production in winter-wheat using the Danish simulation-model DAISY. Fert. Res. 27, 245–259.

Hanson, P.J., Amthor, J.S., Wullschleger, S.D., Wilson, K.B., Grant, R.F., Hartley, A.,Hui, D., Hunt, E.R., Johnson, D.W., Kimball, J.S., King, A.W., Luo, Y., McNulty, S.G.,Sun, G., Thornton, P.E., Wang, S., Williams, M., Baldocchi, D.D., Cushman, R.M.,2004. Oak forest carbon and water simulations: model intercomparisons andevaluations against independent data. Ecol. Monogr. 74, 443–489.

Hartwich, R., Behrens, J., Eckelmann, W., Haase, G., Richter, A., Roeschmann, G.,Schmidt, R., 1995a. Bodenübersichtskarte der Bundesrepublik Deutschland1:1 000 000 (BÜK1000) – Erläuterungen und Textlegende (Beiheft zur Karte).Hanover, Germany.

Hartwich, R., Haase, G., Richter, A., Roeschmann, G., Schmidt, R., 1995b. Soil surveymap of Germany 1:1 000 000 (BÜK1000). Hanover, Germany.

Hoogenboom, G., 2000. Contribution of agrometeorology to the simulation of cropproduction and its applications. Agric. For. Meteorol. 103, 137–157.

Hu, S., Mo, X.G., 2011. Interpreting spatial heterogeneity of crop yield with a processmodel and remote sensing. Ecol. Model. 222, 2530–2541.

Jagtap, S.S., Jones, J.W., 2002. Adaptation and evaluation of the CROPGRO-soybeanmodel to predict regional yield and production. Agric. Ecosyst. Environ. 93,73–85.

Keil, M., Kiefl, R., Strunz, G., 2005. CORINE Land Cover 2000 – Germany. Final Report.Oberpfaffenhofen, Germany.

Kersebaum, K.C., 1995. Application of a simple management model to simulate waterand nitrogen dynamics. Ecol. Model. 85, 145–156.

Kersebaum, K.C., 2007. Modelling nitrogen dynamics in soil-crop systems with HER-MES. Nutr. Cycl. Agroecosyst. 77, 39–52.

Kersebaum, K.C., Hecker, J.-M., Mirschel, W., Wegehenkel, M., 2007. Modelling waterand nutrient dynamics in soil-crop systems: a comparison of simulation modelsapplied on common data sets. In: Kersebaum, K.C., Hecker, J.-M., Mirschel, W.,Wegehenkel, M. (Eds.), Modelling Water and Nutrient Dynamics in Soil CropSystems. Springer, Stuttgart, pp. 1–17.

Kersebaum, K.C., Lorenz, K., Reuter, H.I., Wendroth, O., 2002. Modeling crop growthand nitrogen dynamics for advisory purposes regarding spatial variability. In:Ahuja, L.R., Ma, L.W., Howell, T.A. (Eds.), Agricultural System Models in FieldResearch and Technology Transfer. Lewis Publishers, Boca Raton, USA, pp.229–251.

Liu, J.G., Williams, J.R., Zehnder, A.J.B., Yang, H., 2007. GEPIC – modelling wheat yieldand crop water productivity with high resolution on a global scale. Agric. Syst.94, 478–493.

Lobell, D.B., Burke, M.B., Tebaldi, C., Mastrandrea, M.D., Falcon, W.P., Naylor, R.L.,2008. Prioritizing climate change adaptation needs for food security in 2030.Science 319, 607–610.

Lobell, D.B., Cassman, K.G., Field, C.B., 2009. Crop yield gaps: their importance, mag-nitudes, and causes. Annu. Rev. Environ. Resour. 34, 204.

Manning, G., Fuller, L.G., Eilers, R.G., Florinsky, I., 2001. Soil moisture and nutrientvariation within an undulating Manitoba landscape. Can. J. Soil. Sci. 81, 449–458.

Martinez-Vazquez, P., Sterling, M., 2011. Predicting wheat lodging at large scales.Biosyst. Eng. 109, 326–337.

Miguez, F.E., Maughan, M., Bollero, G.A., Long, S.P., 2012. Modeling spatial anddynamic variation in growth, yield, and yield stability of the bioenergy crops

Miscanthus × giganteus and Panicum virgatum across the conterminous UnitedStates. Glob. Change Biol. Bioenergy 4, 509–520.

