Oral

IAMAS

JCM03 - Permafrost under changing climate

Mapping of peatland in Mongolia along the southern fringe of Siberian permafrost region

Mongolia exhibits extensive development of peatlands and is recognized as one of the top ten countries globally with substantial peatland coverage. It is located in the southern fringe of Siberian permafrost regions and has a widespread distribution of peatlands. Such peatlands mainly developed in permafrost regions in Mongolia due to cold and wet environmental conditions. Permafrost is widely distributed across the Mongolian landscape, covering more than 29.3 % of the national area half of this area was the continuous permafrost. During recent decades, Mongolian peatlands have become severely degraded due to both climate related events and overgrazing of livestock. However, while previous maps of peatland distribution have been widely used, they are largely based on experiential judgment and lack accuracy validation. Thus, the spatial distribution of peatlands in Mongolia remains largely uncertainties. The specific objectives of this study are to (1) identify the main drivers that affect the peatland distribution in Mongolia based on long-term satellite imagery and field measurement data; (2) predict SOC content distribution; (3) create a map of peatland distribution in Mongolia. We conducted a field survey to collect soil samples from peatlands in Mongolian permafrost regions, explicitly targeting the Altai, Khangai, Khentii, and Khuvsgul mountains from June to October 2022 and 2023. Soil samples at 1246 sites were taken at depths between 0-5 cm, 10-15 cm, 15-20 cm, and 20-30 cm. We selected 37 SOC content covariates based on their impact as soil-forming factors. We grouped these covariates into five distinct categories in this study: climatic parameters, optical parameters, including reflectance and biophysical parameters, soil characteristics parameters, topographic parameters, and category parameters. Most of the above-mentioned covariates were acquired from the Google Earth Engine (GEE) platform. This study used machine learning to predict peatlands based on SOC content distribution. Spatial prediction of SOC content was based on Random Forest (RF). We chose the RF model because it had the highest level of predictive accuracy and it is a well-known, widely used technique for digital soil mapping. The best result was reached by the RF model with 37 covariates that explained 89% SOC content variability across the country, with R2 of 0.89, RMSE of 44.62 g/kg and MAE of 22.92 g/kg in a 10-times repeated 5-fold cross-validation procedure. When independent training and test sample sets were used to train and evaluate the model, it showed an R2 of 0.73, RMSE of 58.16 g/kg and MAE of 38.97 g/kg. We created a map of spatial distribution of SOC content over Mongolian region at depths of 0-30 cm and resolution with 250 m. According to the spatial distribution map, its content ranges between 0.20 g/kg- 707.53 g/kg and the average value is 25.81 g/kg. In this study, we used machine learning methods to create a peatlands map in Mongolia. According to this mapping result, we estimated SOC content for all soil samples and identified peat soil more than 120 g/kg. As a result of our study, peatlands are found an area of 29881 km�� in Mongolia.