Workflow 1
Workflow 2
Web app GUI

Inspiration

Urban green areas are very important for urban climate. Local authorities and citizens should have an overview of the development of the urban green areas in their city.

What it does

The goal is to collect and analyze data on urban green spaces and visualize it in an appealing way with interactive maps and graphics so that the resulting information is easy to understand. The methods and outcomes of this project shall be transferable to other cities. This is ensured by a high level of automation of the data processing and the usage of global avaliable datasets.

How we built it

Using AWS SageMaker Studio Labs, Python Processing, ASDI Datasets, Streamlit web app

Challenges we ran into

RAM resources during numpy related computations - maybe Dask can help here
methods development take most of the time

Accomplishments that we're proud of

applied ML classification on global available datasets (training datasets, too!) make the idea transferable to other cities

What we learned

rasterio WarpedVRT is really the easy way to access COGs!
rasterio's reshape_as_image and rehape_as_raster are really handy to reshape numpy arrays to ML algorithm input
Bottleneck is a more efficient cython based drop in replacement for standard numpy computations
fast-glcm is more easy to compute GLCM features than in scikit image
GLCM for textures maybe an important feature for prediction