Users can upload their own photos and receive a real-time weather forecast for Copenhagen.

The app is trained on 1.236 photos taken on 55 different days from March 2020 to February 2021. All photos were taken on the Kalvebod Fælled (hence KAL-VEDDER) recreational area outside of Copenhagen (https://naturstyrelsen.dk/naturoplevelser/naturguider/kalvebod-faelled/). Kalvebod Fælled is part of Naturpark Amager, one of several nature parks in Denmark.

The context for taking the photos was that I went on long hikes during the lock-down and took photos, after one year this provided a unique dataset of the same geographical location taken at different seasons and temperatures, hence a rich and varied dataset well suited for Machine Learning.

There is large potential in weather forecasting based on image recognition, since it is faster than processing large input data sets in a meteorological model, even with supercomputers. Current applications of image-recognition based weather forecasting are primarily applied to 'macro information' meaning radar images of large regions showing cloud formations. It is more rare to have weather forecasting based on 'micro information' such as photos from the ground taken by private individuals. The potential is however huge, for instance in analyzing contextual data such as birds flying at low altitude as an indication that rain is expected. Applying user's own data can also increase engagement in outdoors or social activities, sports, concerts etcetera.

Objective: provide real-time temperature forecasting for Copenhagen based on end-user photos. Increase engagement and user activation on Kalvebod Fælled and more generally for visitors to Denmark's nature parks.

Built With

Share this project: