Inundation Habitats Pilot

Heremans, Stien¹²³ Research Institute for Nature and Forest (INBO)⁴⁵

keywords: remote sensing; inundation; satellites; wetlands; grasslands

This R project provides a complete workflow for acquiring Sentinel-2 satellite imagery, processing it against ground-truth data, and validating an inundation detection model. The analysis places a special emphasis on understanding how model performance is affected by the purity of land cover within each pixel.

---

Key Features 📜

• Automated Data Pipeline: Scripts are designed to run sequentially, creating a reproducible analysis pipeline from raw data to final validation.
• Cloud-Based Image Acquisition: Leverages openEO to process Sentinel-2 data on the Copernicus Data Space Ecosystem, minimizing local processing requirements.
• Pixel Purity Analysis: A core feature is the analysis of how mixed pixels affect model accuracy, providing deeper insights than a simple accuracy score.
• Centralized Configuration: Easily run the entire workflow for different study sites or years by modifying a single config.R file.
• Integrated Data Management: Seamlessly downloads source data from and uploads results to Google Drive, ensuring data is organized and accessible.
• Rich Visualization: Generates a suite of outputs, including boxplots of spectral signatures, performance metric charts, and detailed three-panel comparison maps for spatial validation.

---

Workflow Overview

The project follows a sequential workflow, where the output of one script often serves as the input for the next.

1. Data Acquisition (OpenEO_get_Sen2_data_inundation.R): Downloads Sentinel-2 L2A imagery for a specific site and date using the openEO platform.
2. Ground-Truth Preparation (Create_fractional_coverage_raster.R): Converts labeled vector polygons (ground-truth) into a fractional cover raster that is perfectly aligned with the Sentinel-2 grid.
3. Data Extraction (Extract_labeled_pixel_values.R): Extracts the spectral values from the Sentinel-2 image for pixels identified in the ground-truth raster, creating a tidy dataset for analysis.
4. Tabular Analysis & Validation (Analyze_and_Model_Pixel_Data.R): Uses the extracted data to perform a detailed numerical evaluation of a pre-trained decision tree model, calculating performance metrics (Recall, Precision, F1-Score) stratified by pixel purity.
5. Spatial Analysis & Mapping (Apply_jussila_model_map.R): Applies the decision tree model to the entire study area to produce a classified inundation map. It generates comparison plots to visually assess the spatial accuracy of the model against the ground truth.

---

Getting Started

Prerequisites

• R (version 4.2 or higher)
• RStudio IDE (recommended)
• Access to Google Drive and the necessary permissions for the project folders.
• An openEO platform account for satellite data acquisition.

Installation

1. Clone the repository:

   git clone [https://github.com/inbo/HabPil_inundation_R.git](https://github.com/inbo/HabPil_inundation_R.git)
   cd HabPil_inundation_R

2. Install R Packages: Open the R project and run the following command in the console to install all required packages.

   install.packages(c("terra", "dplyr", "ggplot2", "rpart", "googledrive", "openeo", "tidyr", "scales", "ggnewscale"))

3. Configure Environment:
   • Authenticate Google Drive by running googledrive::drive_auth() in the console and following the prompts.
   • Modify the config.R file to set your desired study_site_name, target_year, and relevant Google Drive folder IDs.

---

Scripts Overview 🔎

This project is organized into a series of primary analysis scripts and helper scripts that contain reusable functions.

Main Scripts

OpenEO_get_Sen2_data_inundation.R

• Purpose: Automates the acquisition of Sentinel-2 L2A satellite imagery for a specific date, study site, and year using the openEO platform.

Create_fractional_coverage_raster.R

• Purpose: Generates foundational fractional cover rasters from labeled polygons. These rasters are essential for understanding mixed pixel compositions.

Extract_labeled_pixel_values.R

• Purpose: Creates a structured dataset for machine learning by extracting spectral values from Sentinel-2 imagery corresponding to the ground-truth labels.

Analyze_and_Model_Pixel_Data.R

• Purpose: Performs the final numerical analysis and validation of a pre-trained inundation model using the extracted pixel data.

Apply_jussila_model_map.R

• Purpose: Transitions from tabular analysis to spatial validation by applying the model to the entire study area raster.

Crosstab_Cal_Area_BWK.R

• Purpose: An accessory script to analyze the Biological Valuation Map (BWK) vector data, focusing on the area and distribution of habitats within grassland and wetland units.

Helper Scripts

• config.R: Contains the input parameters (e.g., study_site_name, target_year) that the user needs to change before running the workflow.
• gdrive-utils.R: Contains functions for downloading data from and uploading results to Google Drive.
• spatial_processing_utils.R: Contains functions for core spatial processing tasks used by the main scripts.

Footnotes

1. author ↩

2. contact person ↩

3. Instituut voor Natuur- en Bosonderzoek (INBO) ↩

4. copyright holder ↩

5. funder ↩