Mirschel, W., Wenkel, K.-O., 2007. Modelling soil–crop interactions with AGROSIMmodel family. In: Kersebaum, K.C., Hecker, J.-M., Mirschel, W., Wegehenkel, M.(Eds.), Modelling Water and Nutrient Dynamics in Soil Crop Systems. Springer,Stuttgart, pp. 59–74.

Moen, T.N., Kaiser, H.M., Riha, S.J., 1994. Regional yield estimation using a cropsimulation model – concepts, methods, and validation. Agric. Syst. 46, 79–92.

Moriondo, M., Bindi, M., Kundzewicz, Z.W., Szwed, M., Chorynski, A., Matczak, P.,Radziejewski, M., McEvoy, D., Wreford, A., 2010. Impact and adaptation oppor-tunities for European agriculture in response to climatic change and variability.Mitig. Adapt. Strateg. Glob. Change 15, 657–679.

Moulin, S., Bondeau, A., Delecolle, R., 1998. Combining agricultural crop modelsand satellite observations: from field to regional scales. Int. J. Remote Sens. 19,1021–1036.

Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models.Part I. A discussion of principles. J. Hydrol. 10, 282–290.

Nearing, G.S., Crow, W.T., Thorp, K.R., Moran, M.S., Reichle, R.H., Gupta, H.V., 2012.Assimilating remote sensing observations of leaf area index and soil moisturefor wheat yield estimates: an observing system simulation experiment. WaterResour. Res. 48, W05525.

Nendel, C., 2009. Evaluation of Best Management Practises for N fertilisation inregional field vegetable production with a small scale simulation model. Eur.J. Agron. 30, 110–118.

Nendel, C., Berg, M., Kersebaum, K.C., Mirschel, W., Specka, X., Wegehenkel, M.,Wenkel, K.O., Wieland, R., 2011. The MONICA model: testing predictability forcrop growth, soil moisture and nitrogen dynamics. Ecol. Model. 222, 1614–1625.

Nendel, C., Kersebaum, K.C., Mirschel, W., Manderscheid, R., Weigel, H.J., Wenkel, K.-O., 2009. Testing different CO2 response algorithms against a FACE crop rotationexperiment. NJAS: Wageningen J. Life Sci. 57, 17–25.

Neumann, K., Verburg, P.H., Stehfest, E., Muller, C., 2010. The yield gap of global grainproduction: a spatial analysis. Agric. Syst. 103, 316–326.

Nippgen, F., McGlynn, B.L., Marshall, L.A., Emanuel, R.E., 2011. Landscape structureand climate influences on hydrologic response. Water Resour. Res. 47, W12528.

Olesen, J.E., Bocher, P.K., Jensen, T., 2000. Comparison of scales of climate and soil datafor aggregating simulated yields of winter wheat in Denmark. Agric. Ecosyst.Environ. 82, 213–228.

Olesen, J.E., Carter, T.R., Díaz-Ambrona, C.H., Fronzek, S., Heidmann, T., Hickler, T.,Holt, T., Minguez, M.I., Morales, P., Palutikof, J.P., Quemada, M., Ruiz-Ramos, M.,Rubaek, G.H., Sau, F., Smith, B., Sykes, M.T., 2007. Uncertainties in projectedimpacts of climate change on European agriculture and terrestrial ecosys-tems based on scenarios from regional climate models. Clim. Change 81,123–143.

Oreskes, N., Shraderfrechette, K., Belitz, K., 1994. Verification, validation, and con-firmation of numerical-models in the earth-sciences. Science 263, 641–646.

Palosuo, T., Kersebaum, K.C., Angulo, C., Hlavinka, P., Moriondo, M., Patil, R., Ruget,F., Rumbaur, C., Takác, J., Trnka, M., Bindi, M., Caldag, B., Ewert, F., Ferrise, R.,Mirschel, W., Olesen, J., Saylan, L., Siska, B., Rötter, R.P., 2011. Simulation of win-ter wheat yield and its variability in different climates of Europe. A comparisonof eight crop growth models. Eur. J. Agron. 35, 103–114.

Parry, M.L., Rosenzweig, C., Iglesias, A., Livermore, M., Fischer, G., 2004. Effects ofclimate change on global food production under SRES emissions and socio-economic scenarios. Glob. Environ. Change 14, 53–67.

