The Innovation Team of Agricultural Remote Sensing from the Institute of Agricultural Resources and Regional Planning (IARRP) of the Chinese Academy of Agricultural Sciences recently reported a significant advance in land surface temperature (LST) retrieval. Their study, titled "A framework of coupling split-window and machine learning (SW-ML) for land surface temperature retrieval from MODIS thermal infrared data," was published in the journal Remote Sensing of Environment.

Land surface temperature is a key parameter governing energy and water exchanges between the land and the atmosphere, playing a crucial role in meteorology, ecology, and hydrology. However, existing LST retrieval methods face notable challenges. Physically based split-window algorithms often lack accuracy and computational efficiency, while data-driven machine learning approaches typically lack generalization capability and interpretability.

To address these challenges, the research team developed a novel framework that combines physical models with machine learning techniques. The approach integrates machine learning–estimated residuals to correct biases in the split-window algorithm, achieving high accuracy and strong robustness under challenging conditions such as high temperature, high humidity, and bare land surfaces.

Figure: Flowchart of the coupled physical model and machine learning framework for land surface temperature retrieval

The framework was systematically evaluated through spatiotemporally independent validation, ten-fold cross-validation, and performance comparisons under various environmental conditions. Additionally, SHAP (Shapley Additive Explanations) analysis was applied to attribute the estimated residuals, enhancing the interpretability of the model.

Results demonstrate that the hybrid framework delivers high accuracy, robustness, and interpretability in LST retrieval, particularly under complex surface and atmospheric conditions. The findings provide important technical support for studies on land–atmosphere interactions, climate change, and ecological processes.

The study's first author is Guan Yongjuan, a doctoral student at IARRP, with Researcher Duan Sibo serving as the corresponding author. The research was supported by the State Key Laboratory of Efficient Utilization of Arid and Semi-arid Arable Land in Northern China and the National Natural Science Foundation of China.

Original paper link: 

https://www.sciencedirect.com/science/article/pii/S0034425726001823