RiverStream
The Project
From 1970 to 2019, floods were the most frequent weather, climate, and water-related disasters, accounting for 44% of all disasters globally. However, flood prediction apps are available only in some countries, leaving many regions unprepared. To address this, we developed RiverStream, an app that can be used worldwide. It allows users to receive and contribute flood data, while AI predicts floods in useful time and notifies users when risk increases. A simulation feature visualizes flood scenarios, raising awareness and preparedness. RiverStream builds a collaborative network to strengthen global disaster response, promoting action, prevention, and education. We challenged ourselves with new technologies such as Flutter for app development, Firebase for implementing posting features, generative AI for creating simulations, and TensorFlow for training AI models. It was the first time for most of us to work with these technologies. We learned mainly through Udemy courses and overcame difficulties by consulting mentors and using ChatGPT when we were stuck. Building the AI model was especially a new experience. We researched which weather parameters would be effective for flood risk prediction and built a classification model using SPI, rainfall, precipitation intensity, and humidity as inputs. We collected 1,000 data points from WeatherUnderground (800 for training, 200 for testing) and trained the model to predict flood risk at three levels: low, medium, and high. This model was then integrated into our app, allowing users to see useful time, location-specific flood risk estimates based on the latest weather data.
