Microclimate Vision

Overview

Repository for the code used for multimodal prediction of climatic parameters using street-level and satellite imagery, developed by the Urban Analytics Lab (UAL) at the National University of Singapore (NUS) in collaboration with Takenaka Corporation.

You can read more about this project on its website too.

The journal paper can be found here.

The task of the model is predicting microclimate data at a target location based on microclimate data at a reference location and imagery inputs.

This model consists of LSTM and ResNet-18 architectures, and predicts air temperature $T_{air}$, relative humidity $RH$, wind speed $\nu$, and global horizontal irradiance $GHI$.

Installation

PyTorch and torchvision

Since Microclimate Vision uses pytorch and torchvision, you may need to install them separately. Please refer to the official website for installation instructions.

Dependencies

conda/mamba env -f environment.yml

The environment is named microclimate-vision, to navigate to it, use:

conda/mamba activate microclimate-vision

Usage

Data preparation

You need to create a dataset integrating microclimate data with street-level and satellite imagery. Please refer to the sample files in the "data" directory. Note that these sample files do not contain observed real data, but rather virtual data created to demonstrate the proper data structure.

Config file

The detailed settings, e.g., hyperparameters of the model and paths to dataset files, are specified in a config file. Please refer to 'configs/sample.yaml'.

Training

python train.py --config path/to/config

Test

python test.py --config path/to/config --model path/to/model --result path/to/result

License

Microclimate Vision was created by Kunihiko Fujiwara. It is licensed under the terms of the CC BY-SA 4.0.

Citation

Please cite the paper if you use Microclimate Vision in a scientific publication:

Fujiwara, K., Khomiakov, M., Yap, W., Ignatius, M., & Biljecki, F. (2024). Microclimate Vision: Multimodal prediction of climatic parameters using street-level and satellite imagery. Sustainable Cities and Society, 105733. doi:10.1016/j.scs.2024.105733

@article{2024_scs_microclimate_vision,
 author = {Fujiwara, Kunihiko and Khomiakov, Maxim and Yap, Winston and Ignatius, Marcel and Biljecki, Filip},
 doi = {10.1016/j.scs.2024.105733},
 journal = {Sustainable Cities and Society},
 pages = {105733},
 title = {Microclimate Vision: Multimodal prediction of climatic parameters using street-level and satellite imagery},
 volume = {114},
 year = {2024}
}

Credits

• Creation of environment.yml: Desai Wang

---