Introduction any large region. Considerable problems can arise Materials and methods

European Journal of Agronomy 11 1999 239–253 www.elsevier.comlocateeja Variability of winter rye grain yield in a glacial plain catchment — modelling and observation G.M. Richter Soil Science Department, IACR-Rothamsted, Harpenden AL5 2JQ, UK Accepted 30 June 1999 Abstract Crop growth of winter rye Secale cereale was simulated on sandy soils in northwest Germany at the meso-scale 6 km 2 catchment, 5 year rotation. Based on site-specific soil data, simulated grain yields YMOD were compared to observed yields Y OBS , and the results were discussed with respect to the N balance. The mean annual yields simulated for rain-fed 3.8–7.3 tha and irrigated conditions 5.2–8 tha varied more strongly than those observed in the region 4.6–6.0 tha. Excluding years with strongly deviating simulation results, the ratio Y OBS YMOD decreased from 0.82 to 0.73 under irrigation. Spatially, the effect of plant available water PAW on simulated yields varied considerably over the years, explaining 5–75 of the yield variability. In reality, yields depended on water availability PAW +irrigation in a year with spring drought, suggesting a time-specific water stress function. In the simulation scenario, there was ample N supply, and variable N mineralization explained 10 of the yield variability. Simulated mean N uptake and export were greater than the observed 160 vs. 140 and 112 vs. 100 kg Nha, respectively, which may explain the overestimated growth rates. The mean harvest index varied less 0.36–0.55 than the fraction of N returned with the residues 0.17–0. 50, and irrigation decreased the variation of both. With respect to scaling methods, soil data could be aggregated, and simulations based on distributed inputs could be substituted by the weighted mean of simulations based on mean inputs. Down-scaling from the EU map created the difficulty of selecting appropriate soil units. The simple factor Y OBS Y MOD was unreliable for scaling simulated yields to assess regional yields because of its temporal variation in response to climatic variables. The results suggest that changes in model structure and parameters are required to describe water stress, sink limitations, and N diffusion or influx rates better. © 1999 Elsevier Science B.V. All rights reserved. Keywords: Scaling-up; Soil spatial variability; Water stress; Weather variability; Yield prediction

1. Introduction any large region. Considerable problems can arise

in transferring models and scaling-up such multi- Crop growth models are used at the field and component systems, and accurate estimates of regional scale to predict yields and the environmen- yields seem difficult to achieve by either explicit or tal impact of land use. Usually, they have been simplified simulations Wolf, 1997; Landau et al., calibrated on a few experimental sites for a few 1998. Macroscopically, models can reproduce the years, and their reliability has not been proved for range and variation of yields at the scale of a whole country when based on spatially distributed inputs Butterfield et al., 1997. However, simula- Tel.: +44-1582-763-133. fax: +44-1582-769-222. E-mail address: goetz.richterbbsrc.ac.uk G.M. Richter tions overestimate observed yields under a wide 1161-030199 – see front matter © 1999 Elsevier Science B.V. All rights reserved. PII: S 1 1 6 1 -0 3 0 1 9 9 0 0 03 5 - 0 240 G.M. Richter European Journal of Agronomy 11 1999 239–253 range of environmental conditions, suggesting the quantifies the effects of 1 water availability on yield and N return to the system and 2 the need for regional correction factors de Koning and van Diepen, 1993. This is a pragmatic, but temporal stability of these outputs under variable weather conditions. The aggregated means of indi- questionable, approach because uncertainties con- cerning input variables, model parameters and vidual simulations are compared to mean field and regional observations, including statistical records structure Russell and van Gardingen, 1997 are lumped into a single factor. Searching for adequate and state-wide experiments in Northern Germany. resolution of inputs and model complexity becomes indispensable for regional studies Smith, 1996; Downing, 1997; van Gardingen et al., 1997.

