A Training Sample Migration Method for Wetland Mapping and Monitoring Using Sentinel Data in Google Earth Engine

Wetlands are one of the most important ecosystems due to their critical services both humans and environment. Therefore, wetland mapping monitoring essential for conservation. In this regard, remote sensing offers efficient solutions availability cost-efficient archived images over different spatial scales. However, a lack sufficient consistent training samples at times is significant limitation multi-temporal monitoring. study, new sample migration method was developed identify unchanged be used in classification change analyses International Shadegan Wetland (ISW) areas southwestern Iran. To end, we first produced map reference year (2020), which had samples, by combining Sentinel-1 Sentinel-2 Random Forest (RF) classifier Google Earth Engine (GEE). The Overall Accuracy (OA) Kappa coefficient (KC) were 97.93% 0.97, respectively. Then, an automatic detection migrate from target years 2018, 2019, 2021. Within proposed method, three indices Normalized Difference Vegetation Index (NDVI), Water (NDWI), mean Standard Deviation (SD) spectral bands, along with two similarity measures Euclidean Distance (ED) Spectral Angle (SAD), computed each pair reference–target years. optimum threshold also derived using histogram thresholding approach, led selecting that likely based on highest OA KC classifying test dataset. resulted high OAs 95.89%, 96.83%, 97.06% KCs 0.95, 0.96, 0.96 2021, Finally, migrated generate Overall, our showed potential when no existed year.