2. Change detection using optical images¶

Floriane PROVOST¹, Pascal LACROIX² & Diego CUSICANQUI²

¹ Institut de Physique du Globe de Strasbourg, EOST/Université de Strasbourg, Strasbourg, France

² Univ. Grenoble Alpes, CNES, CNRS, IRD, Institut des Sciences de la Terre (ISTerre), Grenoble, France.

Copyright

Document version 0.2, Last update: 2025-09-18

© PSF TelRIskNat 2025 Optical team (D. Cusicanqui, R. Basantes & P. Lacroix). This document and its contents are licensed under the Creative Commons Attribution 4.0 International License (CC BY-NC-SA 4.0).

2.1. Learning objectives:¶

• Design a workflow to answer a question

• Learn how to create RGB composite and to compute indexes

• Learn how to create manually a vector file

• Execute spatial/attribute operations to create a vector file

• Create a map that summarise your results

2.2. Introduction¶

We are going to study the wild fires that impacted California in January 2025. The main goal of our study is to estimate how many houses/building were destroyed by the 2025 wild fire?

Regarding the California wild fire event, In 2024, the San Francisco Bay Area experienced several notable wildfires around the region rather than inside the city itself. The Corral Fire near Tracy in the Diablo Range burned about 14,168 acres (≈ 57.34 km²) from June 1–6, while Sonoma County’s Point Fire scorched ~1,207 acres (≈ 4.88 km²) from June 16–24. More infor in CallFire.

Smoke from these and other regional fires periodically drifted into San Francisco, prompting Bay Area Air Quality Management District advisories—especially in mid-June and again in August—due to hazy skies and particle pollution concerns [1]. Overall, impacts in San Francisco were felt mainly through smoke and reduced air quality, while the largest burn areas were in surrounding counties SFGATE. More information about large California wildfires in [2]

First of all, try to draft a workflow/diagram describing the logical steps and operations needed to answer the question. For that, answer to these questions:

• What data do you need?

• What information do you need?

• What operation do you need to execute to obtain the results?

Once answered construct the diagram.

2.3. Raster data¶

We need to map the wild fire extent. To do that we are going to use Sentinel-2 images and manipulate the bands in order to easily detect and map the fires.

For this practice, we are going to use QGIS independent of the operating system. If you have not installed QGIS yet, please refer to the practical Installation.

2.3.1. Sentinel-2 images¶

The Sentinel-2 images are available on the Copernicus Open Access Hub or AWS. For this practical, we provide you two Sentinel-2 images acquired before and after the fire event. The data is located in the two separate folders in data/exercise_1_change_detection named by their acquisition date.

• S2_L1C_T11SLT_20250102T183751 (before the fire event)

• S2_L1C_T11SLT_20250112T183731 (after the fire event)

2.3.2. Sentinel-2 RGB composite¶

Sentinel-2 images are multispectral images, meaning that they contain several bands acquired at different wavelengths. Visit: Sentinel Hub to determine what RGB is more suitable to map wildfires.

Now, create a RGB composite for both dates (before and after the fire event). - Open the menu Raster > Miscellaneous > Create a Virtual Raster… - Define the bands composing the RGB composite. - Click on Place each input file into a separate band. - Create the Virtual raster.

2.3.3. Sentinel-2 band math¶

As fires are burning the vegetation, another option is to compute the Normalised Different Vegetation index (NDVI). This index is a widely used metric for quantifying the health and density of vegetation. Compute the NDVI by opening the function Raster > Raster Calculator. This window should appear as Fig. 2.3.

• Define the mathematical operation you want to realize (here NDVI). Be careful, select bands of the same date.

• Choose to create the raster "on-the-fly" or to save it in your project (preferred). If you choose to create the raster on-the-fly, the raster will appear in the project but will not be saved. If you save the project, close it and open it later, the raster will be empty (because it was not saved).

• Repeat the operation for the two dates of acquisition.

2.4. Vector data¶

2.4.1. Manually create a vector file¶

Vector files can be created manually with the function XXX The following window should appear like Fig. 2.5.

1. Choose the name of the file, the extension (shapefile), and set up carefully the geometry (point, line or polygon) and the CRS. Click OK once all the options have been chosen. The layer should appear in the section Layers of the GIS environment.