2. Materials and methods

Further, appropriate data are needed to test the underlying assumptions of scaling-up. The strategy of this study was to run an integ- rated crop-soil model that had been calibrated Linking crop models to distributed inputs via Geographic Information Systems GIS has great earlier for winter rye Richter, 1996, using distrib- uted input data. For simulation, the inputs were potential not only for spatially optimizing agricul- tural management Bouma, 1997 and assessing supplied by a GIS in several information layers soils, rotation, management, and the model was land-use impacts Richter et al., 1998, but also for model validation. Data provided on the catch- run individually for all ecotopes over a period of 5 years. The ecotopes evaluated here represent ment level will minimize erroneous model outputs caused by input limitations and allow the adequacy combinations of crop rotation positions and crop management factors on homogeneous soil units. of the models’ structure and parameterization to be tested. Furthermore, effects of input aggregation Comparisons of simulated and observed yields were based on field non-homogeneous means on the model output can be analysed. Multiple model non-linearities might obstruct the assump- weighted by area of the ecotopes, thus validating the crop growth model on all fields in the catch- tion that the mean of all individual simulations equals the simulation of the mean Addiscott and ment where yield records by farmers were available. Mirza, 1998. Two hypotheses thus emerge from the literature, concerning the scaling of crop models: hypothesis A suggests that detailed soil 2.1. Soil resources data can be aggregated or drawn from an up-scaled source without losing any precision in estimating The study was done for the Eisenbach catch- ment in Northern Germany on soils with a texture regional crop growth; hypothesis B proposes a regional or technological correction factor that is ranging from medium-fine and loamy sand mfS, lS to sandy loam sL. The area comprised about unique and temporally stable for a specific crop. Earlier studies de Koning and van Diepen, 1993; 500 ha of arable land that had been intensively surveyed and used previously to develop and com- Wolf, 1997 implied these assumptions, which needed to be verified on detailed yield records at pare various models Diekkru¨ger et al., 1995. According to the 1:1 000 000 EC soil map, the the catchment scale over a range of soils and environmental conditions. region is dominated by Ferro-humic Podsols on wind-blown sands and Podsoluvisols on medium Two principal objectives follow from the above regarding scaling-up: 1 to assess the necessary to medium-fine colluvium CEC, 1985. At the 1:5000 scale, soil data were available in digital spatial resolution of the model input and 2 to prove the validity of the underlying biophysical form describing homogeneous polygons and pro- files with a vertical resolution of 0.1 m NLfB, principles at the larger scale. By modelling a generalized non-N-deficient management scenario, 1991; Eckelmann and Oelkers, 1993. The soils of the catchment were grouped according to texture the general validity of the model’s inherent func- tion for water stress is assessed over a wide range classes and volumetric water contents at field capacity h fc . The groupings corresponded to of soil water availability. Specifically, this paper 241 G.M. Richter European Journal of Agronomy 11 1999 239–253 humic podsols mfS; h fc 0.20 in the southern part and subtypes of luvisols lS-sL; h fc 0.20 in the central and northern part of the catchment. The depth of soil accessible to plants ranged from 0.4 to 0.7 m, depending on the presence of imped- ing layers and the maximum rooting depth in sand R dmax =0.7 m. Plant available water PAW in the potential root zone ranged typically from 46 to 119 mm; in areas of deeply ploughed low-moor soils, where the organic matter content is large 8– 15 , PAW reached 150 mm. Soil and land-use maps were overlaid to identify and characterize ecotopes, giving a total of 461 polygons in the 500 ha. During the 5 years, a total of 403 simulation cases were covered by winter rye, which corresponded to a total area of 450 ha. The size of the ecotopes ranged from 0.2 to 7.6 ha, log-normally distributed [Fig. 1a], with a mean of 1.25 ha. The 76 fields with recorded yields had a mean size of 3.6 ±2 ha range of 1–7.5 ha. For comparison of the samples of modelled vs. observed yields, PAW and yield of the fields were calculated as area-weighted averages. The field specific coefficient of variation CV for PAW ranged from 5 to 50 across the catchment. For all ecotopes evaluated for winter rye in the catch- ment during the 5 year rotation, the distribution of PAW was skewed towards smaller values Fig. 1. Frequency distributions of ecotope properties simulated [Fig. 1b]. The average PAW for all ecotopes was growth of winter rye in Eisenbach catchment 1988–1992 after 83 ±26 mm but occurred only rarely in the catch- intersection of soil and land use maps: a ecotope size with a ment. The fitted