Simulations and validations of the soybean yields per unit area using DSSAT crop model in the major soybean producing areas of China and America

Crop models have been widely studied for simulating crop growth and yield production at the field scale. However to upscale simulate crop yield from the field scale to regional scale is a key component for monitoring crop growth over large areas, especially in global main agricultural regions. In this study, a DSSAT-SOYGRO model coupled with remote sensing data was utilized to simulate soybean yield in Jilin Province, China and Iowa State, the U.S., using the meteorological data during the growth period at 0.5°×0.5° resolution, and Green Chlorophyll Vegetation Index (GCVI) at 500 m×500 m resolution. The specific procedure was listed: 1) The parameters of soybean cultivars were determined in these two research regions, according to their historical meteorological data (total rainfall in June-August, mean solar radiation in June-August, and the average daytime maximum temperature in August), recorded growth stages (planted, emerged, blooming, setting pods, dropping leaves, harvested date), application rate of nitrogen fertilizer (50-300 kg/hm2), and yields on site. 2) Random simulations were conducted for daily Leaf Area Index (LAI) spanning a range of sites, years, and nitrogen fertilizer application. The generated data was used to train a multi-linear regression model, according to the soybean cultivars, where stored data results in a coefficients table for later use. 3) The regression equations were applied to estimate soybean yields during 2008-2017, based on the actual measured data from remote sensing data, with the spatial resolution of 500 m × 500 m. The estimated soybean yields were also compared with the survey statistics. The results showed that: in Iowa State, the U.S., the soybean yields were 2 139-4 766 kg/hm2, showing a larger range than the survey yield with 3 002-3 991 kg/hm2, where the Mean Percentage Error (MPE) was 16.8%, Root Mean Square Error (RMSE) was 762.8 kg/hm2, and Mean Bias Error (BSE) was 107.2 kg/hm2; whereas, in Jilin Province, China, the soybean yields were 1 653-2 766 kg/hm2, which was closed to the survey yield with 1 997-2 797 kg/hm2, where the value of MPE was 36.3%, RMSE was 1 088.4 kg/hm2, and BSE was -237.9 kg/hm2. In addition, the inter-year variation patterns from 2008 to 2017 year showed well consistent trends between the estimated yields and survey data of soybeans. In Iowa State, the lowest survey yields of soybean occurred at the year of 2012, similar to the estimated data, while the survey soybean yields were increased since 2012, but decreased since 2016, indicating a good agreement with the estimated yields. In Jilin Province, the increase and decrease trends were also in good agreement with the estimated crop model. At the county scale, the correlation between simulated soybean yields and survey data revealed the higher accuracy in Iowa State compared with Jilin Province, where the correlation coefficient was 0.78 in Iowa State and was 0.59 in Jilin Province. Moreover, there was a higher correlation coefficient in Iowa State, when soybean yields were relatively low. This finding indicates that the yield estimation of soybean over large areas can be achieved in the main agricultural regions of China and the U.S., using the DSSAT crop model combining with remote sensing data, including the meteorological elements and vegetation index. The high-resolution satellite images can be supposed to consider some complicated environment variable information, including the irrigation management, planting case databases and analyses, in further study to improve the accuracy of yield simulation.