2. To edit the layer, click on the created vector file and on the function Toggle Editing (pencil icon).

3. Click then on the function Add Polygon Feature (yellow star icon) to create a new entity.

4. Click on the map to draw the limits of the polygon you want to map.

5. Save the entity by clicking on the function Save Layer Edits (floppy disk icon) to render the layer non editable.

6. If you want to modify the points/shape of your vector, use the function Vertex Tool (icon with a point and a line) and move/modify the points.

2.4.2. Create a vector file from raster data¶

Now, we are going to create a vector file from the NDVI change raster. The idea is to extract the area where the NDVI has decreased significantly (i.e. vegetation has been destroyed by the fire).

a. Determine what operation should be realized to extract the fire extent from the NDVI raster files. Use the Raster Calculator to derive a raster with the fire extent.

2. Vectorize the resulting raster using the function Raster > Conversion > Raster to Polygon. You should obtain a new vector layer that looks like Fig. 2.6. The spatial domain is correctly sampled into geographical entities but many geographical entities are not of interest for this study.

3. Use the function Vector > Geometry > add geometrical attributes to compute the area and perimeter of each entity of the vector layer Vectorized.

4. Use the attribute table to filter entities with areas smaller than 1,000 m², remove them by editing the table Toggle Editing (pencil icon) and deleting the selected entities (red garbage).

5. Repeat the operation to remove entities with areas larger than 68,000,000 m².

6. Regroup the entities in one single entity fire extent by using the function Vector > Geoprocessing Tools > Dissolve. Your final result should look like Fig. 2.7.

7. Create a map showing the extent of the fires in California in January 2025.

2.5. Quantifying impacts¶

2.5.1. Count the number of buildings destroyed¶

To estimate the number of buildings destroyed by the fire, we are going to use a vector file containing the building footprints in California. The file is located in data/exercise_1_change_detection/buildings_CA.shp.

1. Load the vector file buildings_CA.shp in QGIS.

2. Use the function Vector > Geoprocessing Tools > Intersection to extract the buildings that intersect the fire extent.

3. Open the attribute table of the resulting vector layer and count.

2.5.2. Check our mapping¶

We are now going to use Open Street Map (OSM) to check our mapping. OSM data is a geographic database feed by a community of mappers that contribute and maintain data about roads, trails. It is therefore possible to download the raw data directly and reuse it as you wish, for cartographic or other purposes. Every week, a complete file called planet is generated and made available by the OpenStreetMap Foundation. This file is very large and growing fast (20 GB at the end of 2011, 63 GB at the beginning of 2018) because it contains all the data for the entire planet. Extracts by continent, country or region are also available and updated daily by Geofabrik. There are also tools for breaking down a file into smaller geographical areas.

There are several ways of viewing and downloading OSM data:

2.5.2.1. Viewing the ‘static’ OSM map¶

The ‘static’ OSM map represents OSM data in raster form. It is not possible to manipulate the information contained in this layer. However, it allows OSM information to be viewed and can be used as a background map. To download the OSM map into QGIS you need to use the QuickMapServices extension. Display OSM from WMS layers already installed on your computer. Use the function Web > QuickMapServices > OSM > OSM Standard. Look at the areas affected by the fires. What do you notice?

2.5.2.2. Download the OSM database for an area of interest¶

You can download OSM data using QuickOSM plugin to download the vector data from the OSM base maps.

• Click on the function Vector > QuickOSM > QuickOSM (green lupe icon).

• Click on Theme and select the default Urban theme. Define the geographical extent for which you want to download the data. You can use the Canvas extent option to download the data for the area currently displayed on your QGIS canvas.

• Click on Execute query to download the data. Several layers should appear in your Layers section.

You can also use the OSMDownloader plugin to download OSM data. This plugin allows you to download OSM data in a specific format (shapefile, geopackage, etc.) and for a specific area of interest.

• Install the plugin OSMDownloader from the QGIS plugin manager.

• Open the plugin from the menu OSMDownloader.

• Define the area of interest by drawing a rectangle.

• Save the data in a specific format (shapefile, geopackage, etc.) and download.

A number of vector layers should appear in your Layers tab. Manipulate these layers to display the buildings mapped in OSM and compare with your result.

2.6. References¶