Mapping Cambodian Wetlands with Satellite Imagery and Google Earth Engine’s Machine Learning Algorithm.

The wetlands and lakes that make up more than 30% of Cambodia's terrain are home to a diverse range of resources and biodiversity. More than 46% of the population lives and works in these wetlands while 80% of the local population relies on their vital resources for sustenance such as fish, food, water and vegetables. This makes Cambodia one of the nations with the highest reliance on wetland and lake ecosystems in the world. On-going development in the region has boosted the rates of  urbanization. Urban expansion has deteriorated wetland ecosystems through land reclamation and infilling projects as well as hydrological and sediment cycle disruptions. It has also increased the demand for mined sand from the Mekong River. Mapping and monitoring the extent and distribution of wetland ecosystems in order to quantify the impact of human activities on these vital areas is critical for maintaining the ecological balance and promoting the sustainable development of an extensively eco-service dependent country such as Cambodia. In this study we combine spaceborne multispectral and radar remote sensing datasets with machine learning classification models and algorithms within the Google Earth Engine to monitor the changes observed in Cambodian wetlands through time. Our classifier is trained by comparing Sentinel 1 Synthetic Aperture Radar data to corresponding multispectral images captured from Landsat. We then use the classifier to monitor wetland extent through time from 1989 to present using merged Landsat 5 and 8 databases. With our maps and areal statistics, we identify the spatio-temporal trends and changes in wetland cover linked to climatic patterns and local anthropogenic influence connected to sand mining from the Mekong River and land infilling. In the last 15 years, about half the country’s wetlands have disappeared, with 15 out of 25 lakes near the capital completely infilled with sand that can be clearly observed with analysis of satellite data.