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

Floods are among the most devastating natural disasters, causing significant damage to lives, property, infrastructure, and the environment worldwide. Bangladesh is particularly vulnerable to flooding due to its unique geographical and topographical characteristics, as well as its heavy reliance on the monsoon season for agriculture and water resources.

The Problem

Flooding in Bangladesh has severe consequences for the nation and its people. The inundation of vast areas leads to the loss of lives, displacement of millions, and extensive damage to infrastructure, agriculture, and the economy. Homes and businesses are destroyed, crops are ruined, and transportation networks are disrupted. The aftermath of floods also poses significant public health risks, with the spread of waterborne diseases and the contamination of water sources. The recurring nature of floods in Bangladesh hinders sustainable development and perpetuates a cycle of poverty and vulnerability for affected communities.

The Background

Several factors exacerbate Bangladesh’s vulnerability to flooding:

Geographical Location

Bangladesh’s position in the low-lying Ganges-Brahmaputra-Meghna Delta makes it inherently prone to flooding. The country’s flat terrain, with over 80% of the land area less than 10 meters above sea level, offers little natural protection against rising water levels.

Monsoon Season

The monsoon season brings heavy rainfall, which, coupled with the melting of Himalayan glaciers, leads to a massive influx of water into the country’s river systems. This seasonal influx of water often overwhelms the country’s drainage capacity, resulting in widespread flooding.

Climate Change

Climate change has further intensified the frequency and severity of extreme weather events, including floods, in Bangladesh. Rising sea levels, increased glacial melt, and more erratic rainfall patterns due to climate change amplify the risk of flooding.

Deforestation and Land Use Changes

Deforestation and land use changes in the upstream regions of Bangladesh’s rivers have reduced the land’s natural ability to absorb and retain water. This leads to increased surface runoff and faster flow of water into the river systems, heightening the risk of flooding downstream.

The Goal

The Omdena Bangladesh Local Chapter team saw an opportunity to improve flood forecasting and management by harnessing recent technological advances. They aimed to develop a flood prediction model that could enhance disaster preparedness and resource allocation by utilizing rainfall time series and GIS data. The project’s main goal was to provide accurate flood forecasts to help authorities better manage disasters. The team planned a comprehensive workflow, which included collecting data, conducting exploratory analysis, training the model, and deploying it. Ultimately, they wanted to create a working prototype of an alert system that could provide actionable insights to disaster management officials.

Our Approach

Data Collection

Historical Data

Websites of meteorological agencies, such as the Bangladesh Climate Data Portal, along with open-source repositories i.e. Kaggle, were programmatically scraped or manually downloaded to gather historical data on river discharge and rainfall across Bangladesh’s 10 divisions. This information was systematically compiled and stored in a CSV format.

GIS Data

Leveraging the capabilities of the Google Earth Engine platform, geospatial data covering all divisions of Bangladesh were acquired. This encompassed critical factors such as Elevation, Slope, Curvature, Land Use Land Cover, Soil Types, and Land Surface Temperature, among others. The datasets were downloaded in a Raster format.

Exploratory Data Analysis

Historical Data

In the analysis of flooding patterns across districts in Bangladesh, Chittagong and Barisal experience the highest levels of flooding, as demonstrated in the graph below. The bar chart illustrates the mean daily precipitation, detailing average rainfall throughout the year, during rainy periods, and specifically during the monsoon season by division.

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.

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.

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.

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.

Model Development

Flood Prediction Model using Historical Data

We prepared a balanced dataset of historical flood events from 2003-2023, including topological features (e.g., Digital Elevation Model) and weather/meteorological features (e.g., precipitation, wind direction). To capture both static and dynamic aspects, we developed two separate models:

1. Spatial & Weather Condition Model: Using Random Forest, we focused on high recall and Cohen Kappa Score. Land features, especially elevation, effectively predicted flood locations.
2. Temporal Model: Using XGBoost and only weather data, this model precisely predicted floods with balanced recall and exceptional AUC (0.99). Time of year and rainfall patterns were key success factors.

While these models enhance disaster management capabilities, we recognized the need to refine our data collection strategy for more nuanced identification of flood-prone areas based on historical data before deployment.

Flood Susceptibility Model using GIS Data

To prepare the GIS data, a Flood Inventory of 2,766 sample points (1,408 flood points and 1,358 non-flood points) was used, forming a binary classification problem. 11 flood conditioning factors were extracted for each point using GIS software.

The dataset was split into an 80% training set to train a Random Forest classifier, and a 20% test set.

The model excellently predicted which areas were likely to flood. On unseen test data, it correctly identified 90% of both flooded and non-flooded areas, performing much better than random chance. The results show the model is reliable for assessing flood risk.

