Spatial Prediction of Soil Moisture Content in Winter Wheat Based on Machine Learning Model

Soil moisture is one of the important factors affecting the growth of crops. Accurate monitoring and forecasting of soil moisture in the growth period of crops is an important part of agricultural production. In this study, 15 predictors from three aspects of meteorology, topography and soil properties in Baoji were selected to establish the machine learning model to predict the soil moisture in 0 similar to 20cm and 20 similar to 40cm soil layers. The prediction of same data was carried out by three models, which were support vector machine (SVM), random forest (RF) and back-propagation neural network (BPNN). The results showed that the prediction accuracy of SVM were 92.899% and 92.656% in 0 similar to 20cm and 20 similar to 40cm soil layers, the RMSE were 7.521 and 8.011 respectively, while the RF were 87.632% and 87.842% in prediction accuracy, 10.759 and 11.042 in RMSE, and the prediction accuracy of BPNN were 80.570% and 85.323%, the RMSE were 12.147 and 11.165. The study found that the three models have good prediction effect on winter wheat soil moisture in Baoji, reflecting the good application ability of machine learning model in soil moisture prediction. And the prediction accuracy of three models in 0 similar to 20cm soil layer were slightly better than that in 20 similar to 40cm. Compared with the model of RF and BP, SVM has better prediction results. And the analysis of predictors showed that the meteorology has greatest impact on soil moisture and its changes, which the precipitation, air relative humidity and sunshine duration most; the effects of topography is relative, which the slope and elevation have great influence; soil property has little effect on the change of soil moisture, which the thickness of plough layers has slightly stronger influence than other factors.