Scholarlink

<p>Climate-driven rainfall variability poses considerable challenges to infrastructure resilience in Bangladesh, particularly in the monsoon-sensitive northwestern region. This study presents a comprehensive framework for long-term rainfall forecasting and flood-susceptibility mapping, integrating deep learning (LSTM), statistical forecasting (Prophet), and GIS-based vulnerability analysis. Historical rainfall data from Rajshahi and Pabna were utilized to train and evaluate the models, with LSTM achieving better performance compared to Prophet, as indicated by lower RMSE and higher R² values. Forecasts for 2025–2039 indicate alternating spells of rainfall and moderate moisture, mirroring the regular monsoon oscillations. These rainfall estimates were integrated with NDVI, NDWI, and land-sensitivity layers to construct multi-year flood vulnerability zones. Results highlight emerging high-risk hotspots, emphasizing the need for adaptive planning in road, drainage, and agricultural systems. The proposed approach presents a methodical pathway for anticipatory climate-resilient infrastructure planning in Bangladesh and can be further enhanced as additional satellite data and ground observations become available.</p> Graphical Abstract <p></p> <p>The graphical abstract depicts a rainfall forecasting and climate-resilient infrastructure design process in northwest Bangladesh. This begins with the identification of regions in Rajshahi and Pabna, two districts vulnerable to hydroclimatic alterations. Historical rainfall data from 1980 to 2024, obtained from the Bangladesh Meteorological Department (BMD), is subjected to preprocessing. Several statistical performance metrics are used to train and assess two forecasting models: Facebook Prophet and Long Short-Term Memory (LSTM). In 2025, the NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) are used to extract data from the Landsat 9 satellite. GIS spatial overlays incorporate geographical layers to identify high-risk flood areas and categorize them as low- or high-risk. Annually, the forecasts are evaluated to ascertain the long-term variability of precipitation. Results are then connected to infrastructure development plans, which will allow improved policy decisions and mitigate the possible risk of future floods. To turn climate foresight into district-level resilience planning, this end-to-end system showcases the possibilities of an integrated machine learning, remote sensing, and GIS system.</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

Artificial Intelligence-Driven Rainfall Forecasting and GIS-Based Flood Risk Mapping for Climate-Resilient Infrastructure in Bangladesh

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Hasan Ahamed Alif,
  • Md. Jisan Mashrafi,
  • Mohamed Elhag,
  • Sanju Purohit

摘要

Climate-driven rainfall variability poses considerable challenges to infrastructure resilience in Bangladesh, particularly in the monsoon-sensitive northwestern region. This study presents a comprehensive framework for long-term rainfall forecasting and flood-susceptibility mapping, integrating deep learning (LSTM), statistical forecasting (Prophet), and GIS-based vulnerability analysis. Historical rainfall data from Rajshahi and Pabna were utilized to train and evaluate the models, with LSTM achieving better performance compared to Prophet, as indicated by lower RMSE and higher R² values. Forecasts for 2025–2039 indicate alternating spells of rainfall and moderate moisture, mirroring the regular monsoon oscillations. These rainfall estimates were integrated with NDVI, NDWI, and land-sensitivity layers to construct multi-year flood vulnerability zones. Results highlight emerging high-risk hotspots, emphasizing the need for adaptive planning in road, drainage, and agricultural systems. The proposed approach presents a methodical pathway for anticipatory climate-resilient infrastructure planning in Bangladesh and can be further enhanced as additional satellite data and ground observations become available.

Graphical Abstract

The graphical abstract depicts a rainfall forecasting and climate-resilient infrastructure design process in northwest Bangladesh. This begins with the identification of regions in Rajshahi and Pabna, two districts vulnerable to hydroclimatic alterations. Historical rainfall data from 1980 to 2024, obtained from the Bangladesh Meteorological Department (BMD), is subjected to preprocessing. Several statistical performance metrics are used to train and assess two forecasting models: Facebook Prophet and Long Short-Term Memory (LSTM). In 2025, the NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) are used to extract data from the Landsat 9 satellite. GIS spatial overlays incorporate geographical layers to identify high-risk flood areas and categorize them as low- or high-risk. Annually, the forecasts are evaluated to ascertain the long-term variability of precipitation. Results are then connected to infrastructure development plans, which will allow improved policy decisions and mitigate the possible risk of future floods. To turn climate foresight into district-level resilience planning, this end-to-end system showcases the possibilities of an integrated machine learning, remote sensing, and GIS system.