Model Deployment

The finalized models were then integrated into a Streamlit application and was enhanced to include important information for users to comprehend.

Flood Prediction Model using Historical Data

After seeing how important rainfall is in causing floods, we decided to focus on predicting the average daily rainfall for selected cities in Bangladesh. By forecasting rainfall, we hope to help people better prepare for potential floods. We built different AI models and fine-tuned each one using past weather data from each city to get the best possible predictions.

For example, one of our models was able to predict a single day’s rainfall with an average error of only about 2.8 mm. This shows how close the model’s predictions are to the actual rainfall on average. Also, the model’s predictions matched the real data quite well, with a 73% correlation. We saw similarly promising results for other cities too.

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 has made significant progress in improving flood prediction and management in Bangladesh. Some of the project’s key achievements are:

Comprehensive Data Collection

Successfully gathered and integrated historical rainfall data and GIS data from various sources, creating a robust dataset for flood prediction and susceptibility analysis.

Insightful Analysis

Conducted in-depth analysis of historical rainfall patterns, land cover distribution, soil composition, and temperature trends, providing valuable insights into factors influencing flood risk in Bangladesh.

Accurate Flood Prediction Models

Developed highly effective models for flood prediction using historical data, achieving exceptional performance metrics such as high recall, balanced accuracy, and impressive AUC scores.

Reliable Flood Susceptibility Mapping

Created a reliable flood susceptibility model using GIS data and Random Forest classifier, accurately identifying flood-prone areas with a 90% success rate on unseen test data.

Interactive Web Application

Deployed the flood prediction and susceptibility models on a user-friendly Streamlit application, enabling users to explore rainfall forecasts and interactive flood susceptibility maps for better decision-making.

Region-Specific Insights

Identified high-risk regions like Khulna Division and comparatively safer areas like Dhaka Division, emphasizing the need for tailored flood mitigation and adaptation strategies based on regional vulnerabilities.

Potential Next Steps

While the FloodGuard project has made significant strides in flood prediction and susceptibility mapping in Bangladesh, there are several areas for future improvement to enhance the system’s effectiveness and impact:

Enhancing Data Resolution

Increasing the spatial and temporal resolution of the collected data, particularly for rainfall and GIS datasets, to enable more precise flood predictions and susceptibility mapping at a finer scale.

Incorporating Real-Time Data

Integrating real-time data streams, such as live rainfall measurements, river water levels, and satellite imagery, into the models to provide up-to-date flood forecasts and enable early warning systems.

Expanding Data Sources

Exploring additional data sources, such as socio-economic indicators, population density, and infrastructure data, to assess the potential impact of floods on communities and guide targeted mitigation efforts.

Integrating with Hydrological Models

Combining the developed models with hydrological models that simulate water flow and inundation to provide more comprehensive flood risk assessments and support flood management decision-making.

Developing Early Warning Systems

Utilizing the flood prediction models to develop and implement early warning systems that provide timely alerts to communities, enabling them to take proactive measures to protect lives and minimize damage.

Potential Applications in Other Industries

The flood prediction models and susceptibility mapping methodology developed in the FloodGuard project have the potential to be applied in various industries and sectors beyond disaster management. Some of the potential applications include:

Agriculture

The flood prediction models can help farmers and agricultural organizations make informed decisions about crop planting, harvesting, and irrigation schedules based on the likelihood of flooding in their areas. This can help minimize crop losses and optimize agricultural productivity.

Insurance

Insurance companies can leverage the flood susceptibility maps to assess the risk of flooding in different regions and adjust their premiums and coverage accordingly. This can help them manage their risk exposure and provide more accurate pricing for flood insurance policies.

Urban Planning

Urban planners and developers can use the flood susceptibility maps to identify areas that are less prone to flooding when planning new infrastructure projects, such as housing developments, industrial parks, and transportation networks. This can help ensure that new developments are built in safer locations and are more resilient to flooding.

Water Resource Management

The flood prediction models can be used by water resource management authorities to optimize the operation of dams, reservoirs, and other water control structures. By anticipating potential flood events, they can make informed decisions about water release and storage to minimize downstream flooding and ensure adequate water supply.

Emergency Response

Emergency response organizations can use the flood prediction models and susceptibility maps to develop more effective disaster response plans. By identifying high-risk areas and understanding the likelihood of flooding, they can pre-position resources, establish evacuation routes, and coordinate relief efforts more efficiently.

Climate Change Adaptation

The FloodGuard methodology can be applied to assess the potential impact of climate change on flood risk in Bangladesh and other regions. By incorporating climate change projections into the models, policymakers and planners can develop long-term adaptation strategies to build resilience against future flood risks.

Want to work with us too?

Let’s see if we are a good fit