Google AI Expands Flood Forecast Initiative in India
India is largely affected by floods every year. To have a beforehand knowledge about the possibility of floods, Google AI announced the up-scale of the Flood Forecast Initiative towards India as well as Bangladesh.
With Artificial Intelligence being an active influence and dominant constituent in almost all domains, calamity prediction was not to be disregarded. Google initially launched the Flood Forecast Initiative in India in 2018, restricted to only a few parts of Bihar.
However, very recently, they announced the expansion of this AI-driven tool to all parts of India and Bangladesh, making it accessible to over 200 million people across 250,000 square kilometers in India alone. It uses machine learning algorithms and tools to predict the possibility of floods in any region and can alert the local authorities and citizens to aid early evacuation or necessary measures needed to be taken.
Besides being a partner to the Indian Central Water Commission (CWC), Google also joined forces with Bangladesh Water Development Board, helping over 40 million people. Until very recently, CWC raised an alert for floods one day before the day it was predicted to happen, but with Google AI, these alerts can be notified up to three days before the disaster.
According to studies by Google with Yale, 65% of people who were notified before the hour took measures to protect themselves and secure their belongings.
Accurate and effective flood forecasting might be helpful for local authorities to alert the people residing in surrounding areas and take precautionary measures to save themselves from a major loss.
The multinational tech-giant claims they have now sent over 30 million flood forecast notifications to Android users.
Besides focusing on expansion, Google is also aiming to get more accurate forecasts and the mediums of how locals receive notifications.
The company said that they aim to build local networks so that updates can be made easily available to all those as well who might not receive them the conventional way, which is through smartphones.
Unlike the majority of Google’s models that use data from various users for that platform, this model formulates the historical data for rain and water level in rivers. Google AI is using Artificial Intelligence, Machine Learning, analysis of hydrological observation of data, and geospatial mapping to provide location-specific flood warning systems. It works on a very elaborate yet simple machine learning model involving the creation of the Inundation model which inputs a forecast of the level of water in a river and stimulates the water behavior across that floodplain. This results in highly spatially accurate predictions of the risk map. These maps help identify what areas may be flooded and what areas may be safe. Creation of the Inundation model depends on the following major components:
1. Real-Time Water Level Management
To efficiently run these models, it is essential to have access to steady information regarding the water levels in real-time which is aided by the local government.
2. Elevation map creation
The map of the terrain is equally important as knowing the water levels of the river. For this, high-resolution digital elevation models are used. Processing begins by collecting a large variety of satellite images from Google Maps and aligning these images while optimizing for satellite camera corrections and coarse terrains, these images are correlated. The corrected camera models are used to create depth maps for each image that are fused together at each instance of location. Finally, all objects such as trees, bridges, etc. are removed.
3. Hydraulic modeling
The modeling of the actual prediction model takes place in two steps. First, the physics-based hydraulic computations are performed which updates the location and velocity of water through time. And the incorporation of historical measurement-based predictive inundation models. Based on the dataset by SAR (Synthetic Aperture Radar) imagery, historical water level measurements are correlated with historical inundations, enabling the identification of consistent corrections to the hydraulic models. The model provides information about the areas that will be affected by floods and the areas that will be safe. It also determines the depth of the flood, that is when and how much are the floodwaters likely to rise.
In areas where successful depth maps can be created throughout the floodplain, the tool shares updates with the users surrounding those areas.
The look of alerts and their function to ensure maximum reach has also been modified. Information is now represented in multiple formats such that it is readable as well as visually representable besides supporting multiple languages like Hindi, Bengali, and 7 other languages that are locally accepted.