distribution was bimodal, reflect- class width of 0.25 ha, and b plant available water PAW, mm. Note log-scale for a. ing the two different soil types, which have mean values and standard deviation for PAW of 63 ±9 and 109 ±16 mm. The soil organic matter SOM The frequency distributions of PAW and initial N r showed that the texture classification scheme is content in the cultivated horizon A p ranged from 1 in the recently cultivated forest soils located more differentiated than that for organic matter Fig. 2. The actual distributions of PAW and N r in the southern part to values up to 15 in drained and deep-ploughed low bog soils in the central in the sub-sample of simulated winter rye fields appear when the parameters are weighted for the part of the catchment. Converting SOM to carbon 11.72 and assuming a C:N ratio of 17, a value spatial size of individual ecotopes: 60 of the modelled area had a PAW below 80 mm, whereas typical of naturally acid, ameliorated soils McVoy et al., 1995, the amount of potentially mineraliz- in the remaining 40 , it was more widely spread around 100 mm [Fig. 2a]. After the SOM content able soil organic N N r was set at 13 of the total N Kersebaum and Richter, 1991. It ranged from had been initialized at the mean class values, 45 of the area had an initial amount of potentially 150 to 900 kg Nha in the A p 0.2–0.3 m. A total of 13 ha 3 of the area contained N r contents mineralizable N of approximately 600 kgha [Fig. 2b]. For about 10 , SOM was classified as greater than 1000 kg Nha. The parameters PAW and N r were positively related r 2=0.25, p0.001. extremely low h1; 300 kg Nha and far less 242 G.M. Richter European Journal of Agronomy 11 1999 239–253 30 were sown to winter rye each year, totalling about 360 ha from 1988 to 1991. Yields were recorded for total fields by the farmers and made available in a database for approximately 55 of this area; there were few 10 yield records for 1992 90 ha. Farmers’ records also revealed irriga- tion quantities and frequencies, ranging from 40 to 250 mm for the rye crop. The annual average irrigation quantities were 64, 86, 91 and 85 mm for the years 1988–1991. For the simulation, the crop management was generalized for all years. The sowing date of winter rye was set to 15 October and the date of harvest to 17 August. N fertilization was assumed to be constant at 155 kg Nha given in three rates 45, 45 and 65 kg Nha. Further information on the land use can be found in an earlier paper Richter et al., 1998. 2.3. Weather conditions Daily data for temperature, sunshine, precipita- tion, vapour pressure deficit and potential evapora- tion calculated according to Thompson et al. 1981 were taken from a weather station in the catchment. Table 1 summarizes the most important climatic driving variables accumulated during three different phases of plant development: 1 vegeta- a b tive growth from sowing to double ridge EC 25, Fig. 2. a Distribution of plant available water PAW, mm and 2 early reproductive growth until the beginning b potentially mineralizable organic N N r , kg Nha in the of anthesis EC 61, and 3 reproductive growth cultivation horizon A p across the catchment, based on the cumulative relative area of simulated ecotopes. from anthesis until yellow ripeness EC 87. The distributions of rainfall, temperature and radiation than 10 as extremely high h5; 1000 kg Nha during the growth periods were distinctly different. in N r . During the simulation, the initially discrete Winter 19871988 was wet with a very low rainfall distribution of N r changed into a continuous distri- in the following spring. In 19881989, rainfall bution, which represented the actual N r before barely refilled the soil water storage capacity sowing of winter rye. This change was caused by during the winter and continued to be very scarce the N returned to the soil according to the parti- up to harvest. In the third year 19891990, the tioning of nitrogen in the crop residues and fertiliz- period up to anthesis was by far the warmest and ers. For details on the fractions of decomposable sunniest of all years, and this continued until and recalcitrant nitrogen, see Richter et al. 1998. harvest. The following 2 years were much cooler during vegetative growth, though in 19911992, 2.2. Land use and management the cumulative temperature during grain filling was the highest of the 5 years, corresponding to Land-use maps showed that between 40 ha 1991 and 150 ha 1990 of the catchment 8– an average daily temperature of almost 20 °C. 243 G.M. Richter European Journal of Agronomy 11 1999 239–253 Table 1 Cumulative mean air temperature T air , potential evapotranspiration ET , precipitation Rain, sunshine hours SS and global radiation Rad during the different phases of crop development, sowing to double ridge ECa 25, beginning of anthesis EC 61 and yellow ripeness EC 87 Year EC a T air °C ET mm Rain mm SS h Rad MJm 2 19871988 0–25 757 34 b 343 242 518 25–61 743 188 33 476 1064 61–87 1099 165 171 381 1053 Total 2599 387 547 1100 2636 19881989 0–25 710 22 § 189 284 456 25–61 858 135 70 598 1301 61–87 1136 196 92 522 1255 Total 2704 353 352 1403 3012 19891990 0–25 783 78 228 384 505 25–61 930 201 96 634 1317 61–87 1122 207 88 453 1158 Total 2855 486 411 1470 2980 19901991 0–25 671 110 207 450 728 25–61 735 163 77 460 1099 61–87 953 164 130 390 976 Total 2359 437 413 1301 2802 19911992 0–25 667 79 238 378 591 25–61 691 172 93 380 1044 61–87 1277 262 88 546 1363 Total 2635 513 419 1304 2998 a Eucarpia decimal scale for plant development Zadoks et al., 1974. b Limited measurements due to methodological reasons frost. 