Meta analysis on the evaluation and application of DSSAT in South Asia and China: Recent studies and the way forward

Abstract

The Decision Support System for Agrotechnology transfer (DSSAT) is a global modelling platform that encompasses crop models for more than 40 different crops. The models have been used extensively throughout the world, including South Asia and China. From the web of science database, we reviewed 205 papers that were published from January 2010 to February 2022 containing examples of the evaluation and application of the DSSAT crop simulation models. In South Asia and China, more than 50 traits and variables were analyzed for various experiments and environmental conditions during this period. The performance of the models was evaluated by comparing the simulated data with the observed data through different statistical parameters. Over the years and across different locations, the DSSAT crop models simulated phenology, growth, yield, and input efficiencies reasonably well with a high coefficient of determination (R2), and Willmott d-index, together with a low root mean square error (RMSE), normalized RMSE (RMSEn), mean error (ME) or percentage error difference. The CERES models for rice, wheat and maize were the most used models, followed by the CROPGRO models for cotton and soybean. Grain yield, anthesis and maturity dates, above ground biomass, and leaf area index were the variables that were evaluated most frequently for the different crop models. The meta-analysis of the data of the most common simulated variables (Anthesis, maturity, leaf area index, grain yield and above ground biomass) for the four commonly used DSSAT models (CERES-Rice, CERES-Wheat, CERES-Maize and CROPGRO-Cotton) showed that the models predicted anthesis with an RMSE of ~2 (CERES-Maize) and -4 days (CERES-Wheat), a normalized RMSE of ~2.5 (CERES-Maize) and -3.8% (CERES-Rice), and a R2 ~ 0.98-0.99. The maturity was predicted with an RMSE~ 3.0 (CERES-Maize)-6.1 days (CROPGRO-Cotton), normalized RMSE~2.3 (CERES-Wheat)-5.0% (CERES-Rice) and R2 ~ 0.90-0.99. The leaf area index was predicted with an RMSE~ 0.3-0.7, normalized RMSE~6 (CROPGRO-Cotton)-16% (CERES-Maize) and R2 ~ 0.75-0.98. The model performance for simulating grain yield was best with CROPGRO-cotton with a normalized RMSE of 4.4%, RMSE of 138.8 kg and R2 of 0.99.  The lowest R2 and highest RMSEn was found for CERES-Wheat. Among all the variables that were evaluated, above ground biomass was least accurately simulated with a RMSEn as high as 18% and R2 as small as 0.50 by CERES-Wheat.  The models were used for studying the crop response under various soil, weather, and management conditions. The review will be helpful to identify the research gap in the use of crop models for different crops in South Asia and China. It can also aid scientists to target their research for specific applications to address food and nutrition security based on sustainable management practices.