lasseke / vt-land-cover-classification

Using machine learning to model the distribution of Vegetation Types in Norway

Project on using supervised classification to predict the distribution of Norwegian Vegetation Types from environmental background variables. The code heavily relies on the Google Earth Engine Python API (https://developers.google.com/earth-engine/api_docs) and numerous supporting Python packages (see environment.yml).

Installation suggestion to run the project code locally

Install Anaconda/Miniconda and Git. In a terminal where the conda and git commands are available (e.g., Anaconda Prompt), run:

cd [path/to/download/target/directory]
git clone https://github.com/lasseke/vt-land-cover-classification.git
cd ./vt-land-cover-classification
conda env create -f environment.yml
conda activate dmvtnor-env

Subsequently, navigate to the "notebooks" directory for the analysis scripts used in Keetz et al. (in prep.). The notebooks may need to be executed sequentially to reproduce results (ascending file numbering).

Project structure overview

Directory	File(s)	Summary
data/dict/	colors.json	Defines colors shared across different plots.
	predictors.json	Defines metadata (long names, etc.) for the predictor variables.
	spectral_indices.json	Defines long names and band calculation formulas for the spectral indices.
	vt_classes.json	Defines metadata (long names, ecosystem group, etc.) for the Vegetation Type classes.
data/misc/	vtdata_5f_spatial_cv_indices.pkl	Stores 10-fold leave-location-out cross-validation indices for VT feature matrix entries.
data/interim/	*	Stores interim outputs of processed datasets. Not included.
data/processed/	*	Stores final outputs of processed datasets. Not included.
data/raw/	*	Stores original datasets. Not included.
notebooks/	00-*.ipynb	Notebook for minor data-preprocessing.
	01-*.ipynb	Notebooks to retrieve and export data in required formats.
	02-*.ipynb	Notebooks to preprocess the data for model fitting (clean, generate feature matrix, create shared spatial cross validation indices, etc.).
	03-*.ipynb	Notebooks to calculate and visualize data statistics (predictor correlation, etc.).
	04-*.ipynb	Notebooks for model experiments.
	A-*.ipynb	Notebooks for experiments run on an HPC cluster for better performance.
src/	*.py	Python helper code.