Awards & Nominations
AeroSparx has received the following awards and nominations. Way to go!
Automated Detection of Hazards
Automated Detection of Hazards: Machine Learning and algorithmic approaches to detect Floods, Fires
Summary
Natural phenomena like flood, fire, and algae blooms are threats to the ecosystems and hamper the chain of the life cycle, making its timely detection and categorization an important factor for mitigation and recovery. Our aim with this effort is to detect and characterize the mentioned hazards, by which researchers and decision-makers can better understand their impacts and scope. We have carried out case studies associated with these phenomena and used satellite imagery and machine learning approaches to classify threshold values that define the boundary between flooded and non-flooded, severity fire, algae bloom classes quickly and accurately. This well-formed approach can reduce the time
How We Addressed This Challenge
We developed a ML Based Web Application where we did various case studies of Floods, Fires and Algae blooms. In this web Application, once the sentinel 2 images of any area of interest is fed it uses the powerful machine learning algorithm to detect the natural phenomenon in the satellite imagery.
This detection is important because we can highlight the effect of these natural phenomena over the vegetation, flora, fauna, human population etc and help us take measures to protect these phenomena causing any impacts over them.
We have made a flask based web application which has an interface where users can choose any of 3 phenomena like Flood, Fire and Algae Bloom. Once these phenomena are selected the user is redirected to a dashboard where he can choose any one of the 3 case studies for which we have done analysis. The Rest API then passes the request parameter as the location selected which returns the sentinel class of the image which has values ranging from 1 to 5 based on the severity of the particular disaster. The Matplotlib library is then used to plot a map for this sentinel class.
In future we can leverage our application to fetch the satellite imagery from GoogleEarthEngine for any particular location and process the data to give output for the same. We can also leverage our machine learning models in predicting disasters in any area beforehand.
How We Developed This Project
Abstract :
Floods and fires are the most frequent natural disasters occurring all over the globe. These are not only affecting human life but also causing severe damage to the environment. Especially the harmful algae blooms that cause more damage to aquatic life. So we have chosen flood, fire and Harmful Algae Bloom detection in this challenge.
Approach
Flood:
- Initially, we have used the Green Band (B3), Near Infrared (NIR) Band (B8) of the data, before and after flood satellite images gathered using Sentinel-2 to calculate the Normalized Difference Water Index (NDWI) which differentiates the water body and dry land from satellite images.
- Then we have generated ground truth for the satellite images using the NDWI of pre-flood and post-flood.
- Using the ground truth generated from the NDWI, we have developed the machine learning model to classify the flood area and non-flood area.
- Fine-tuning of the developed machine learning model is done using the grid search method.
Fire:
- We have used the Near Infrared (NIR) Band (B8A) and the Shortwave Infrared (SWIR) Band (B12) from Sentinel-2 satellite imagery, both pre-fire and post-fire for calculating the Difference Normalized Burn Ratio (DNBR).
- The basis of this approach is the fact that natural vegetation affected by fires reflect characteristically in the above mentioned bands of the spectrum.
- The DNBR is then used to categorize the focus area into various severity levels.
- The categorized data is then used to train and develop a machine learning model, which classifies the test area into the aforementioned levels.
- Fine-tuning is done using the grid search method to optimize the performance of the model.
Harmful Algae Blooms:
- Initially we have used the sentinel 2 satellite imagery and segregated the bands and read its values for calculating to NDWI using green band (b3) and NIR band (Band 8). We have classified the water bodies from the image.
- Post classifying the water bodies, we have used following algorithm
- Al10SABI - Measures chlorophyll - Blue Band(B2), Green Band(B3), Red Band(B4) and NIR Band(B8 or B8A)
- TurbBow06RedOverGreen - Measures Turbidity - Red Band(B3) and Green Band(B3)
- Be16FLHGreenRedNIR - Measures BlueGreenAlgae/PycoCynin- Red Band(B4), Green Band(B3) and NIR Band(B8)
- Equal Weightage to all these 3 algorithms is given to compute the water quality index which is then normalized between 0 to 1 using min max scaling where 0 represents pure water and 1 represents highest concentration of Harmful Algae Bloom.
Tools and Technologies Used :
We have used a gradient boosting framework that uses tree-based algorithms to solve the problem ie., detection of flooded areas. We used python as the coding language, QGIS, EarthPy, LightGBM, Flask, RasterIO, GDAL software’s are used to develop a Machine learning model that detects the flooded area and non-flooded area.
Problems and Achievements
The image consisted of Atmospheric imbalance and we had to use various algorithms and tools to do its atmospheric corrections. Once the preprocessing using QGIS and ArcGis was done we were able to achieve the accuracy of almost 96-99% for various floods, fires and Algae Bloom Detection.
How We Used Space Agency Data in This Project
We have used Copernicus Sentinel-2 data in our project, which has been developed and operated by the European Space Agency (ESA). This mission comprises a constellation of two polar-orbiting satellites placed in the same sun-synchronous orbit, phased at 180° to each other. It aims at monitoring variability in land surface conditions, and its wide swath width (290 km) and high revisit time (10 days at the equator with one satellite, and 5 days with 2 satellites under cloud-free conditions which result in 2-3 days at mid-latitudes), which support monitoring of Earth's surface changes.
Apart from this, we have also used the GitHub link provided by https://github.com/nasa/spaceapps-phenomena_detection/tree/dev/data/labeled to use the data for processing fires and Algae Blooms in our algorithm.