Downscaling the APSIM crop model for simulation at the within-field scale

Abstract

CONTEXTMost crop models are designed for point-based modeling and to simulate agronomic variables on their native spatial footprint, i.e. typically as a uniform field-scale value. Precision agriculture needs crop model simulations at sub-field scales to support differential management application. Spatialization processes are used to change the simulation scale of crop models. OBJECTIVEThe objective of this study is to investigate the spatialization of a complex crop model by using a spatial calibration approach to modify its native spatial footprint and to evaluate if it is relevant to use this kind of crop model at the within-field scale. METHODSAPSIM was spatialized to simulate durum wheat yield at different spatial scales (field, within-field and site-scale) on an experimental field under Mediterranean conditions in southern Italy. Ancillary soil data were used to derive potential management (modeling) zones at different scales, which were then used to spatially calibrate soil and biomass parameters in APSIM to spatially predict yield in two different production years (one year was used for calibration and the other for evaluation). Spatialized crop model performances were evaluated using the spatial balanced accuracy (SBA) score, a metric to evaluate the global preservation of patterns between maps. RESULTS AND CONCLUSIONSThe spatial structure of the yield data influenced the effectiveness of the spatial calibration process. When the agronomic variable (durum wheat yield) was spatially structured, a spatialized APSIM approached performed best (5-zone modeling scale, SBA = 0.17) and outperformed the field-scale (native footprint) model (SBA = 0.19). In contrast, when the target agronomic variable was more random (less spatially structured), the uniform field-scale modeling performed best and spatial calibration had no benefit. The spatialized APSIM performances were mainly based on the reliability of the delineated zones that undeniably affected the quality of the spatialized model outputs. Thus, more research is needed on how best to model scale-dependent processes to have more reliable modeling at the within-field scale. SIGNIFICANCEBased on the example of a complex crop model like APSIM, this study showed that spatial calibration can be effective and has a role to play in the spatialization of complex crop models.