Paydar, Z., Gallant, J., 2008. A catchment framework for one-dimensional models:introducing FLUSH and its application. Hydrol. Process. 22, 2094–2104.

Pedersen, A., Zhang, K.F., Thorup-Kristensen, K., Jensen, L.S., 2010. Modelling diverseroot density dynamics and deep nitrogen uptake – a simple approach. Plant Soil326, 493–510.

Porter, J.R., Challinor, A., Ewert, F., Falloon, P., Fischer, T., Gregory, P., van Ittersum,M.K., Olesen, J.E., Moore, K.J., Rosenzweig, C., Smith, P., 2010. Food security: focuson agriculture. Science 328, 172–173.

Priyashantha, K.R.S., Maule, C.P., Elliott, J.A., 2007. Influence of slope position and hogmanure injection on fall soil P and N distribution in an undulating landscape.Trans. Asabe 50, 45–52.

Rebaudo, F., Dangles, O., 2011. Coupled information diffusion-pest dynamics modelspredict delayed benefits of farmer cooperation in pest management programs.PLoS Comput. Biol. 7, e1002222.

Reuter, H.I., Kersebaum, K.C., 2009. Applications in precision agriculture. In: Hengl,T., Reuter, H.I. (Eds.), Geomorphometry: Concepts, Software, Applications. Else-vier, Amsterdam, pp. 623–636.

Reuter, H.I., Kersebaum, K.C., Wendroth, O., 2005. Modelling of solar radiationinfluenced by topographic shading – evaluation and application for precisionfarming. Phys. Chem. Earth 30, 143–149.

Rihani, J.F., Maxwell, R.M., Chow, F.K., 2010. Coupling groundwater and land surfaceprocesses: idealized simulations to identify effects of terrain and subsurfaceheterogeneity on land surface energy fluxes. Water Resour. Res. 46, W12523.

Rivington, M., Matthews, K.B., Bellocchi, G., Buchan, K., 2006. Evaluating uncertaintyintroduced to process-based simulation model estimates by alternative sourcesof meteorological data. Agric. Syst. 88, 451–471.

Rosenzweig, C., Jones, J.W., Hatfield, J.L., Ruane, A.C., Boote, K.J., Thorburn, P., Antle,J.M., Nelson, G.C., Porter, C., Janssen, S., Asseng, S., Basso, B., Ewert, F., Wallach,D., Baigorria, G., Winter, J.M., 2013. The Agricultural Model Intercomparisonand Improvement Project (AgMIP): protocols and pilot studies. Forest. Agr.Meteorol. 170, 166–182.

Rötter, R.P., Carter, T.R., Olesen, J.E., Porter, J.R., 2011. Crop-climate models need anoverhaul. Nat. Clim. Change 1, 175–177.

Author's personal copy

C. Nendel et al. / Field Crops Research 145 (2013) 67–77 77

Rötter, R.P., Palosuo, T., Kersebaum, K.C., Angulo, C., Bindi, M., Ewert, F., Ferrise, R.,Hravlinka, P., Moriondo, M., Nendel, C., Olesen, J.E., Patil, R., Ruget, F., Tacác, J.,Trnka, M., 2012. Simulation of spring barley yield in different climatic zones ofNorthern and Central Europe: a comparison of nine crop models. Field CropsRes. 133, 23–36.

Rykiel, E.J., 1996. Testing ecological models: the meaning of validation. Ecol. Model.90, 229–244.

Sadler, E.J., Gerwig, B.K., Evans, D.E., Busscher, W.J., Bauer, P.J., 2000. Site-specificmodeling of corn yield in the SE coastal plain. Agric. Syst. 64, 189–207.

Salo, T. et al. Comparing the performance of eleven agro-ecosystem models in pre-dicting crop yield response to nitrogen under Finnish weather conditions. FieldCrops Res., in preparation.

Saue, T., Kadaja, J., 2009. Modelling crop yield response to precipitation redistribu-tion on slopes. Biologia 64, 502–506.

Savary, S., Teng, P.S., Willocquet, L., Nutter, F.W., 2006. Quantification and modelingof crop losses: a review of purposes. Annu. Rev. Phytopathol. 44, 89–112.

Schmidhuber, J., Tubiello, F.N., 2007. Global food security under climate change. Proc.Natl. Acad. Sci. U.S.A. 104, 19703–19708.

Shaeffer, D.L., 1980. Model evaluation methodology applicable to environmentalassessment models. Ecol. Model. 8, 275–295.

