Prediction of Soil Moisture in Grassland Based on Information Gain and Model Modification

Abstract

Accurate prediction of grassland soil moisture is the key to understanding and responding to the destruction of grassland ecosystems. In order to improve the prediction effect of grassland soil moisture, this study proposed a grassland soil moisture prediction model based on information gain and model modification based on multi-dimensional and small sample size data from a testing station in Xilingol Grassland. First, we conduct correlation analysis on the 22-dimensional feature data of Xilin Gol grassland based on the Pearson correlation coefficient, and screen out the 15-dimensional feature data that can represent the whole; then calculate the information gain of the features on soil moisture, and screen out the 6-dimensional high information gain features; Secondly, by introducing the evolutionary boundary constraint processing scheme, Levy flight strategy and group fitness variance strategy to jointly improve the pathfinder algorithm (CLPPFA), it improves its global optimization capability, thereby optimizing extreme learning machine (ELM) related parameters and constructing a grassland soil moisture prediction model, and use 6-dimensional high information gain feature data to predict the preliminary results of grassland soil moisture; Finally, by establishing an ARIMA error correction model, the error prediction value and the preliminary prediction value are superimposed to obtain the final prediction result. The results show that the improved extreme learning machine's fitting degree R2 for grassland soil moisture prediction is 0.937, which is better than the PFA_ELM model and the ELM model; the ARIMA model is introduced to analyze the error sequence of the preliminary prediction results of CLPPFA_ELM, and ARIMA (2,0,2) The model is error corrected. The prediction fit R2 of the CLPPFA_ELM-ARIMA model is 0.988, which is significantly improved compared to the prediction effects of the CLPPFA_ELM, SVR, RF, BP and ridge regression models. In summary, it is shown that the model has good fitting effect and generalization ability in grassland soil moisture prediction. This model provides model reference and technical support for formulating effective grassland management and protection strategies.