Application of Geo-CropSim framework for rainfed sugarcane yield assessment in Thailand

Abstract

The adoption of hybrid approaches that use remote sensing datasets in crop simulation models for yield assessment at spatial scale has been constrained by lack of crop type maps and high-resolution remote sensing datasets. To address that, Geo-CropSim, a geospatial crop simulation modeling framework, was developed to produce high quality remote sensing products and use them in Environmental Policy Integrated Climate (EPIC) model to estimate crop yields at various spatial scales. In this study, Geo-CropSim was implemented within the Chaiyaphum province, Thailand to estimate sugarcane yields under rainfed conditions for the 2018 at 500-m resolution. Results showed that the use of Harmonized Landsat Sentinel (HLS) data along with sentinel-1 SAR data allowed to produce a land cover map with an overall accuracy of 96%. Similarly, utilization of HLS data helped to estimate emergence dates and seasonal leaf area index capturing spatial differences driven by differences in climatic conditions. Geo-CropSim yield estimates ranged from 55-98 Mg/ha and showed a typical north-south gradient of decreased yields reflecting lower rainfall and soil quality in the south. Simulated yields showed good a correlation with district and field observed yields with RMSE values of 11.7 Mg/ha and 15.75 Mg/ha, respectively. Overall, simulated yields had uncertainty indicated by mean absolute percentage error (MAPE), less than 20% which is considered as acceptable accuracy for regional yield assessment. Even though it is required to test the framework's performance for long-term consistency over multiple regions, this study demonstrated Geo-CropSim's potential to produce sugarcane yields under rainfed conditions with reasonable accuracy.