FloodGuard: Harnessing the Power of AI and GIS to Protect Bangladesh from the Fury of Floods

Omdena’s FloodGuard project uses AI and GIS to predict and map floods in Bangladesh with high accuracy (AUC 0.99). The system empowers authorities and communities to prepare early, reduce damage, and build long-term climate resilience.

The Problem

Flooding in Bangladesh brings devastating consequences every year. It destroys homes, displaces millions of people, and causes severe damage to infrastructure, agriculture, and the economy. When vast areas are submerged, lives and livelihoods are lost.

The impact doesn’t end when the water recedes. Floods also trigger public health crises by spreading waterborne diseases and contaminating drinking water. As a result, affected communities face both immediate and long-term challenges.

In addition, the destruction of crops and transport networks disrupts local markets and weakens the economy. Businesses shut down, and families lose their sources of income.

Because floods occur so frequently, recovery efforts are often short-lived. Communities are forced into a cycle of loss and rebuilding, which slows down progress and hinders sustainable development.

The Background

Several factors exacerbate Bangladesh’s vulnerability to flooding:

Flooding in Bangladesh happens for several reasons. A combination of geography, weather, and human activity makes the country extremely vulnerable. Understanding these factors is key to creating better flood prevention strategies.

Geographical Location

Bangladesh lies in the Ganges-Brahmaputra-Meghna Delta, one of the largest river deltas in the world. Because the terrain is mostly flat, rising water levels quickly spread across wide areas. More than 80% of the land sits below 10 meters above sea level, leaving little natural protection from floods.

Monsoon Season

During the monsoon season, heavy rainfall and melting glaciers from the Himalayas add huge amounts of water to the rivers. As a result, drainage systems often cannot handle the overflow. This leads to widespread flooding, especially in low-lying regions flooding.

Climate Change

Climate change makes these floods worse. Rising sea levels, unpredictable rainfall, and faster glacial melting all increase the frequency and intensity of flooding. Therefore, communities that were once safe are now becoming more exposed to extreme weather events.

Deforestation and Land Use Changes

Another factor is deforestation and poor land use in upstream areas. When forests are cleared, the soil loses its ability to absorb water. This creates more surface runoff and faster river swelling downstream. Consequently, flooding becomes more intense and harder to control.

The Goal

The Omdena Bangladesh Local Chapter set out to improve flood forecasting and management using AI and GIS technologies. The team’s main goal was to create a flood prediction model that could help authorities prepare better and allocate resources more effectively.

They wanted the model to deliver accurate and actionable flood forecasts, giving disaster management teams the information they need to protect communities. To achieve this, the team designed a clear and practical workflow. It included data collection, exploratory analysis, model training, and deployment.

Ultimately, the project aimed to build a working prototype of an early warning system. This system would alert disaster management officials in advance, helping them take quick, informed action to reduce damage and save lives.

Our Approach

1. Data Collection

The team began by gathering both historical and geospatial (GIS) data to build a strong foundation for the flood prediction models. This data would help identify flood patterns, regional risks, and environmental conditions that influence flooding in Bangladesh.

Historical Data

The team collected historical information from trusted sources such as the Bangladesh Climate Data Portal and open repositories like Kaggle. These datasets included rainfall and river discharge records from all ten divisions of Bangladesh.

Data was either programmatically scraped or manually downloaded and then organized in CSV format for easy access. This systematic approach allowed the team to compile accurate, consistent data for model training.

GIS Data

Next, the team used Google Earth Engine to collect detailed geospatial data for all divisions of the country. This data covered key environmental features such as elevation, slope, curvature, land use, soil type, and land surface temperature.

All datasets were downloaded in raster format to ensure compatibility with GIS analysis tools. This combination of historical and geospatial data created a solid foundation for predicting flood risks with precision.

2. Exploratory Data Analysis

The team performed a detailed exploratory data analysis (EDA) to understand flood patterns and environmental factors across Bangladesh. This process combined historical data and GIS-based insights to uncover trends that could improve the accuracy of the flood prediction models.

Historical Data

The analysis revealed that Chittagong and Barisal experience the most frequent and severe flooding. The team created visualizations showing mean daily precipitation across all divisions, highlighting rainfall levels throughout the year and during the monsoon season.

Distribution of Mean Daily Precipitations Across Divisions in Bangladesh

Complementing this, the below graph represents the annual mean rainfall from 1985 to 2016. Several fluctuations are evident over the years, with peaks corresponding to likely years of higher flood events. The variance in annual rainfall can be associated with climatic cycles and potentially with significant flood events, especially in districts with higher average rainfall.

