Temperature-based solar radiation models for use in simulated soybean potential yield

Abstract

Solar radiation is the main meteorological element required for crop yield simulation. However, it is not widely measured as air temperature and rainfall. This study evaluated some temperature-based solar radiation models for estimation of daily solar radiation (Rs), and how the estimates may affect soybean yield potential. The evaluated models were Annandale (AN), Hargreaves (HA), Modified Hargreaves (HA-1), Hunt (HU), Bristow and Campbell (BC), Chen (CH), Donatelli and Campbell (DC) and De Jong and Stewart (JS). This research was carried out using historical data from six sites in the Triangulo Mineiro region, where measured values of Rs, minimum and maximum air temperature and rainfall were available. The dataset (2009-2014) was separated into two sub-data sets, one for calibration (2009) and the other for evaluation of performance (2010-2014). The Rs estimated data were used in SoySim software to estimate potential soybean yield. Statistical indexes: (a) root mean square error (RMSE), (b) relative root mean square error (RRMSE), (c) coefficient of determination (R2) and (d) mean error (ME) were used as indicators of the agreement between observed and estimated Rs data. After evaluating the performance, Rs estimated values for each model were used to simulate the soybean potential yield. Although the eight models have presented similar performance for estimating Rs values, when these data were used for simulation of the potential soybean yield, the performances diverged considerably. In this way, only the BC, CH, DC and JS models showed satisfactory performance in yield simulation with R2 and RRMSE varying from 0.76 to 0.80 and 3 to 4%, respectively. Therefore, the findings suggest that, before choosing the model to estimate Rs, it is important to define the purpose of use of solar radiation data.