Scholarlink

<p>Kaziranga National Park (KNP), a UNESCO World Heritage Site in Assam, India, is a renowned biodiversity hotspot home to endangered species like the one-horned rhinoceros. However, annual monsoon floods have disrupted its delicate ecosystem, highlighting the need to analyse flood patterns and assess susceptibility. This study has utilized Google Earth Engine to generate Flood Frequency (FF) maps for KNP from 228 Sentinel-1 SAR images (2015–2022). These FF maps are crucial for identifying flood and non-flood-prone areas. A dataset of 1,246 flood and 421 non-flood samples has been rigorously validated using Sentinel-2 Optical images, NDWI and MNDWI maps. This flood inventory supported the Flood Susceptibility Map (FSM) development using 12 key parameters. Selection of parameters has been refined using Multicollinearity and Boruta algorithm, identifying 8 and 12 significant factors, respectively. Five ensemble machine learning models, Random Forest (RF), Stochastic Gradient Boosting (SGB), Boosted Regression Tree (BRT), DeepBoost (DB) and LogitBoost (LB) have been applied. Boruta selected parameters outperformed those identified by Multicollinearity, achieving higher Area Under Curve values of 0.99 and 1 for RF and SGB models, compared to 0.91 and 0.89 for RF and BRT models. FSM results have highlighted areas near the Brahmaputra as highly flood-prone, with elevation playing a significant role in flood dynamics. Using the Multicollinearity parameters, RF and BRT models have classified 75.39% and 81.49% of the area as Highly flood susceptible, respectively. In contrast, Boruta parameters have yielded notably lower classifications, with RF identifying 36.18% and SGB delineating 32.09% of the area as Highly flood susceptible.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Flood Susceptibility Assessment of Kaziranga National Park Using Ensemble Machine Learning Algorithms and Sentinel-1 SAR Data

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Prasad Balasaheb Wale,
  • Thota Sivasankar,
  • Ratna Sanyal,
  • Surajit Ghosh,
  • Hari Shanker Srivastava

摘要

Kaziranga National Park (KNP), a UNESCO World Heritage Site in Assam, India, is a renowned biodiversity hotspot home to endangered species like the one-horned rhinoceros. However, annual monsoon floods have disrupted its delicate ecosystem, highlighting the need to analyse flood patterns and assess susceptibility. This study has utilized Google Earth Engine to generate Flood Frequency (FF) maps for KNP from 228 Sentinel-1 SAR images (2015–2022). These FF maps are crucial for identifying flood and non-flood-prone areas. A dataset of 1,246 flood and 421 non-flood samples has been rigorously validated using Sentinel-2 Optical images, NDWI and MNDWI maps. This flood inventory supported the Flood Susceptibility Map (FSM) development using 12 key parameters. Selection of parameters has been refined using Multicollinearity and Boruta algorithm, identifying 8 and 12 significant factors, respectively. Five ensemble machine learning models, Random Forest (RF), Stochastic Gradient Boosting (SGB), Boosted Regression Tree (BRT), DeepBoost (DB) and LogitBoost (LB) have been applied. Boruta selected parameters outperformed those identified by Multicollinearity, achieving higher Area Under Curve values of 0.99 and 1 for RF and SGB models, compared to 0.91 and 0.89 for RF and BRT models. FSM results have highlighted areas near the Brahmaputra as highly flood-prone, with elevation playing a significant role in flood dynamics. Using the Multicollinearity parameters, RF and BRT models have classified 75.39% and 81.49% of the area as Highly flood susceptible, respectively. In contrast, Boruta parameters have yielded notably lower classifications, with RF identifying 36.18% and SGB delineating 32.09% of the area as Highly flood susceptible.