Mapping Surface Water Dynamics and Land Cover Types for Global Aquatic Lands

Abstract

<p>Monitoring surface water dynamics and vegetation types is important to aquatic ecosystem management. With the advantages of high spatiotemporal resolution and wide coverage, remote sensing has played a vital role in Global Aquatic Land Cover (GALC) mapping. In recent years, several global water products have been produced. However, these datasets are mainly focused on open water areas, while the surface water dynamics and land cover types in vegetated aquatic areas have always been under-represented, calling for more attention to improve GALC mapping in these areas. This study aims to propose a framework to characterize the intra-annual surface water dynamics and specific land cover types for GALC, with a focus on vegetated areas.</p> <p>In the proposed framework, we firstly characterized the specific vegetation/non-vegetation types for GALC using different combinations of features derived from optical, RADAR, and multiple ancillary data. By cross-validating our results at globally distributed aquatic sample sites, we found that combining all the data sources achieved the highest overall accuracy (83.2%). By combining SAR features from the ALOS/PALSAR mosaic and Sentinel-1 data with Sentinel-2 data, the misclassification of highly mixed and spectrally similar types, such as shrubs, trees, and herbaceous vegetation could be reduced.</p> <p>Next, we mapped the surface water dynamics for 2019 at a 10 m-resolution at the same globally distributed aquatic sample sites. The reference data on the monthly water status (i.e., water or non-water) were obtained from existing global water products and mangrove datasets. Then, the monthly water status was fine-tuned based on optical and SAR features. We mapped the monthly water presence using the derived reference data using the Random Forest classification of the fused Sentinel-1 and Sentinel-2 data. Finally, the monthly maps were aggregated to derive four water persistence types (i.e., permanently flooded, temporarily flooded, waterlogged, and non-flooded) by defining the water frequency.</p> <p>The proposed framework was also demonstrated at four wetlands featured by different hydrological and climatic conditions. Compared with JRC’s Global Surface Water product, the monthly water maps created in this study performed better in detecting water in vegetated areas. However, mapping the surface water dynamics in vegetated areas still faces challenges. For example, as the vegetation density increased from 0-25% to 75-100%, the producer’s accuracy of permanently flooded shrubs and trees dropped by 80% and 43%, respectively. Results also showed that waterlogged trees/shrubs/herbaceous types had significant misclassifications with non-flooded classes.</p> <p>The framework proposed in this study is able to improve assessments of global water resources, especially in vegetated aquatic ecosystems, and help reveal the impact of global climate change on the distribution of surface water.</p>