Monitoring of climate-change-induced floods and impact analysis on agricultural systems in the Mojana region of Colombia through automatic change detection and machine learning

The Mojana region in Colombia is one of the most complex inland water deltas in the world as it receives inflows from important rivers in the country, including the Cauca river. The current climate change affects this unique natural environment and alters the water balance. During the latest “El Niño-Southern Oscillation” (ENSO) cold phase (La Niña) several floods occurred. The flood happened on 2 September 2021 was the longest event ever recorded in the Mojana region and severely impacted on local agricultural systems (crops, livestock-raising areas, etc.). However, whether and how these systems have recovered and been brought back to prior production level is unknown. Given that the recent past trend suggests that in near future other and more frequent flood events may happen, an analysis of the recovery process may unveil strategies for resilience. Because the region is isolated and poorly accessible, we analysed the whole collections of Earth Observation data (optical and SAR) from the Landsat missions, Sentinel-1 and Sentinel-2 in the period 1999- 2024. The time series analysis highlighted major transformations in the agricultural systems in the area affected by the flooding. The data were processed with machine learning (ML) for mapping affected areas, recovery assessment, and analyse of how surfaces evolved temporally. Cloud denoising of optical data with ML is essential in tropical countries, in order to contrast the low availability of cloud-free optical data. From a technical point of view, one of the conclusions is that the use of ML in dual-polarization SAR data helps to achieve better land classifications and contributes to improve accuracy and mapping products.