HRL Imperviousness — Time Series Reconstruction (Change Backcalculation)

Overview

This repository contains a Jupyter Notebook demonstrating a method for reconstructing spatially consistent historical Imperviousness Density (IMD) layers of the CLMS High-Resolution Layer Imperviousness (HRL Imperviousness), using the most recent (and most accurate) status layer as a baseline together with the Imperviousness Density Change (IMDC) layers.

The notebook shows why the intuitive approach — subtracting detected change backwards in time — produces physically impossible values (e.g. negative imperviousness), and introduces a robust alternative: binary mask substitution, which uses the change layers only as a switch ("did this pixel change?") rather than as a quantity to subtract.

Features

• Interactive site selection (Innsbruck at 10 m, North Italy at 100 m) with an overview map.
• Side-by-side visualisation of the two reconstruction methods and their pixel-value distributions.
• Interactive folium map to toggle between the original, subtraction, and binary-mask substitution layers for any reconstructed year.
• A discussion of the limitations of the method and when it should (not) be used.

Getting Started

Launch via Binder

The easiest way to explore the notebook — no installation required:

The first launch can take a few minutes while Binder builds the environment; subsequent launches are faster thanks to caching.

Launch locally

Clone the repository:

git clone https://github.com/copernicus-land/clms_nvlcc_backcalculation.git
cd clms_nvlcc_backcalculation

Create the conda environment from environment.yml (this includes GDAL, which is easiest to install via conda-forge):

conda env create -f environment.yml
conda activate nvlcc-backcalculation

Then launch Jupyter and open the notebook:

jupyter lab

Launch with Docker

If you have Docker installed, run_docker.sh builds an image from environment.yml and serves the notebook with JupyterLab — no local conda setup needed:

./run_docker.sh

Then open http://localhost:8888/lab/tree/change_backcalculation.ipynb. The first run takes a few minutes while the environment is built; later runs start in seconds. The repository is mounted into the container, so changes you save in the notebook are written back to your local copy. Stop the server with Ctrl+C.

By default the server only accepts connections from your own machine and needs no token. To reach it from other computers on your network, use --lan:

./run_docker.sh --lan         # optionally followed by a port, e.g. ./run_docker.sh --lan 9000

This listens on all network interfaces and protects the server with a random token; the script prints the links (including the token) to open from another machine. Anyone with the link can run code on the host, so only share it with people you trust.

Repository Structure

• change_backcalculation.ipynb – main notebook. Focuses on the idea: site selection, the two reconstruction methods, and their comparison; loading, plotting, and widget boilerplate live in helpers/.
• helpers/raster_utils.py – raster I/O, resampling, and image-conversion functions.
• helpers/data_loading.py – site definitions and status/change raster loading.
• helpers/analysis.py – invalid-pixel summary used to compare the two methods.
• helpers/plotting.py – all matplotlib/folium map and chart builders.
• helpers/ui.py – dropdown-selector widget factory.
• data/ – small demo dataset: cropped IMD/IMDC GeoTIFFs for two study areas (Innsbruck, North Italy) plus site_definition.yaml, which defines their extents and resolutions.
• environment.yml – conda environment definition (used locally, by Binder, and by Docker).
• run_docker.sh – builds a Docker image from environment.yml and serves the notebook.

Data

The data/ folder contains a small cropped subset of the Imperviousness Density (IMD) and Imperviousness Density Change (IMDC) layers of the CLMS HRL Imperviousness, provided only to make this notebook self-contained and runnable. It is not the full dataset.