Introduction Material and methods

38 S. Asseng et al. European Journal of Agronomy 12 2000 37–54

1. Introduction

field measurements in different climates Keating et al., 1995; Probert et al., 1995, 1998; Asseng Optimising economic returns in high N input et al., 1998a,b; Meinke et al., 1998 and have been systems, while minimising loss of N to the environ- used in agronomic studies of farming system sustai- ment, is one of the challenges for modern agricul- nability in the eastern states of Australia Keating tural production Whitmore and Van Noordwijk, et al., 1995; Probert et al., 1995, 1998 and Western 1995. Field studies have been conducted to optim- Australia Asseng et al., 1998a. ise N fertiliser in wheat production, but the results This paper aims 1 to evaluate the performance are always season-specific e.g. Spiertz and Ellen, of the APSIM Nwheat model against a wide range 1978; Spiertz and Van de Haar, 1978; Ellen and of detailed field measurements in the temperate Spiertz, 1980; Chaney, 1990; Darwinkel, 1998. maritime climate of the Netherlands, and 2 to Crop models have been used to optimise manage- use the model in a simulation experiment to study ment practices under variable environments Van interactions between N-fertiliser application and Keulen and Seligman, 1987; Stapper and Harris, grain yield, grain protein and soil residual N. 1989; Keating et al., 1991; Meinke et al., 1993; Savin et al., 1995; Thornton et al., 1995; Asseng et al., 1998a and application of a simulation

2. Material and methods

model may be useful in extrapolating results from such experiments. However, such simulation 2.1. APSIM Nwheat model models require rigorous testing against a wide range of field experimental data before they can APSIM is a software tool that enables modules be used with confidence Monteith, 1996; O’Leary or sub-models to be linked to simulate agricul- and Connor, 1996. Groot and Verberne 1991 tural systems McCown et al., 1996. Examples of have published a detailed data set, covering modules include crops, pastures, soil water, nitro- different sites, seasons and N regimes in the gen and erosion. Four modules, i.e. wheat crop Netherlands to validate simulation models for NWHEAT , soil water SOILWAT , soil N nitrogen dynamics in crop and soil. Parts of these SOILN and crop residue RESIDUE are most data have been used to validate models, with relevant to the simulation of wheat-based cropping various success De Willigen, 1991. Whereas most systems. These modules have evolved from experi- of these models dealt with the above-ground pro- ences in Australia with the CERES crop and soil cesses more or less adequately, simulation of models Ritchie et al., 1985; Jones and Kiniry, below-ground processes was more problematic. In 1986, and the PERFECT model Littleboy et al., particular, processes of N mineralisation and 1992, as modified by Probert et al. 1995, 1998. immobilisation, which are related to soil microbio- The main differences of the APSIM Nwheat model logical activities, were poorly simulated De to the CERES wheat model are summarised by Willigen, 1991. Probert et al. 1995 and Asseng et al. 1998b. The APSIM Agricultural Production Systems Documented model source code in hypertext Simulator Nwheat model comprises crop growth, format can be obtained by writing to Dr B.A. soil water, nitrogen and crop residue modules Keating CSIRO Division of Tropical Agriculture, McCown et al., 1996 and enables simulation of Brisbane, Australia, e-mail: brian.keatingc- their interactions. The wheat crop, soil water and netns.tcp.csiro.au or can be viewed at www.apsim- soil nitrogen modules Keating et al., 1995; Probert help.tag.csiro.au. et al., 1995, 1998 have been derived from the CERES models of Ritchie et al. 1985 and Jones 2.2. Experimental data and Kiniry 1986, which have been widely tested and applied e.g. Ritchie et al., 1985; Otter-Nacke The APSIM Nwheat model with modules et al., 1986; Savin et al., 1994, 1995; Thornton et al., NWHEAT, SOILWAT, SOILN and RESIDUE in 1995; Toure´ et al., 1995. Results of the APSIM Nwheat model have been compared with various the APSIM framework was tested using field meas- 39 S. Asseng et al. European Journal of Agronomy 12 2000 37–54 Table 1 Experimental sites used for simulation — observation comparisons Groot and Verberne, 1991 Location Latitude °N Longitude °W Soil type Rainfall mm The East 52.4 5.4 Silty loam 646 