CLIM4cities is a European Space Agency (ESA)-funded project under the call for Artifical Intelligence (AI) Trustworthy Applications for Climate (ESA Contract No. 4000143628/24/I-DT - AI TRUSTWORTHY APPLICATIONS FOR CLIMATE). It aims to pioneer the development of Machine Learning (ML) and Artificial Intelligence (AI) models designed to downscale air and land surface temperature predictions in urban areas. This initiative serves as a preliminary step towards the implementation of cost-effective Integrated Urban Climate and Weather components into local Digital Twin Systems.
By leveraging crowdsourced data obtained from citizens owned weather stations, Earth Observation and weather forecasting models, we offer spatio-temporal data fusion models that can solve the unmet need for a low-cost, efficient and scalable Urban Climate prediction system. To achieve this, CLIM4cities has tailored its solution to the requirements of local early adopters, who state the need for tools that offers both early warning weather forecast capabilities, as well as scenario-making capabilities to evaluate climate adaptation measures, namely the impact of blue-green infrastructures on the Urban Heat Island effect.
This repository holds a demonstration code of the air temperature downscaling model developed during this project for four Danish Functional Urban Areas.
The full description of the model development and validation can be found at the preprint of the article: Downscaling Urban Thermal Signals with Machine Learning: A Case Study of Danish Functional Urban Areas
The model application for Urban Heat Island quantification was also described at the paper: Added Value for Urban Heat Island Quantification from Machine Learning Downscaling of Air Temperatures
- Install the Climate Data Operators (CDO) package. On Ubuntu, you can do so by running:
sudo apt-get update
sudo apt-get install cdo
-
Install conda (if not already installed)
-
create a conda environment by one of the following methods:
conda env create -f environment.yml
or:
conda create --name clim4cities python=3.11
pip install -r requirements.txt
pip install -e.
Before running the model inference download the geospatial predictors and the UTS-D model using these links:
Use the inference.ipynb notebok at \notebooks to explore the model behaviour.
If you use this code or data, please cite our paper:
@article{castro6172920downscaling,
title={Downscaling Urban Thermal Signals with Machine Learning: A Case Study of Danish Functional Urban Areas},
author={Castro, Maria and Paix{\~a}o, Jo{\~a}o and Gir{\~a}o, In{\^e}s and Marques, Bruno and Magnus Koktvedgaard Zeitzen, Rune and Cunha, Rita and Fonteles, Caio and Thejll, Peter and S{\o}rup, Hjalte JD and Paletta, Quentin and others},
journal={Available at SSRN 6172920}
}
Ana Oliveira - ana.oliveira@colabatlantic.com
Maria Castro - maria.castro@colabatlantic.com