Detecting Spatial Variability of Paddy Rice Yield by Combining the DNDC Model with High Resolution Satellite Images

Yield estimation over large areas is critical for ensuring food security, guiding agronomical management, and designing national and international food trade strategies. Besides, analyzing the impacts of managed cropping systems on the environment is important for sustainable agriculture. In this study, the agro-ecosystem model DNDC (DeNitrification-DeComposition) and FORMOSAT-2 (FS-2) satellite imagery were used to detect spatial variabilities of paddy rice yield in the Qixing Farm in 2009. The Qixing Farm is located at the center of the Sanjiang Plain in north-east China, which is one of the important national food bases of China. The site-specific mode of the DNDC model was adapted due to its advantages of better transferability and flexibility. It was generalized onto a regional scale by programming a set of scripts using the Python programming language. Soil data were prepared as model inputs in 100m raster files. The spatial variabilities in modelled yields were well detected based on the detailed soil data and an accurate rice area map. Rice yield was also derived from multiple vegetation indices based on the FS-2 imagery. The DNDC model integrates environmental factors and predicts yield depending on all model input data, whereas the RS method mainly considers in-season crop information. Based on the vegetation indices, the RS-derived yield represents a response to the environmental factors and human activities which may exceed the DNDC capability. It was found that the highest coefficient of model determination (CD) and index of agreement (IA) for the modelled yield were 2.63 and 0.74, respectively, while for the RS-derived yield, the highest CD and IA were 1.2 and 0.55, respectively. Results from both methods were comparable and each method has its own advantages.