Model Adaptation to Local Conditions

The Decision Support System for Agrotechnology Transfer (DSSAT), with its suite of CERES- and CROPGRO-based crop models, was developed to help researchers understand crop responses to various management options, soils, and weather conditions (Tsuji et al., 1998; Jones et al., 2003). The CENTURY soil organic matter model, originally developed to simulate soil C dynamics in temperate grasslands (Parton et al., 1987), has since been used in a wide range of conditions including tropical systems (Paustian et al., 1992; Parton et al., 1988, 1994; Woomer, 1993; Anderson and Ingram, 1993; International Centre for Research in Agroforestry, 1994). Recently, the CENTURY model was linked with the DSSAT cropping system model to improve capability for simulating cropping systems with low inputs (Gijsman et al., 2002a; Jones et al., 2003). This linked DSSAT-CENTURY model was used in this study.

Answers are sought to the following questions: (1) Does the model adequately simulate growth and yield of the crops under the soils, climate, and management conditions being considered? (2) Does the model adequately simulate changes in soil processes (including SOC) under those same conditions? One can be relatively sure that existing models, even robust, widely used models like DSSAT, will not perform well in a new location unless an effort is made to adapt them to local conditions. Adapting the model requires: (1) assembly of local data on soil, weather, and crop performance under field conditions, (2) estimation of crop model parameters for local cultivars, (3) estimation of critical soil parameters not normally measured (particularly soil hydraulic properties), and (4) evaluation of model ability to simulate crop (e.g., phenological development, yield, and biomass) and soil (e.g., water, SOC, and N) responses under local conditions.

Although appropriate data for this area were limited (i.e., no long-term experiments with observed SOC changes), available data were used to simulate these preliminary estimates of SOC sequestration. In this study, continuous use of maize was assumed to demonstrate the approach. The first step was to adapt the model for simulating maize cultivars normally grown in Mali agronomic experiments. The second step was to adjust runoff characteristics for CT vs. RT so that published differences in runoff and crop yield between these two systems were correctly simulated. The final step was to adjust initial C fractions so that SOC under CT was at steady state. These procedures allowed us to confirm that the model correctly simulates absolute yield levels as well as differences between the two systems that are being compared.

16.2.1.1 Soil Data

Soil samples collected by Mamadou Doumbia and Russ Yost in March 2002 were used to develop necessary soil profile inputs to the model. A composite of soils sampled from the fields of Zan Diarra, (Lat 12.55 N, Long 6.47 W) and of Yaya Diassa (Lat 11.14 N, Long 5.35 W) was used to create a soil input file with parameters listed in Table 16.1. Soil water-

Table 16.1 Selected Soil Inputs in Zan Diarra Samples and Yaya Diassa Soils

Soil Wilting Field Bulk

Depth SOC Sand Silt Clay Pointa Capacityb Density

(cm) (%) (%) (%) (%) pH (cm3 cm-3) (cm3 cm-3) (g cm-3)

a Initial soil C was assumed to be 7016 kg ha-1 in the top 20 cm. b Calculated using the method described by Jagtap et al. (2004). Source: From M. Doumbia, personal communication, 2003.

holding characteristics were estimated from soil texture using the nearest neighbor method of Jagtap et al. (2004). SOC composition in the DSSAT-CENTURY model is initialized by partitioning total C into three pools based on rates of decomposition: microbial, slow, and stable, with default fractions for grassland and previously-cultivated soils of 02:64:34 and 02:54:44, respectively. Since we had no measurements that would allow us to estimate these fractions directly, we assumed that soil C under CT was at a steady state. Thus, we varied these fractions for CT simulations until we achieved a steady-state level of SOC. When fractions of 02:41:57 were used for Mali soil, climate, and CT management, SOC remained at 0.24% for the 10-year period of simulations (see results for CT in Figure 16.1).

16.2.1.2 Weather Data

Historical daily weather data are needed for simulating experiments conducted in the past and evaluating model predictions vs. observations. Observed daily weather data were obtained in order to compare simulated maize results with

Soil Organic Carbon 0-20 cm

Soil Organic Carbon 0-20 cm

Time, yrs

Figure 16.1 Plot of annual change in soil organic carbon (SOC) over 10 years under conventional tillage and fully implemented ridge tillage using initial SOC composition of calibrated stability (02:41:57).

Time, yrs

Figure 16.1 Plot of annual change in soil organic carbon (SOC) over 10 years under conventional tillage and fully implemented ridge tillage using initial SOC composition of calibrated stability (02:41:57).

Table 16.2 Calibration of Local Maize Variety Sotubaka: Three Years of Observed and Simulated Grain Yield and Days to Anthesis

Time to Anthesis (days) Maize Grain Yield (kg ha-1) 1999 2000 2001 1999 2000 2001

Simulated 62 61 57 5486 4138 5514 Observed 63 61 58 5100 3900 6070

Source: From Coulibaly, Ntji, personal communication, 2002.

those obtained by Coulibaly (i.e., Table 16.2). We also generated 10 years of daily weather data by interpolation between nearest existing weather stations using MarkSim, version 1 (P. Jones et al., 2002). The generated daily data include rainfall, maximum temperature, minimum temperature, and solar radiation. Small amounts of N (13 kg ha-1 100 cm-1 infiltrated rainfall) (Campbell, 1978; Pieri, 1992; Vitousek et al., 1997) were applied to all simulated crops according to infiltration of rainfall.

16.2.1.3 Agronomic Experiment Data

Agronomic yield trial data for a 3-year maize study were obtained from Njti Coulibaly in Mali, including soil, weather, and management of the crops in each year. That experiment was simulated using the DSSAT CERES-Maize model, and genetic coefficients were estimated using measured anthesis dates and yields for the 3 years (Jones et al., 2002a). Data in Table 16.2 demonstrate that the model describes anthesis dates and yields across the 3 years with errors less than 10%. Although additional tests are desirable, this exercise demonstrated that the model can simulate growth and yield responses to typical growing conditions in Mali under conventional management.

Detailed measurements from experiments comparing RT vs. CT were not available. Thus, a computer experiment was conducted over a 10-year period: (1) to adjust field runoff parameters for RT vs. CT, and (2) to compare predicted grain and biomass yield values for RT and CT with those responses reported by Gigou et al. (2000). Maize was planted each year in the computer experiment as soon as the soil in the top 30 cm of soil reached 60% of plant available water, but no earlier than June 18 to ensure adequate moisture for germination. Harvest was assumed to occur at maturity, which was simulated for each season. Plant density was set at three plants m-2 in all simulations, and row width was 75 cm. Crops simulated under CT had no manure or N fertilizer applications, and 90% of the crop residue was removed after harvest. Management of RT included application of inorganic N (40 kg ha1 applied in two doses), return of 90% of crop residue to the soil, and addition of 3 metric tons ha-1 of (dry) manure.

Runoff parameters for the RT field were set by assuming that the ridges were sufficiently constructed to reduce runoff by 45% relative to CT management (M. Doumbia, personal communication, 2003; Gigou et al., 2000). Runoff curve numbers of 90 and 96 were selected for RT and CT, respectively. Simulated runoff for the 10 years was 45% less for RT vs. CT management, and differences in yield and residue biomass between CT and RT closely matched differences reported by Gigou et al. (2000) (Table 16.3).

0 0

Post a comment