Jeemi Sung
Jeemi Sung has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
Changhyun Jun
1 shared publicationJongjin Baik
1 shared publication- Seoyeong Ku1 shared publication
- Seunghyun Hwang
Department of Civil and Environmental Engineering, Chung-Ang University, Seoul 06974, Republic of Korea
1 shared publication
Latest Publications
- Open Access
- 11 February 2026
- Published by Elsevier in SSRN Electronic Journal
A Machine Learning Framework for Identifying Potential Candidate Sites for Inland Wetland Restoration Using Remote Sensing Data
Inland wetlands provide essential hydrological–ecological functions, yet their conservation is increasingly challenged by land-use and climate change, underscoring the need for robust, large-scale approaches to identifying wetland restoration candidates. This study proposes a data-driven framework for classifying inland wetlands and identifying stable restoration candidates by integrating multi-source remote sensing and reanalysis datasets (MODIS, GPM, and ERA5-Land) within an eXtreme Gradient Boosting-based binary classification model. To explicitly account for type-dependent environmental characteristics, inland wetlands across South Korea were classified into four categories: mountain, lake, river, and constructed wetlands. For each wetland type, the 10 largest sites from the national inland wetland inventory were selected for type-specific model training using 13 hydrometeorological, vegetation, and carbon-related variables derived from monthly data (2017–2022). The trained models were applied to seasonally averaged 2023 datasets, and stable restoration candidates were identified by extracting the intersection of pixels classified as wetlands under a binary classification criterion set at a 99% confidence level, thereby ensuring consistency despite seasonal variability. Predictive performance varied with training data availability: river wetlands achieved the highest accuracy (F1 = 0.88), whereas mountain wetlands showed the lowest performance (F1 = 0.82) among wetland types, due to limited training samples and environmental heterogeneity. When the trained models were applied nationwide across South Korea, seasonally consistent restoration candidates accounted for 3.57% of mountain wetlands, 3.24% of lake wetlands, 2.80% of constructed wetlands, and 0.52% of river wetlands. These results provide a nationwide spatial baseline for identifying potential sites for inland wetland restoration.
- 11 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Machine Learning Framework for Identifying Potential Candidate Sites for Inland Wetland Restoration Using Remote Sensing Data
Inland wetlands provide essential hydrological–ecological functions, yet their conservation is increasingly challenged by land-use and climate change, underscoring the need for robust, large-scale approaches to identifying wetland restoration candidates. This study proposes a data-driven framework for classifying inland wetlands and identifying stable restoration candidates by integrating multi-source remote sensing and reanalysis datasets (MODIS, GPM, and ERA5-Land) within an eXtreme Gradient Boosting-based binary classification model. To explicitly account for type-dependent environmental characteristics, inland wetlands across South Korea were classified into four categories: mountain, lake, river, and constructed wetlands. For each wetland type, the 10 largest sites from the national inland wetland inventory were selected for type-specific model training using 13 hydrometeorological, vegetation, and carbon-related variables derived from monthly data (2017–2022). The trained models were applied to seasonally averaged 2023 datasets, and stable restoration candidates were identified by extracting the intersection of pixels classified as wetlands under a binary classification criterion set at a 99% confidence level, thereby ensuring consistency despite seasonal variability. Predictive performance varied with training data availability: river wetlands achieved the highest accuracy (F1 = 0.88), whereas mountain wetlands showed the lowest performance (F1 = 0.82) among wetland types, due to limited training samples and environmental heterogeneity. When the trained models were applied nationwide across South Korea, seasonally consistent restoration candidates accounted for 3.57% of mountain wetlands, 3.24% of lake wetlands, 2.80% of constructed wetlands, and 0.52% of river wetlands. These results provide a nationwide spatial baseline for identifying potential sites for inland wetland restoration.