This repo includes notebooks used in the study of Deep Learning Approach for Predicting Urban Heat Island Effects with High-frequency Urban Sensing Data.

The flowchart of this reserch is shown below:

You can follow the guide to batch download of AoT data with waggle: https://github.com/waggle-sensor/waggle/blob/master/data/README.md

nodes.csv is csv file contains the information about AoT nodes

node_with_band.csv file contains the information about AoT nodes combined with the processed the band value from remote sensing data

sensors.csv file contaisn the AoT sensor information

offsets.csv and provenance.csv offset and provenance of the data

data.csv. --- can be downloaded with link above, data to large to store in github

Landsat data used in this reseach can be found in: https://earthexplorer.usgs.gov/

Please specify the timerange in the summer of 2018-2020, Chicago area, and Landsat 8.

If you don't want to run the whole thing and wish to have intermediate result to run the modelling part directly, please contact Fangzheng Lyu with flu8@illinois.edu

Those two step are conducted with waggle data tool at: https://github.com/waggle-sensor/data-tools

Specifically, data fitering are done with slice_data_range funciton and the data reduction are conducted with data-reduction-tool function

Landsat_processing.ipynb is used to do band extraction from remote sensing data

data_formatting.ipynb is used to do interpolation for remote sensing data

Missing_Data.ipynb is used to fill in missing value with miceforest: Fast Imputation with Random Forests method.

Regression.ipynb

SVM.ipynb

ann_MAE.ipynb for evaluating with MAE

ann_MSE.ipynb for evaluating with MSE

RF.ipynb

Heatmap_ANN.ipynb and Heatmap_RF.ipynb are used to generate the data used for the heatmap visualization.