This project focuses on a critical aspect of floating solar installations: identifying artificial water bodies. Solar developers often prefer to build on artificial water bodies (e.g. hydro-power reservoirs, water treatment plants) as they typically have an easier time obtaining permission. To help find these artificial water bodies, I developed a machine learning model capable of classifying any given water body as artificial or natural.
- GeoPandas
- Shapely
- Rasterio
- Rioxarray
- Sat-search
- Matplotlib
I worked with two datasets, Water Bodies in Rhode Island and Sentinel-2 Data, to identify and visualize artificial water bodies in Rhode Island. Using a Jupyter Notebook, I created a script that leverages GeoPandas, Shapely, and other libraries to process and visualize geospatial data.
To identify artificial water bodies, I designed an innovative a shape-based approach. I calculated the rectangularity of each water body and considered it artificial if its area was more than 85% (perfect circle would have pi/4 about 78% overlap), of the area of its minimum bounding rectangle. After identifying artificial water bodies, I downloaded cropped Sentinel-2 images for the three largest and three smallest water bodies in Rhode Island.
There are several challenges when working with shape-based approaches for artificial lake identification, such as false positives and difficulties in data availability. To address these issues, I proposed alternative solutions such as supervised classification, clustering, and incorporating additional features. By continuously integrating diverse data sources and leveraging supervised classification systems, I can enhance data and progressively improve accuracy.
This project showcases my commitment to harnessing the power of data science and artificial intelligence to accelerate renewable energy adoption. By identifying artificial water bodies, I empower solar developers with the information they need to make informed decisions and contribute to a more sustainable future.