International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 07 | Jul 2025
p-ISSN: 2395-0072
www.irjet.net
Geosheild-Geospatial data-driven Flood and Landslide Prediction Kshitija Thakare, Hindavi Patil, Pratibha Salave, Juhi Agrawal 1 Student, G H Raisoni College of Engineering and Management, Pune, India , 2 Student, G H Raisoni College of Engineering and Management, Pune, India 3 Student, G H Raisoni College of Engineering and Management, Pune, India 4 Assistant Professor, G H Raisoni College of Engineering and Management, Pune, India -----------------------------------------------------------------------------***-----------------------------------------------------------------------------
Abstract—Natural disasters like floods and landslides pose severe threats to human lives and infrastructure, particularly in regions with complex topography. This study proposes a machine learning-based predictive framework for flood and landslide risk assessment, integrating spatial and temporal data to enhance early warning systems. The framework consists of two key components: (1) a temporal prediction model utilizing Long Short-Term Memory (LSTM) networks to forecast water levels and rainfall intensity, and (2) a spatial analysis model leveraging Inverse Distance Weighting (IDW) for generating flood and landslide hazard maps. By incorporating historical weather patterns, dam levels, land use data, and socio-economic indicators, the system provides real-time risk assessment and mitigation strategies. Experimental validation demonstrates significant improvements in prediction accuracy over conventional models, enabling proactive disaster response and enhanced preparedness. This approach has potential applications in disaster management, urban planning, and policy-making, ensuring better resilience in vulnerable regions. Index Terms— Flood Prediction, Landslide Detection, Machine Learning, LSTM, IDW, Spatial Analysis, Early Warning System, Disaster Management.
I. INTRODUCTION
leverages multiple data sources, including historical weather patterns, dam levels, land use data, and socio-economic indicators. The framework is designed to improve prediction accuracy and provide real-time alerts, ensuring better preparedness for disaster-prone regions. This study proposes a two-component approach: (1) a temporal prediction model utilizing Long Short-Term Memory (LSTM) networks to forecast water levels and rainfall intensity, and (2) a spatial analysis model using Inverse Distance Weighting (IDW) to generate detailed flood and landslide hazard maps. By combining these methodologies, the system aims to deliver reliable early warnings, reduce economic losses, and support decision-makers in implementing timely evacuation and mitigation strategies. The experimental results demonstrate significant improvements over traditional prediction models, making this framework an effective tool for disaster risk reduction. This research has wide-ranging applications in disaster management, urban planning, and policy-making, ultimately fostering resilience in vulnerable communities. The findings emphasize the need for data-driven approaches in predicting and mitigating natural disasters, ensuring a more effective and systematic response to floods and landslides.
II. LITERATURE REVIEW
Natural disasters such as floods and landslides have devastating impacts on human life, infrastructure, and the economy, particularly in regions with complex topography. The unpredictability of these events poses significant challenges for disaster management authorities, making early warning systems crucial for proactive mitigation and response. Accurate prediction of floods and landslides requires the integration of both spatial and temporal data to enhance real-time monitoring and forecasting capabilities. Traditional disaster prediction models primarily rely on historical data and static hazard maps, which often lack the adaptability needed for dynamic risk assessment. To overcome these limitations, this research introduces a machine learning-based predictive framework that
The The development of landslide detection and monitoring systems using various sensor technologies and machine learning approaches has been extensively researched. One significant study focused on Wireless Sensor Networks (WSNs) for real-time environmental disaster monitoring. The system utilized Linear Regression Algorithms for landslide prediction, demonstrating high scalability and adaptability [1]. Another study explored deep learning models for landslide prediction, integrating an Attention-Based Temporal Convolutional Network (TCN) linked with a Recurrent Neural Network (RNN). This model effectively predicted landslide instability margins (LIMs) by utilizing rainfall simulation and sensor-based data collection [2]. Remote sensing imagery has also been widely used in landslide detection. A study proposed Landslide SegNet, a deep learning model based on an EncoderDecoder Residual (EDR) architecture. This approach achieved
© 2025, IRJET
ISO 9001:2008 Certified Journal
|
Impact Factor value: 8.315
|
|
Page 712