Integration of ecological knowledge with Google Earth Engine for diverse wetland sampling in global mapping

Abstract

Accurate wetland extraction using remote sensing technology poses significant challenges due to the complex hydrological dynamics, diverse landscapes, and varied wetland types. Constructing a reliable sample set is a critical first step in overcoming these challenges for large-scale wetland mapping. To meet the demand for global wetland mapping, this study (1) proposes a multi-level wetland classification system suitable for remote sensing, incorporating the soil moisture, vegetation cover and temporal dynamic characteristics of wetlands; (2) introduces a theoretically plausible wetland sample identification method based on the ecological, geographical and temporal dynamic characteristics of wetland ecosystems; (3) develops an approach that combines the Inundation-Frequency and Ecological Remote Sensing Indicators for global wetland sampling based on global climatic zones. The global wetland sample set was finally produced with 64,486 samples. The dataset revealed that seasonal marsh, swamp, mangrove, floodplain, salt marsh, tidal flat and permanent marsh accounted for 22.99%, 20.05%, 18.06%, 14.58%, 12.38%, 10.62% and 1.29% of the total sample set, respectively. Furthermore, the water body sample set comprised 13,402 samples, distributed among permanent (45.50%), seasonal (31.35%) and temporary (23.15%) water bodies. The proposed knowledge-based method, which makes use of big earth-observing data and the Google Earth Engine platform, has been demonstrated to have the capability to generate reliable wetland samples with a high degree of accuracy. This represents the first effort to create a global wetland sample set, which has the potential to offer critical support for comprehensive wetland mapping initiatives

Keywords

• Global wetland samples
• Remote sensing time series
• Google Earth Engine
• Knowledge-based method
• Wetland remote sensing