A mechanism-guided machine learning method for mapping gapless land surface temperature

More accurate, spatio-temporally, and physically consistent land surface temperature (LST) estimation has been a main interest in Earth system research. Developing physics-driven mechanism models and data-driven machine learning (ML) models are two major paradigms for gapless LST estimation, which have their respective advantages and disadvantages. In this paper, a mechanism-guided ML model, which combines the strengths in the mechanism model and ML model, is proposed to generate gapless LST with physical meanings and high accuracy. The hybrid model employs ML as the primary architecture, under which the input variable mechanistic guidance is incorporated to enhance the interpretability and extrapolation ability of the model. Specifically, the light gradient-boosting machine (LGBM) model, which only uses remote sensing data as input, serves as the pure ML model. Mechanistic guidance (MG) is coupled by further incorporating key Community Land Model (CLM) forcing data (cause) and CLM simulation data (effect) as inputs into the LGBM model. This integration forms the MG-LGBM model, which incorporates surface energy balance (SEB) guidance underlying the data in CLM-LST modeling within a biophysical framework. Results indicate that, MG-LGBM model shows a good accuracy for the sample-based validation, with a root-mean-square error of 1.23–2.03 K, and a Pearson correlation coefficient of 0.99. Validation with four independent ground measurements shows that MG-LGBM can generate clear-sky LST that is comparable to the original Moderate Resolution Imaging Spectroradiometer- (MODIS) LST under fully clear-sky conditions and can correct for the likely cloud-contaminated LST pixels. The generated LST also presents a high accuracy (RMSE = 2.91–3.66 K and R = 0.97–0.98) under cloudy-sky conditions. Compared with a pure mechanistic method and pure ML methods, the MG-LGBM model improves the prediction accuracy and mechanistic interpretability of LST. It also demonstrates a good extrapolation ability in the regions without valid samples, suggesting that the predictions of MG-LGBM model not only exhibit low errors on the training dataset but also align consistently with the known mechanistic laws in the unlabeled set. Compared with other popular ML methods and sophisticated gapless products, the MG-LGBM model delivers a superior validation accuracy and image quality. The proposed method represents an innovative way to map accurate and mechanistically interpretable gapless LST, and could provide insights to accelerate knowledge discovery in land surface processes and data mining in geographical parameter estimation.