a PAGV 52.3 5.3 Silty loam 646 a The Bouwing 51.6 5.4 Silty clay loam 763 b a Average annual rainfall between 1974 and 1988. b Average annual rainfall between 1954 and 1996. urements from 18 treatments at three locations N treatments are listed in Table 2. These data sets include frequently measured and two seasons, published by Groot and Verberne 1991. These data were used for a statistical shoot biomass, shoot N content, leaf area, grain yield and its components, grain N, phenological analysis of model performance. An additional qualitative test, involving the comparison of simu- development stages, changes in soil water, ground- water table depth and soil N and are described in lated with measured data, was carried out with 1 grain yields from a long-term winter wheat detail by Groot and Verberne 1991. To account for a continuous wet and dry depos- field experiment at Lovinkhoeve K.B. Zwart, AB-DLO, pers. comm., 2 yield data from ition of N in the Netherlands of about 35–50 kg N ha−1 year−1 Neeteson and Hassink, experiments by Spiertz and Ellen 1978 and Spiertz and Van de Haar 1978 and 3 grain 1997, a monthly N deposition of 3 kg N ha−1 of ammonium nitrate for sites at the Eest and PAGV yields from the Gelderland region Crop Estimates, Statistics Netherlands, Voorburg. All grain yields and 4 kg N ha−1 for the Bouwing was assumed for all model simulations. refer to dry matter weight, and all grain protein contents are based on grain dry weight. The soil characteristics are listed in Tables 3a and 3b. Simulations were initialised using the first soil water and soil N measurements of each season 2.2.1. Field measurements after Groot and Verberne 1991 Groot and Verberne, 1991. Hence, all the first soil measurements were excluded from the statisti- Measured data sets from field experiments from two Polder sites, a silty loam at the Eest and at cal performance test. Estimated quantities of residues used to ini- PAGV near Swifterbant, and from a third site, a silty clay loam at the Bouwing near Wageningen tialise the model are presented in Table 4. SOILN module parameters for the effects of the Netherlands, were used for model validation. A site description is given in Table 1. soil water content on the processes of mineralisa- Table 2 N-fertiliser treatments N1, N2 and N3 after Groot and Verberne 1991 a Location Year N1 kg N ha −1 N2 kg N ha −1 N3 kg N ha −1 The East 1983 60 120+40 1984 50+60 50+60+40 50+60+40 PAGV 1983 80 60+80 60+140+40 1984 80 80+60+40 80+120+40 The Bouwing 1983 60 120+40 1984 70 70+60+40 70+120+40 a Each number represents an application amount at a particular application time. The first application occurred in February, the second application between tillering and the beginning of stem elongation, and the third application between flag leave stage and anthesis. 40 S. Asseng et al. European Journal of Agronomy 12 2000 37–54 Table 3a Soil characteristics for the Eest, PAGV and the Bouwing a Layer depth cm The Eest PAGV The Bouwing Silty loam Silty loam Silty clay loam LL m 3 m−3 DUL m 3 m−3 OC kg kg−1 OC kg kg−1 LL m 3 m−3 DUL m 3 m−3 OC kg kg−1 0–5 0.09 0.34 0.030 0.023 0.18 0.39 0.028 5–10 0.09 0.34 0.030 0.023 0.18 0.39 0.028 10–20 0.09 0.34 0.030 0.023 0.18 0.39 0.028 20–30 0.10 0.34 0.030 0.023 0.18 0.39 0.028 30–40 0.12 0.35 0.020 0.018 0.18 0.37 0.014 40–60 0.14 0.37 0.020 0.018 0.20 0.37 0.014 60–80 0.14 0.39 0.018 0.018 0.20 0.37 0.012 80–100 0.14 0.39 0.018 0.018 0.20 0.37 0.012 100–130 0.14 0.39 0.010 0.010 0.20 0.37 0.010 130–200 0.14 0.39 0.010 0.010 0.20 0.37 0.010 Total mm 264 757 392 746 a LL: lower limit for plant available soil water; DUL: drained upper limit; OC: soil organic carbon. Data after Groot and Verberne 1991. LL and DUL for PAGV are the same as for the Eest. Table 3b cients [after Ritchie et al. 1985] for Arminda are Relative soil water conductivity K s — proportion of water given in Table 5. above DUL flowing to the next deeper layer with a time step of 1 day and relative macro flow water conductivity K m — proportion of water in macro pores flowing to the next deeper 2.2.2. Grain yields from a long-term winter wheat layer with a time step of 1 day for the Eest, PAGV and the experiment at Lovinkhoeve Bouwing Simulation results were compared with grain yields from a long-term winter wheat experiment Layer depth The Eest, PAGV, The Eest, The cm the Bouwing PAGV Bouwing at Lovinkhoeve near the Eest site, same soil