Multi-temporal remote sensing of inland surface waters: A fusion of sentinel-1&2 data applied to small seasonal ponds in semiarid environments

• Surface water attributes in small water bodies were modelled across semiarid regions. • Models including Sentinel-1&2 metrics achieved satisfactory classification precision. • Sentinel-1&2 models were competitive with Landsat for local spatiotemporal water inference. • Sentinel data offer ecological and sustainable applications for small-sized water bodies. Inland freshwaters are essential in maintaining ecological balance and supporting human development. However, comprehensive water data cataloguing remains insufficient, especially for small water bodies (i.e., ponds), which are overlooked despite their ecological importance. To address this gap, remote sensing has emerged as a possible solution for understanding ecohydrological characteristics of water bodies, particularly in water-stressed areas. Here, we propose a novel framework based on a Sentinel-1&2 local surface water (SLSW) model targeting very small (<0.5 ha, Mdn ≈ 0.031 ha) and seasonal water bodies. We tested this framework in three semiarid regions in SW Iberia, subjected to distinct seasonality and bioclimatic changes. Surface water attributes, including surface water occurrence and extent, were modelled using a Random Forests classifier, and SLSW time series forecasts were generated from 2020 to 2021. Model reliability was first verified through comparative data completeness analyses with the established Landsat-based global surface water (LGSW) model, considering both intra-annual and inter-annual variations. Further, the performance of the SLSW and LGSW models was compared by examining their correlations for specific periods (dry and wet seasons) and against a validation dataset. The SLSW model demonstrated satisfactory results in detecting surface water occurrence ( μ ≈ 72 %), and provided far greater completeness and reconstructed seasonality patterns than the LGSW model. Additionally, SLSW model exhibited a stronger correlation with LGSW during wet seasons (R 2 = 0.38) than dry seasons (R 2 = 0.05), and aligned more closely with the validation dataset (R 2 = 0.66) compared to the LGSW model (R 2 = 0.24). These findings underscore the SLSW model’s potential to effectively capture surface characteristics of very small and seasonal water bodies, which are challenging to map over broad regions and often beyond the capabilities of conventional global products. Also, given the vulnerability of water resources in semiarid regions to climate fluctuations, the present framework offers advantages for the local reconstruction of continuous, high-resolution time series, useful for identifying surface water trends and anomalies. This information has the potential to better guide regional water management and policy in support of Sustainable Development Goals, focusing on ecosystem resilience and water sustainability.