Spectral-gap-aware domain adaptation for cross-sensor evapotranspiration retrieval under chronic ground-truth scarcity: a climate-analytics evaluation of irrigation-advisory reliability in data-sparse basins

Cross-sensor evapotranspiration (ET) retrieval in data-sparse basins is constrained by a spectral gap: the wavebands available to a target sensor differ from those of the source sensor on which a retrieval model was trained, and ground-truth flux towers are too few to fine-tune the model in the target domain. The framework is formalised as a transfer-risk decomposition (Theorem 3.1) and a gap-regularised estimator (Theorem 4.2), and its implications for irrigation-advisory reliability are examined through a climate-analytics evaluation using NASA POWER monthly reanalysis for Kampala, Uganda (0.35°N, 32.58°E), 2000–2023, comprising 312 monthly observations with a mean 2 m air temperature of 22.443 °C and a decadal trend of −0.1564 °C. Advisory reliability is bounded not by model capacity but by the spectral-gap operator norm and by the temporal autocorrelation of the target-domain residual, and that a gap-aware estimator can reduce the advisory false-alarm rate relative to naive transfer when the gap norm is below a stated threshold. The contribution is a formal reliability criterion that links sensor configuration, ground-truth scarcity and irrigation-advisory error, and that can be evaluated before deployment in a new basin.