type as at the Eest with three N treatments 0, 38 and K m − K s − K s − 188 kg N ha−1 between 1974 and 1987 K.B. Zwart, AB-DLO, pers. comm.. The wheat variety 0–5 1 0.5 0.5 5–10 1 0.5 0.5 Arminda was sown in most years, except in 1974 10–20 1 0.5 0.5 when Clement, and 1975, 1977 when Lely was 20–30 1 0.5 0.5 sown. Since Arminda, Clement and Lely are rela- 30–40 1 0.5 0.5 tively similar in phenology and yield parameters, 40–60 1 0.4 0.5 the genetic coefficients of Arminda were used for 60–80 1 0.3 0.5 80–100 1 0.1 0.5 all the simulations. In every fourth year, no yield 100–130 1 0.1 0.3 data were available from the long-term experiment 130–200 0.6 0.001 0.001 due to a specific rotation K.B. Zwart, AB-DLO, pers. comm.. For two years, in 1976 and 1977, additional yield measurements from N-fertiliser experiments tion and nitrification, and the potential decomposi- tion rates for soil organic matter 0.00015 day−1 at the same location by Spiertz and Ellen 1978 and Spiertz and Van de Haar 1978 were included and microbial biomass 0.0081 day−1 were used as suggested by Probert et al. 1998 for clay soils in the comparison. Each year was re-initialised with the soil water and sowing conditions from in Queensland, Australia. The sown variety was Arminda. Genetic coeffi- the Eest in 1982 Groot and Verberne, 1991; 41 S. Asseng et al. European Journal of Agronomy 12 2000 37–54 Table 4 Initial plant residues for 17 October 1982 and 1983, for the Eest, PAGV and the Bouwing Location The Eest PAGV The Bouwing 1982, 1983 1982 1983 1982, 1983 Above ground Amount t ha−1 4.0 4.0 2.0 4.0 C:N ratio 50 50 50 50 Relative potential decomposition rate d−1 0.05 0.05 0.05 0.05 Root residues Amount t ha−1 1.5 1.5 1.0 1.5 C:N ratio 50 50 50 50 Residue type Potatoes Sugarbeets Sugarbeets Potatoes Table 5 Wheat genotype coefficients for cv. Arminda Arminda Coefficient Explanation 4.0 p1v Sensitivity to vernalisation [1 lowest–5 highest] 4.0 p1d Sensitivity to photoperiod [1 lowest–5 highest] 640 p5 Thermal time base 0°C from beginning of grain filling to maturity °Cd 32 Grno Coefficient of kernel number per stem weight at the beginning of grain filling [kernels g stem−1] 2.5 Fillrate Potential kernel growth rate [mg kernel−1 day−1] 3.0 stwt Potential final dry weight of a single stem, excluding grain g stem−1 100 phint Phyllochron interval Table 4; Section 2.2.1. A low mineral N content 4 kg N ha−1 of ammonium nitrate was included Neeteson and Hassink, 1997. Since none of the at sowing 5 kg N ha−1 was assumed for each year, and a monthly N deposition of 3 kg N ha−1 initial conditions or proportions of soil types were known for the wheat growing area in the of ammonium nitrate was included Neeteson and Hassink, 1997. Gelderland region, initial soil and sowing condi- tions from the Bouwing 1982 Groot and Verberne, These comparisons between simulated and observed yields allowed another qualitative model 1991; Table 4; Section 2.2.1 were used as the nearest approximation. This allowed an additional test of N, season and growth interactions. However, caution should be paid to the interpreta- qualitative model test of season–yield interactions over a long number of seasons, but as mentioned tion of model performance with estimated initial conditions, since initial conditions could have been in Section 2.2.2, test results will be limited by the estimation of initial conditions, but also by very different in some years and might not reflect the true model performance. different earlier farming practices. 2.3. Weather data 2.2.3. Grain yields from the Gelderland region A simulation was carried out with 200 kg N ha−1 using Wageningen weather data Daily data for solar radiation, maximum and minimum temperature and rainfall from from 1975 to 1996. Simulation results were com- pared with grain yields from winter wheat yields Swifterbant from 1974 to 1988 near the Eest and PAGV and Wageningen from 1975 to 1996 near of the Gelderland region between 1975 and 1996 Crop Estimates, Statistics Netherlands, the Bouwing were made available by the Department of Meteorology of Wageningen Voorburg. A monthly N deposition of 42 S. Asseng et al. European Journal of Agronomy 12 2000 37–54 Agricultural University. Fig. 1 presents summar- 2.5. Simulation experiments ised average monthly weather data for A simulation experiment was carried out to Wageningen. study the effect of rate and timing of N-fertiliser applications