Comprehensive evaluation of soil moisture in farmland based on Ensemble Empirical Mode Decomposition and Partial Least Squares Regression

Abstract

The North China Plain is an important agricultural production base for China with its flat terrain and ease of cultivation. However, its severe drought problems limit the use of its resource advantages. Crop growth is affected by multi-source compound stresses such as soil moisture stress, pest and disease stress, and heavy metal stress, and accurate screening and monitoring of soil moisture stress is the key to the research. In this paper, the Normalized Difference Vegetation Index (NDVI) long time series curves of winter wheat were constructed using the NDVI as the response parameter by combining the remote sensing image data from the GF-1 satellite and Landsat satellite. Using the Ensemble Empirical Mode Decomposition (EEMD) algorithm to decompose the long time series, make the statistical description of each decomposed Intrinsic Mode Function (IMF), and combined it with the analysis of soil moisture stress mechanism to achieve an effective screening and extraction of soil moisture stress. Partial Least Squares Regression (PLSR) was used to establish the quantitative relationship between remote sensing monitoring indicators and ground-based indicators for soil moisture monitoring and prediction. The results show that: 1) Among the six decomposed IMF components, the statistical descriptors of IMF1 and IMF2 are the most consistent with the characteristics of the mechanism analysis, and the soil moisture stress sequences synthesized from them can better reflect the soil moisture stress conditions in the study area; 2) Chlorophyll Response to Soil Moisture Stress (CR_SMS) and Wheat Moisture Content Response to Soil Moisture Stress (WMCR_SMS) can effectively reflect the response of chlorophyll content of winter wheat leaves and wheat moisture content to soil moisture stress in the study area; 3) The coefficient of determination of the quantitative inversion model based on PLSR is 0.879, with a high degree of model fit and low error. However, the combination of the EEMD algorithm and PLSR modelling can effectively identify and extract soil moisture stress and achieve accurate monitoring and quantitative inversion of soil moisture in cropland, so as to provide reference for irrigation and rational use of water resources in farmland.