International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 11 | Nov 2025
p-ISSN: 2395-0072
www.irjet.net
REAL TIME FLOOD DETECTION AND FORECASTING SYSTEM Sanjivani B. Adsul1, Aditya Ghurye2, Mahesh Dakore3, Mrunali Dhoke4, Kunal Dagade5 , Garvit Khandelwal6 1Professor, Department Artificial Intelligence & Data Science, VIT Pune, Maharashtra, India
2,3,4,5,6 Students, Department of Artificial Intelligence & Data Science, VIT Pune, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper presents an advanced real-time flood
operate as a software-based simulator, using seed-based synthetic data generation for academic or research demonstrations, or as a hardware-integrated solution with real-time sensors connected via Arduino microcontrollers. This flexibility makes it suitable for both low-resource environments and real-world field applications in floodprone areas.
detection and forecasting system (RT-FDFS) that integrates machine learning, deep learning, and real-time weather data analysis for effective flood prediction. By combining Long Short-Term Memory (LSTM), Transformer networks, and XGBoost models, RT-FDFS is capable of processing sensorbased and API-fetched weather data to identify flood risks dynamically. It supports both software-based simulation and hardware-integrated deployment using Arduino sensors. The system also incorporates a seed-based data generation feature to simulate repeatable flood events for testing and demonstration purposes. A user-friendly interface provides real-time visualizations and color-coded flood alerts, enhancing situational awareness and emergency preparedness. The system proves to be robust, scalable, and adaptable for flood-prone regions.
Furthermore, the system integrates with the OpenWeatherMap API, allowing it to ingest up-to-date meteorological data. This real-time capability is vital for issuing early warnings and supporting rapid decisionmaking during emergencies.
2. LITERATURE REVIEW Sharma et al. [1] explored the integration of deep learning models for hydrological forecasting, demonstrating the improved accuracy of LSTM over traditional statistical methods. Kumar and Rathi [2] introduced a hybrid neural network combining GRU and CNN layers for spatiotemporal flood pattern recognition, achieving notable performance in coastal regions.
Key Words: Flood Forecasting, Deep Learning, IoT, Realtime Simulation, Disaster Management.
1.INTRODUCTION Floods remain one of the most devastating and frequent natural disasters worldwide, resulting in significant loss of life, economic damage, and disruption of critical infrastructure. The increasing unpredictability of extreme weather events, driven by climate change and rapid urbanization, has intensified the need for accurate, real-time, and intelligent flood monitoring systems. Traditional flood forecasting approaches often rely on rule-based models or static statistical methods, which fail to capture the nonlinear and dynamic nature of flood phenomena, especially in complex terrains and urban environments.
Patel et al. [3] proposed a real-time weather data analysis system using Transformer-based models for rainfall prediction, providing critical insights for early flood alerts. Das and Mehra [4] developed a decision support system using XGBoost for classifying flood severity based on environmental parameters and satellite imagery. Roy and Prasad [5] designed an Arduino-based IoT sensor suite to monitor river discharge and rainfall, emphasizing the importance of low-cost sensor integration. Similarly, Chandra et al. [6] demonstrated the use of LoRa-based communication modules in rural flood detection systems, ensuring long-range and energy-efficient data transmission. Rao et al. [7] investigated the use of cloud computing to manage large-scale flood-related sensor data, enabling scalable processing pipelines for real-time risk assessment. Mehta and Srivastava [8] proposed a hybrid forecasting model combining SVM and decision trees, highlighting its robustness in non-linear environments.
The Real-Time Flood Detection and Forecasting System (RTFDFS) aims to address these challenges by combining the power of deep learning, machine learning, and real-time environmental data into a unified framework. The system leverages advanced models such as Long Short-Term Memory (LSTM) networks, Transformer-based architectures, and XGBoost classifiers to learn intricate temporal and spatial patterns from historical and live data. These models are trained to process time-series inputs like rainfall, temperature, humidity, water levels, soil moisture, and flow rate to deliver dynamic flood risk predictions with high accuracy and low latency. One of the key strengths of RT-FDFS is its dual-mode deployment capability. It can
© 2025, IRJET
|
Impact Factor value: 8.315
Verma et al. [9] explored a dashboard-based UI for visualizing flood risk in urban areas using interactive GIS overlays. Bansal and Kapoor [10] integrated satellite image segmentation with LSTM models for predicting flash floods
|
ISO 9001:2008 Certified Journal
|
Page 203