Data sets for testing ML algorithms for deforestation mapping in Brazilian Amazonia.

This project contains data sets for testing machine learning algorithms for deforestation mapping and monitoring in Brazilian Amazonia. These data sets consists of image time series of one Sentinel-2 tile (20LMR) for year 2022, as well R scripts used to perform land use classification together with training data sets.

Access to large data sets using git lfs

The data set is a regular series of images covering MGRS tile 20LMR, with 23 time instances for the period 2022-01-05 to 2022-12-23. Each time instances contains 10 bands (B02, B03, B04, B05, B06, B07, B08, B8A, B11, B12). Image names folow the standard SENTINEL-2_MSI_20LMR_<band>_<year-month-date>.tif. The data set is 10 GB in size.

Since the data set is large, you need to use git lfs (support for large file sizes using git). Install git lfs according to the instructions at git lfs site.

How to get the data set

To get the data, you will need to clone the rondonia20LMR package from e-sensing github repository to a local directory (/data_dir in what follows). Open a local terminal and run the commands below:

% cd /data_dir
% git-lfs clone https://github.com/e-sensing/rondonia20LMR.git

Please replace data_dir with your preferred choice in all guidelines below.

Image data cube for Sentinel-2 tile 20LMR for year 2022

The directory

`data_dir/rondonia20LMR/inst/extdata/images` 

contains a regular series of images covering MGRS tile 20LMR, with 23 time instances for the period 2022-01-05 to 2022-12-23. Each time instances contains 10 bands (B02, B03, B04, B05, B06, B07, B08, B8A, B11, B12). Image names folow the standard SENTINEL-2_MSI_20LMR_<band>_<year-month-date>.tif.

Training samples for deforestation mapping

The directory data_dir/rondonia20LMR/inst/extdata/samples contains time series of SENTINEL-2 data to be used for classification with machine learning methods which are available when the package is loaded. All satellite image time series have the following columns:

• longitude (East-west coordinate of the time series sample in WGS 84).
• latitude (North-south coordinate of the time series sample in WGS 84).
• start_date (initial date of the time series).
• end_date (final date of the time series).
• label (the class label associated to the sample).
• cube (the name of the image data cube associated with the data).
• time_series (list with the values of the time series).

Using the data in R and RStudio

Before running the code in R, please install the sits package from CRAN. If there are problems with the installation, please follow the instructions in the sits book.

After installing sits and downloading the data set, please open the file

/data_dir/rondonia20LMR/sits_classification.R

This file contains a script that shows the use of sits package for mapping land use and land cover in the image data cube associated with the rondonia20LMR tile for year 2022.

For description of how these scripts work, please see chapter "Introduction" in the sits reference book.

Using the data in Python

Python users will find the images in directory

data_dir/rondonia20LMR/inst/extdata/images

A version of the training data for Python users is available as a CSV file in

data_dir/rondonia20LMR/inst/extdata/samples/samples_rondonia.csv. 

Reproducible code for paper on Bayesian smoothing

This repository contains data and code to reproduce the paper "Bayesian Inference for Post-Processing of Remote-Sensing Image Classification". The code to reproduce the results presented in the paper is available at the following address

data_dir/rondonia20LMR/R/bayes_paper_code.R

Viewing the data in QGIS

To visualise the images in QGIS, we provide a project file in

data_dir/rondonia20LMR/inst/extdata/qgis/bayes_paper.qgz

License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).