Study on Soil Moisture Prediction based on the VMD-PCA-Informer Model

Accurate soil moisture prediction is crucial for the effective management of agricultural water resources, ensuring agricultural management and decision-making, crop monitoring and early warning, as well as securing food supply and safety. Factors such as precipitation, evaporation, temperature, humidity, pressure, and wind speed have a significant impact on soil moisture, making it difficult to predict accurately. In this study, a soil moisture prediction method combining Principal Component Analysis (PCA), Variational Mode Decomposition (VMD), and the Informer model is proposed. Firstly, PCA is applied to multidimensional remote sensing data that influence wind speed for dimensionality reduction. Then, the soil moisture data is decomposed into modes using VMD. Finally, the reduced feature data from PCA, the decomposed modes of soil moisture, and the original soil moisture data are merged as inputs to the Informer network for soil moisture prediction. To showcase the benefits and practicality of the proposed approach, a comparison was conducted against other methods of mode decomposition as well as time series prediction models. Experimental data show that the average MAE of the VMD-PCA-Informer model is 0.275, the average RMSE is 0.448, and the R2_score is 0.865 at the study sites. These results show that the VMDPCA-Informer model has better prediction accuracy than the other models.