Publications

Li, SL; Jiang, SQ; Song, N; Han, Y; Wang, JL (2025). Two-step fusion framework for generating 10 m resolution soil moisture with high accuracy in the cotton fields of southern Xinjiang. INDUSTRIAL CROPS AND PRODUCTS, 226, 120582.

Abstract
A timely and accurate high-resolution evaluation of soil moisture dynamics is imperative for drought monitoring and irrigation planning in cotton fields, especially in water-deficient areas. Nonetheless, the existing regional scale soil moisture content determination techniques require major refinements to precision, efficiency and cost. Given this, a new two-step fusion framework was developed to estimate soil moisture content at a resolution of 10 m in cotton fields by integrating ground-based soil moisture and relevant input variables using an automated machine learning (AutoML) method. The input variables include Sentinel-2 multispectral (MS) data, Sentinel-3 thermal infrared (TIR) data downscaled using the data mining sharpener (DMS) algorithm, topographic data and soil property data. In the first step of the framework, the DMS algorithm is employed to downscale Sentinel-3 land surface temperature (LST) data from a 1 km resolution to 10 m, thereby providing high-resolution TIR band information to address spatial scale mismatches. In the second step, an AutoML workflow is developed to automatically select the optimal model for predicting soil moisture content. Validation conducted in a typical cotton irrigation area in southern Xinjiang demonstrated the framework's high accuracy, with Pearson correlation coefficient (R) value of 0.909, root mean square error (RMSE) of 1.98 % and normalized RMSE (NRMSE) of 9.87 %. Overall, the proposed framework exhibits superior accuracy and efficiency, highlighting its strong potential for application in water resource management and irrigation planning in cotton fields.

DOI:
10.1016/j.indcrop.2025.120582

ISSN:
1872-633X