Actual Evapotranspiration Dataset Merged from Multiple Data Sources for Xinjiang (1990–2022)

This work generated a long-term monthly ET product at resolutions of 0.1° (1990-2022) by integrating three widely used datasets: The land surface component of the fifth-generation ECMWF reanalysis (ERA5-Land), the second Modern-Era Retrospective analysis for Research and Applications (MERRA-2), and the Global Land Evaporation Amsterdam Model 4.0 (GLEAM). Here, we used triple collocation (TC) approach to merge data considering the error cross-correlation (ECC) conditions and applied machine learning methods to correct the merged product. The statistical performance of the merged ET estimates was evaluated by comparing them with the Budyko ET estimates. The results showed that compared to the three input products , the ET estimates from the TC-based merging framework exhibited the highest correlation with Budyko ET and the smallest bias. In addition to the merging framework demonstrating good performance, the correction model (Feature Tokenizer Transformer, FTT) demonstrated effective performance in correcting the merged ET product over Xinjiang, achieving a correlation coefficient of 0.94 and a root mean square error (RMSE) of 2.85 mm relative to Budyko ET. The ET simulated by TC-FTT shows a predominant increasing trend across Xinjiang, with the annual mean ET increasing at a rate of 0.64 mm yr⁻¹ from 1990 to 2022. In contrast, decreasing trends are observed in eastern Xinjiang and the western Junggar Basin. Notably, both the increasing and decreasing trends are greater in summer than in the other seasons.