Stehfest, E., Heistermann, M., Priess, J.A., Ojima, D.S., Alcamo, J., 2007. Simulation ofglobal crop production with the ecosystem model DayCent. Ecol. Model. 209,203–219.

Supit, I., van Diepen, C.A., de Wit, A.J.W., Wolf, J., Kabat, P., Baruth, B., Ludwig, F.,2012. Assessing climate change effects on European crop yields using the CropGrowth Monitoring System and a weather generator. Agric. For. Meteorol. 164,96–111.

Therond, O., Hengsdijk, H., Casellas, E., Wallach, D., Adam, M., Belhouchette, H.,Oomen, R., Russell, G., Ewert, F., Bergez, J.E., Janssen, S., Wery, J., van Itter-sum, M.K., 2011. Using a cropping system model at regional scale: low-dataapproaches for crop management information and model calibration. Agric.Ecosyst. Environ. 142, 85–94.

Thiessen, A.H., 1911. Precipitation averages for large areas. Mon. Weather Rev. 39,1082–1084.

UN, 2011. World population prospects: the 2010 revision – highlights and advancetables. Working Paper No. ESA/P/WP.220. New York.

van Bussel, L.G.J., Ewert, F., Leffelaar, P.A., 2011. Effects of data aggregation on sim-ulations of crop phenology. Agric. Ecosyst. Environ. 142, 75–84.

van Ittersum, M.K., Rabbinge, R., 1997. Concepts in production ecology for analysisand quantification of agricultural input–output combinations. Field Crops Res.52, 197–208.

van Keulen, H., Penning de Vries, F.W.T., Drees, E.M., 1982. A summary model forcrop growth. In: Penning de Vries, F.W.T., van Laar, H.H. (Eds.), Simulation ofPlant Growth and Crop Production. PUDOC, Wageningen, pp. 87–97.

VanLoocke, A., Twine, T.E., Zeri, M., Bernacchi, C.J., 2012. A regional comparison ofwater use efficiency for miscanthus, switchgrass and maize. Agric. For. Meteorol.164, 82–95.

Vermeulen, S.J., Aggarwal, P.K., Ainslie, A., Angelone, C., Campbell, B.M., Challinor,A.J., Hansen, J.W., Ingram, J.S.I., Jarvis, A., Kristjanson, P., Lau, C., Nelson, G.C.,Thornton, P.K., Wollenberg, E., 2012. Options for support to agriculture and foodsecurity under climate change. Environ. Sci. Policy 15, 136–144.

Wassenaar, T., Lagacherie, P., Legros, J.P., Rounsevell, M.D.A., 1999. Modelling wheatyield responses to soil and climate variability at the regional scale. Clim. Res. 11,209–220.

Watson, D.F., Philip, G.M., 1985. A refinement of inverse distance weighted interpo-lation. Geo-Processing 2, 315–327.

Wegehenkel, M., 2000. Test of a modelling system for simulating water balancesand plant growth using various different complex approaches. Ecol. Model. 129,39–64.

Wehrmann, J., Scharpf, H.C., 1979. Mineral nitrogen concentration of soil as a gaugeof need for nitrogen-fertilizer (N-min method). Plant Soil 52, 109–126.

Wendroth, O., Kersebaum, K.C., Schwab, G., Murdock, L., 2011. Spatial relationshipof soil properties, crop indices, and nitrogen application pattern with wheatgrowth and yield in a field. In: Ahuja, L.R., Ma, L.W. (Eds.), Methods of IntroducingSystem Models into Agricultural Research. ASA, CSSA, SSSA, Madison, WI, USA,pp. 229–259.

White, J.W., Hoogenboom, G., Kimball, B.A., Wall, G.W., 2011. Methodologies forsimulating impacts of climate change on crop production. Field Crops Res. 124,357–368.

Wolf, J., van Diepen, C.A., 1995. Effects of climate change on grain maize yield poten-tial in the European Community. Clim. Change 29, 299–331.

Xiong, W., Holman, I., Conway, D., Lin, E., Li, Y., 2008. A crop model cross calibrationfor use in regional climate impacts studies. Ecol. Model. 213, 365–380.

Zhang, X.C., Liu, W.Z., Li, Z., Chen, J., 2011. Trend and uncertainty analysis of simulatedclimate change impacts with multiple GCMs and emission scenarios. Agric. For.Meteorol. 151, 1297–1304.