An LSTM neural network for improving wheat yield estimates by integrating remote sensing data and meteorological data in the Guanzhong Plain, PR China

Crop growth condition and production play an important role in food management economic development. Therefore, estimating yield accurately timely is of vital importance for regional security. The long short-term memory (LSTM) model represents a deep network structure to incorporating crop processes, which has been proven accommodate different types representations data, recognize sequential patterns over time spans, capture complex nonlinear relationships. LSTM was developed estimate wheat the Guanzhong Plain by integrating meteorological data two remotely sensed indices, vegetation temperature index (VTCI) leaf area (LAI) at main stages. Considering characteristics memorizing series information, we adopted steps yield. results showed that accuracy estimation highest (RMSE = 357.77 kg/ha R2 0.83) under input combination (meteorological indices). We evaluated optimal performance compared with back propagation neural (BPNN) support vector machine (SVM). As result, outperformed BPNN (R2 0.42 RMSE 812.83 kg/ha) SVM 0.41 867.70 kg/ha), since its recurrent can incorporate relationships between multi-features inputs To further validate robustness method, correlations estimated measured irrigation sites rain-fed from 2008 2016 were analyzed, demonstrated proposed serve as effective approach type sampling better adaptability interannual fluctuations climate. Our findings reliable promising improving estimation.