2.4. Modelling crop growth module describes photosynthesis, dry matter pro- duction and partitioning van Keulen et al., 1982. Plant phenology is a function of the sum of The N dynamics model MINERVA used to simulate rye growth and yield at the catchment biologically effective temperature analogous to ARCWHEAT1 Weir et al., 1984. The external scale followed the concept of Kersebaum 1995. As submodels, it contains: 1 a water balance and driving forces for the assimilation and development process are mean global radiation and air temper- water flux model based on the simple field capacity concept, 2 a model for mineralization of two ature. Radiation was derived from daily sunshine duration using a standard method based on fractions of N in the crop residues and SOM using first-order decay functions dependent on water daylength. Photosynthetic active radiation PAR is assumed to be 50 of the global radiation, and content and temperature, 3 a nitrate transport model describing mass flux and diffusion, and 4 light penetration into the stand is a function of leaf area index, LAI Goudriaan, 1977. The rela- a crop growth and N uptake model based on the explicit simulation of plant phenology and dry tionship between photosynthetic efficiency and mean air temperature developed by Groot 1987 matter production. The latter is described in detail here, including its interaction with climatic and for winter wheat was adapted to account for better growth of winter rye at low temperatures soil variables. For details of the soil processes, see the earlier descriptions of the model Kersebaum, Kavanagh, 1989. The parameters of winter rye development were adapted from the literature 1995; Richter et al., 1998 and GIS interface Beblik, 1996. DLG, 1987; van Dobben 1979 and verified earlier Richter, 1996. The sums of biologically effective The crop growth model used was a synthesis of a number of modelling concepts. A SUCROS mean air temperatures delineating the critical 244 G.M. Richter European Journal of Agronomy 11 1999 239–253 Table 2 Cumulative biologically effective temperature °C days with respect to emergence and base temperature Base, °C for winter rye to reach certain development stages EC ; adapted from the handbook of the German Agricultural Society DLG, 1987 1. Leaf Double ridge 1. Node Begin of anthesis Yellow ripe Dead ripe EC 11 25 31 61 87 92 °C days −80 250 370 700 1200 1620 Base °C 1 1 1 1 −7 −10 stages of rye phenology are presented in Table 2. 2.5. Statistical analysis The temperature requirement of rye for reaching Outliers were eliminated Sachs, 1980, p. 209 anthesis is 30 less than for winter wheat van on the basis of extreme SOM content and poorly Dobben, 1979, and the base temperature during founded relationships between yield and high soil grain filling and maturation is also less 7 vs. 9 °C . N mineralization rate. Simulation results of total The assimilate allocation rates into the compart- dry matter production, yield and N uptake were ments of root, leaf, stem and ear are reported weighted according to relative area. Weighted field elsewhere Richter, 1996. means were compared with farmers’ yield records, The modelled crop growth rate is limited by regional statistics and other field trials during water and nitrogen availability. Water fluxes from 1988–1992. The performance of the model was the soil–plant system are determined by rainfall evaluated statistically using the root mean square and potential evapotranspiration ET p . ET is error RMSE for modelled versus observed yields partitioned into evaporation and transpiration [Eq. 1 using a programme by Smith et al. 1996: according to surface cover LAI and reduced to actual rates by functions of relative water content in the soil. The ratio of actual to potential transpi- RMSE = 100 Y 9 obs S ∑ i=1 n Y MOD −Y OBS 2n. 1 ration determines the reduction in growth rate Groot, 1987. Water uptake by the plant is deter- The overall mean yield simulated for the catch- mined by the rooting depth and root density, both ment was calculated after temporal scaling of the changing with time, and the root efficiency, which distributed simulated yields using annual correc- varies with relative water content in the respective tion factors CF representing the ratio of mean soil layers. These relationships were taken observed versus mean simulated yields unchanged from studies on winter wheat Groot, Y OBS Y MOD . 1987; Whitmore and Addiscott, 1987. Nitrogen uptake is modelled as the process of convective transpiration and diffusive flux when the N

3. Results