Forest species mapping supported with machine learning.

Source code for paper Boreal forest species mapping supported with machine learning using spectral, terrain and texture features

We observed the influence of various feature types, including values of initial bands, vegetation indices, terrain features and texture features on the performance of popular machine learning algorithms, namely Random Forest, Support Vector Machine, eXtreme Gradient Boosting, Ridge classifier, k Nearest Neighbour, Decision trees, Extra trees and Logistic regression mode.

Map demonstrate classified forest map by Ridge Regression

Installation

Clone this repository

Install python packages

matplotlib
numpy
seaborn
sklearn
pandas
ipython
jupyter

How to run in Jupyter? Easy!

requirement - Docker 🐳

Clone repository to your local machine git clone https://github.com/mishagrol/ForestMapping.git
Go to folder

cd ForestMapping

Run Jupyter in Docker bash run_in_docker.sh
Open Jupyter in browser at localhost:8890, token is SecretToken

Data

Inventory plots data

Availdable by request - m.gasanov[@]skoltech.ru

Sentinel2 L2A data

2019 year data Y.Disk

2020 year data Y.Disk

Terrain data

Availdable by request due to file size (near 16 Gb) - m.gasanov[@]skoltech.ru

Texture data

Texture features generated by EO-learn package - https://github.com/sentinel-hub/eo-learn

Source Code

TrainML.ipynb - jupyter to conduct ML model training with satellite data, spectral, texture and terrain features

TrainDL.ipynb - jupyter to conduct CNN training with satellite data, spectral, texture and terrain features

Plots

BarPlots, Maps and other visualization created with python.

To reproduce plots install python package

SciencePlots

Open Plots.ipynb file with Jupyter-notebook

License

Distributed under the MIT license. See LICENSE for more information.

mishagrol/ForestMapping