Trends in Annual Mean Rainfall in Bangladesh from 1985 to 2016

GIS Data

Several GIS layers, including land surface, soil water content, sand content, and land surface temperature, were analyzed, yielding insightful findings. The below chart categorizes the land cover across Bangladesh, showing a predominance of water and flooded vegetation. This distribution is critical for understanding flood behavior, as areas with significant water bodies and flooded vegetation are likely more susceptible to flooding events.

Distribution of Land Cover Types in Bangladesh

The team also gathered data on the average amount of water held in the soil and sand across different regions of Bangladesh. This information provides valuable insights into the composition of the soil and its ability to absorb and retain water, which are important factors in assessing the risk of flooding. Areas with soil that holds more water may be more likely to experience flooding, as the water cannot easily drain away. On the other hand, regions with high sand content might have better drainage, potentially reducing the risk of floods.

Average Soil Water Content

Average Soil Sand Content

Lastly, the team collected data on temperature patterns across Bangladesh. Temperature can affect how quickly water evaporates from the land and returns to the atmosphere through a process called evapotranspiration. By studying these temperature patterns, the team can better predict how much water is likely to be in the soil at different times. This information helps them understand which areas might be more prone to flooding when heavy rains occur.

Land Surface Temperature Distribution in Bangladesh

Model Development 1

The team developed two main types of models to predict and assess flood risks across Bangladesh. These were based on historical data and GIS data to capture both temporal and spatial patterns.

Flood Prediction Model using Historical Data

To create the first model, the team prepared a balanced dataset covering flood events between 2003 and 2023. The dataset included topological features such as elevation and meteorological data like rainfall and wind direction.

To capture both static and dynamic aspects of flooding, the team built two complementary models:

• Spatial and Weather Condition Model (Random Forest):
  This model focused on land and weather features. It achieved high recall and a strong Cohen Kappa Score, showing that elevation and terrain characteristics are effective predictors of flood-prone areas.

• Temporal Model (XGBoost):
  This model used only weather data and produced excellent results, including a balanced recall and an impressive AUC score of 0.99. Seasonal patterns and rainfall intensity proved to be key factors in its accuracy.

Together, these models improved disaster management capabilities by offering reliable flood forecasts. However, the team also recognized that refining the data collection process would allow for even more precise identification of flood-prone regions.

Flood Susceptibility Model using GIS Data

The team also built a Flood Susceptibility Model using geospatial data. They created a Flood Inventory with 2,766 data points, which included 1,408 flood points and 1,358 non-flood points. This dataset represented a binary classification problem.

Each point contained 11 flood-conditioning factors extracted through GIS tools. The dataset was divided into an 80% training set and a 20% testing set to evaluate performance.

Using a Random Forest classifier, the model accurately predicted 90% of flooded and non-flooded areas on unseen data. The results proved that this approach is both robust and effective for assessing flood susceptibility across large regions.

Model Deployment 2

After building and testing the models, the team integrated them into a Streamlit web application. This platform allowed users to easily visualize flood predictions, explore rainfall forecasts, and view susceptibility maps in an interactive format.

Flood Prediction Model using Historical Data

Recognizing that rainfall is a key factor in flooding, the team focused on predicting average daily rainfall for several major cities in Bangladesh. By forecasting rainfall patterns, users can take early precautions to reduce the impact of floods.

To achieve this, the team built multiple AI models and fine-tuned them with historical weather data from each city. The results were highly promising. One model achieved an average prediction error of just 2.8 mm and a 73% correlation between predicted and actual rainfall values.

Other cities showed similarly accurate results, demonstrating that the model can provide reliable short-term rainfall forecasts. These forecasts give communities and disaster management authorities valuable time to prepare and respond effectively.

Flood Susceptibility Model using GIS Data

The team took the results from the Random Forest model and turned them into a map that people can easily understand. They used a technique called Inverse Distance Weighting (IDW) to estimate flood risk between the specific points the model analyzed. This created a smooth, continuous map surface showing flood susceptibility across the whole country.

The team then used a tool called leafmap to make the map interactive on a web application. Users can explore which areas of Bangladesh are most likely to flood. They can either look at the entire country or zoom in on a specific region, as shown in the image below.

Bangladesh’s flood susceptibility map shows that the Khulna Division, particularly near the coastline, has a high risk of flooding exceeding 50%. This finding is consistent with scientific studies that have identified the Satkhira district within Khulna as one of the most flood-prone regions in the country. On the other hand, the Dhaka Division seems to be the least susceptible to flooding, possibly due to its inland location and higher elevation, which provide some inherent protection against floods.

However, while Dhaka may be less vulnerable to large-scale natural flooding, urban flood risks caused by infrastructural challenges remain a concern. This analysis highlights the varying flood risks across regions and emphasizes the importance of developing region-specific flood mitigation and adaptation strategies to protect vulnerable communities and infrastructure.

