پايش تغييرات محدوده‌ي منابع دائمي برف ايران با كمك افزايش قدرت تفكيك تصاوير سنجنده‌ي MSS لندست با روش‌هاي يادگيري ماشين(در بازه‌ي 2023-1975)

Perennial an‎d long-lasting mountain snow resources in Iran play a critical role in regulating streamflow regimes, recharging aquifers, an‎d sustaining water-dependent ecosystems. However, long-term, national-scale monitoring of these resources remains challenging due to sparse in-situ observations, cross-sensor heterogeneity, an‎d the coarse spatial resolution of historical Lan‎dsat MSS imagery. This research aims to generate a consistent an‎d reliable time series of changes in the extent an‎d area of Iran’s persistent snow resources over 1975–2025 by leveraging the multi-sensor Lan‎dsat archive an‎d machine-learning techniques. To this end, Lan‎dsat MSS images were first enhanced to 30 m resolution using deep learning–based super-resolution (SR) to enable homogeneous snow extraction over the past five decades. Subsequently, focusing on warm-season imagery (June–August) as a proxy for persistent snow, annual snow/cloud/other maps were produced via supervised classification an‎d annual snow area was quantified. For the super-resolution stage, three architectures, SRGAN, ESRGAN, an‎d SwinIR were eva‎luated. Based on quantitative an‎d qualitative assessments, SwinIR achieved the best reconstruction performance (PSNR = 22.29 dB, SSIM = 0.639). Moreover, classification-accuracy assessment for representative MSS-period years indicated that incorporating super-resolution not only improved spatial detail an‎d reduced missed-snow pixels, but also increased the mean Overall Accuracy (OA) from 0.793 to 0.824 an‎d the Kappa coefficient from 0.686 to 0.733. Time-series analysis of snow area across the full period (44 years) revealed strong interannual variability alongside a clear long-term decline, with values ranging from 9.196 to 505.885 km² (mean = 164.94 km²). The average snow area was approximately 214–233 km² during earlier epochs (MSS an‎d TM) an‎d decreased to roughly 102–113 km² in more recent epochs (ETM+ an‎d OLI), indicating a substantial reduction of persistent snow resources in recent decades. Overall, the proposed framework mitigates inconsistencies in historical observations an‎d delivers a harmonized 30 m product, enabling robust national-scale climate–hydrology analyses an‎d supporting water-resources planning.

5

158314