EMIT-ECOSTRESS

This repository hosts the code for the EMIT-ECOSTRESS project from Caltech's CS 101 course in Fall 2023. The project goal is to use machine learning to predict urban land surface temperature as measured by the NASA thermal imager ECOSTRESS by using the surface's reflectance spectra as recorded by the NASA hyperspectral imager EMIT. The repository contains code for dataset creation and data processing as well as all modeling and analysis.

Quick Start

1. Clone this repository.

2. Download the data from the team's Google Drive here: https://drive.google.com/drive/folders/1F0khkxABuI1tzEYNzjSlvQq6dORTq9Zq?usp=drive_link. (Alternatively, you can follow the data_prep/data_download notebook 01_Finding_Concurrent_Data_UrbanHeat.ipynb, followed by the data_prep/dataset_creation notebooks Data_Matching.ipynb, Collapsing_Dataset.ipynb, and Dataset_Splitting.ipynb to build your own copy of the dataset.)

3. Place the data in a directory alongside the cloned repository

4. Start a new virtual environment and install the dependencies found in the requirements.txt file above using pip install -r requirements.txt.

5. Follow the modeling/Patch_to_Pixel.ipynb notebook (or the modeling/CNN.ipynb notebook) to train your own temperature prediction models!