Key Achievements

The FloodGuard project achieved remarkable progress in improving flood prediction and management in Bangladesh. Each milestone contributed to building a reliable and accessible flood monitoring system that can guide disaster preparedness and decision-making.

Comprehensive Data Collection

The team successfully gathered and combined historical rainfall and GIS data from multiple trusted sources. This integration created a robust dataset that serves as the backbone of the project’s flood prediction and susceptibility analysis.

Insightful Analysis

Through in-depth data exploration, the team identified key insights into rainfall patterns, land cover, soil composition, and temperature trends. These findings helped pinpoint the environmental and climatic factors most strongly linked to flood risk in Bangladesh

Accurate Flood Prediction Models

The AI models achieved high recall, balanced accuracy, and strong AUC scores. These results demonstrate that data-driven approaches can effectively predict floods, giving authorities more time to act and protect vulnerable communities AUC scores.

Reliable Flood Susceptibility Mapping

Using GIS data and a Random Forest classifier, the team created a reliable flood susceptibility model that accurately identified flood-prone areas with 90% accuracy on unseen test data. This model provides a valuable foundation for future regional risk mapping.

Interactive Web Application

The project culminated in a Streamlit-based web application that brings flood data to life. Users can view rainfall forecasts and explore interactive flood susceptibility maps, helping them make informed decisions before and during flood events.

Region-Specific Insights

The analysis revealed high-risk regions such as the Khulna Division and relatively safer zones like the Dhaka Division. These insights stress the importance of region-specific flood mitigation and adaptation strategies to reduce disaster impact and strengthen resilience.

Potential Next Steps

While the FloodGuard project achieved major milestones, there are still opportunities to make it even more powerful and effective. The next phase will focus on refining data, improving model accuracy, and expanding real-world usability.

Enhancing Data Resolution

Increasing the spatial and temporal resolution of rainfall and GIS datasets will make flood predictions more precise. With finer data, the model can provide localized forecasts that help communities prepare for specific, high-risk areas rather than general regions

Incorporating Real-Time Data

Integrating real-time information—such as live rainfall, river water levels, and satellite imagery—will keep forecasts current and relevant. As a result, the models can provide up-to-date alerts and power effective early warning systems.

Expanding Data Sources

Adding more data types, including population density, socio-economic indicators, and infrastructure data, will give a clearer picture of how floods impact communities. These additions can help authorities design targeted mitigation and relief strategies.

Integrating with Hydrological Models

Combining FloodGuard’s predictive models with hydrological simulations can improve the accuracy of water flow and inundation forecasts. This integration will enable comprehensive flood risk assessments and enhance planning for disaster management teams.

Developing Early Warning Systems

The ultimate goal is to create an automated early warning system that sends timely alerts to vulnerable communities. With early notifications, people can take preventive actions, reduce losses, and improve safety before floods strike. Many mission-driven groups already lead this kind of community resilience work—see the top non-profit organizations working for the environment making measurable impact worldwide.

Potential Applications in Other Industries

The technology developed in the FloodGuard project has potential far beyond disaster management. Its flood prediction models and mapping methods can help several industries improve planning, reduce risks, and make more data-driven decisions.

Agriculture

Farmers can use flood forecasts to plan planting, harvesting, and irrigation schedules more effectively. By understanding when and where floods are likely to occur, they can protect crops, reduce losses, and boost productivity. As a result, the agricultural sector can become more resilient to extreme weather conditions.

Insurance

Insurance companies can apply flood susceptibility maps to assess risk more accurately. These insights allow insurers to adjust premiums and coverage based on real data. This leads to fairer pricing, better risk management, and improved support for clients in flood-prone regions.

Urban Planning

City planners and developers can use FloodGuard’s maps to identify safer zones for infrastructure projects such as housing, roads, and industrial areas. In addition, these tools help prevent construction in high-risk flood zones, saving both resources and lives in the long run.

Water Resource Management

Authorities responsible for dams, reservoirs, and drainage systems can use the models to anticipate heavy rainfall and rising water levels. With this foresight, they can make informed decisions about water release and storage, reducing flood risk downstream while ensuring steady water supply.

Emergency Response

Disaster response organizations can rely on FloodGuard’s prediction tools to create better evacuation plans and allocate resources effectively. Early knowledge of high-risk zones enables faster, more targeted rescue and relief efforts, saving lives when every second counts.

Climate Change Adaptation

The FloodGuard framework can also help policymakers and researchers assess how climate change affects flood risk over time. By including future climate projections, the models can guide long-term adaptation strategies that build resilience against rising sea levels and extreme weather events.

Want to work with us too?

Let’s see if we are a good fit