Downscaling Microwave‐Based Evapotranspiration With a Fourier‐Supervised Multi‐Source Fusion Network in Central‐Southern East Asia

Abstract Evapotranspiration (ET) is a critical component of the land‐atmosphere energy and water cycle. Satellite remote sensing has proven to be highly effective for large‐scale ET estimation across heterogeneous landscapes, but producing high‐resolution, all‐weather ET remains difficult. Passive microwave remote sensing enables daily all‐weather ET retrievals, but suffers from coarse spatial resolution. Meanwhile, optical remote sensing provides finer spatial resolution but is hindered by prevailing cloud contamination, limiting its temporal availability. To leverage the complementary strengths of both data sources, we propose a Fourier‐supervised Multi‐source Fusion ET Downscaling Network (FMED‐Net), which fuses daily 0. 25° microwave ET retrievals (ETEDVI) and 8‐day 0. 05° optical ET (ETMODIS), along with various auxiliary data to produce all‐weather daily 0. 05° ETFMED. Integrating super‐resolution techniques and Fourier domain transformations, FMED‐Net effectively captures complex relationships in both the spatial and frequency domains across multi‐source data. The proposed method is applied to central‐southern East Asia during 2016–2018. Validations against in situ measurements at 10 eddy covariance sites suggest improved performance of 0. 05° ETFMED compared with the original 0. 25° ETEDVI (Nash–Sutcliffe Efficiency +30. 27%, Kling–Gupta Efficiency +19. 93%, Bias −37. 43%). FMED‐Net also performs evidently better than the four representative machine learnings and deep learning models in both overall accuracy and the ability to preserve fine spatial structures. Furthermore, it effectively captures detailed spatial and temporal ET dynamics and maintains robust performance under cloudy conditions, compensating for the limitations of 8‐day optical ET products. These advantages highlight its potential for developing future global, daily, and all‐weather ET products based on multi‐source fusion.