on predicted grain yield, grain protein and soil residual N for a silty loam at the Eest, 2.4. Quantification of model performance using weather data from the nearest weather sta- tion at Swifterbant, 1974–1988. Six N-fertiliser Model performance was quantified for data treatments A, B, C, D, E and F were applied at from the Eest, PAGV and the Bouwing Groot different growth stages described by a decimal and Verberne, 1991. Since there is no single means code DC Zadoks et al., 1974 as shown in to quantify the performance of a model, a range Table 6. of performance indicators was calculated. To assess The initial, crop residue and sowing conditions model performance, a correlation analysis was for this simulation experiment were identical to carried out, and the coefficient of determination the Eest 19821983 simulations Groot and for the 1 to 1 or y=x line [r 2 1:1, which is NOT Verberne, 1991; Table 4; Section 2.2.1, except for for the fitted regression line] was computed, which the initial soil mineral N profile. Simulations were measures the true deviation of the estimates from initialised before each simulation with the relative observations. The slope m is presented to quan- low soil N profile from PAGV in 1982 Groot and tify a possible over- or underestimation, calculated Verberne, 1991; Section 2.2.1 to avoid any addi- from a best-fit regression line forced through the tional N effects from residual soil mineral N of a origin. Note that a traditional regression analysis previous year. was carried out only to show the slope, but not to A second simulation experiment was carried out produce a coefficient of determination for a fitted to study the effect of zero 0 and maximum max line, which is not relevant to a model validation N-fertiliser applications Table 6 on predicted Mitchell, 1997. Another indicator of model per- grain yield, grain protein and soil residual N. formance, which also represents the true deviation Simulations were done for a silty loam at the Eest of the estimates from observations, the root mean using weather data from the nearest weather sta- square deviation RMSD, i.e. the root of the mean squared error of prediction MSEP after Table 6 Wallach and Goffinet 1989, was computed to Simulation experiment treatments with different N provide a measure of the absolute magnitude of applications a the error. Treatment N application kg N ha−1 at growth stage: February DC23 DC31 DC39 A 140–N min B 140–N min 45 45 C 140–N min 45 45 20 D 140–N min 45 45 40 E 140–N min 45 45 60 F 140–N min 45 45 80 Max 220–N min 65 65 60 a Treatments A–F received a N application based on Fig. 1. Average monthly solar radiation ———, maximum 140 kg N ha−1 minus soil mineral N Nmin of the soil profile 0–100 cm on 17 February of each year. Treatment maximum – – – – and minimum · · · · · · temperature and rainfall entire bars based on 31 years of weather data from max N received a N application based on 220 kg N ha−1 minus soil mineral N of the soil profile 0–100 cm on 17 February Wageningen, the Netherlands. The dark lower parts of the bars show the minimum recorded amount of monthly rainfall. of each year. 43 S. Asseng et al. European Journal of Agronomy 12 2000 37–54 Table 7 Summary of the APSIM Nwheat model performance at the Eest, at PAGV, and the Bouwing in the Netherlands Model attribute Number of paired data points Observed range r 2 1:1a m b RMSD c Grain d yield t ha−1 63 0.4–8.3 0.90 0.96 0.8 Kernels d m−2 27 12726–27916 0.55 0.95 2604 Kernel d weight mg 27 9–37 0.79 1.04 3.9 Anthesis date day 6 14 0.76 0.98 3.7 Biomass t ha−1 129 0.03–20 0.97 1.02 1.2 LAI m m−2 126 0–5.5 0.65 1.28 1.2 Crop N kg ha−1 128 2–276 0.82 1.03 28.5 Grain d N kg ha−1 63 10–204 0.88 0.98 22.4 Grain protein 18 7.1–15.6 0.59 1.00 1.6 Soil mineral N kg ha−1 546 0–78 0.46 0.75 9 a r2 1:1=r2 for the 1 to 1 line y=x. b Slope of linear regression forced through the origin. c Root mean square deviation. d Including pre-maturity harvests. tion at Swifterbant, 1974–1988 and a silty clay been summarised in Table 7. The observed grain yields ranged from 4.5 t ha−1 or when including loam at the Bouwing, using weather data from the nearest weather station at Wageningen, 1955–1996. pre-maturity harvests from as low as 0.4 t ha−1 to 8.3 t ha−1 Fig. 2. The yield simulations cap- tured the response to season and N-fertiliser effects with a coefficient of determination of r2 1:1 of

3. Results