FIREDpy-NRT

Project funded by NASA Applied Sciences Program and developed by ESOC and Earth Lab, CIRES - CU Boulder

Signal evaluation first stage (Summer 2023)

Done by Behzad Vahedi

Our project aims to extend EarthLab’s "Fire Event Delineation for Python (FIREDpy)" library by updating it for Near-Real-Time (NRT) fire event perimeter mapping through the fusion of optical and radar remote sensing data. Our main motivations are to achieve closer-to-real-time temporal resolution and improve the FIREDpy outputs' spatial resolution.

In the first stage of the project, we focused on answering two questions:

• Q1. Is applying BRDF correction necessary to generate NRT fire perimeters?
• Q2. How does fire affect the signal in Sentinel-2 and Landsat-8 images, and how do these satellites compare?

Below, I will explain how I approached each question. To run the scripts in this repository, you should create a conda environment using the environment.yml file. Follow the instructions provided here to do so.

Q1. BRDF Correction

The flowchart below demonstrates the steps I took to answer the first question, with the caveat that I only tested BRDF on Sentinel-2 images (using the Sen2nbar package) since I couldn't find any Python implementation of this correction for Landsat images.

Now, let's go through the different steps in the flowchart

The steps in the first row of this flowchart are done using the FiredPy package. This resulted in a shapefile of 20 fire events in the Americas, which you can find in the fire_events folder.

Once the fire events are extracted, you need to get pre- and post-fire image pairs for each event. This is done using the optical_scenes.py script. This script uses the fire's footprint and its start and end date to search for optical images. It can download the images and/or save their footprints. You can run this script using the command below:

python optical_scenes.py --event_id "202" --satellite "sentinel"

or alternatively,

python optical_scenes.py -id "202" -s "sentinel"

In these commands, the value of the event_id (or id) argument must be the ID of the fire event for which satellite images are to be searched. This comes from the id field of the fire events shapefile. The value of the satellite (or s) argument must be the name of the satellite, either "sentinel" or "landsat".

Important Note:

Before running the optical_scenes.py script, you should create separate accounts for Landsat and Sentinel API and set your username and password in the corresponidng variables in the config.py file.

You can set or change the search criteria in the config.py file. These parameters are:

• max_cloud_cover: Maximum cloud cover percentage. default: 40%.
• delta_days_landsat: The number of days to use as a buffer before and after the fire event when searching for Landsat images. default: 70 days.
• delta_days_sentinel: The number of days to use as a buffer before and after the fire event when searching for Sentinel-2 images. default: 40 days.
• download_scenes: Whether to download images or not. default: False. CAUTION: depending on the number of the images found, this could take a long time and might reqiure a huge amount of storage.
• save_footprints: Whether to save image footprints or not. default: True.
• update_json: deprecated.
• data_dir: The directory where the images and/or their footprints will be saved. See below for more information.

I suggest first running the script with the default values for download_scenes and save_footprints. This will give you the footprints and all the available metadata of all the images that satisfy the search criteria in a geojson format. You can then look at the image footprint and metadata, decide which images you'd like to download, and then rerun the script to download only those images (with download_scenes=True). However, if you decide to do so, you should remember to modify the get_landsat() and get_sentinel() functions inside optical_scenes.py following the instructions provided there

Directory Organization

data_dir is the directory where you want the images to be saved. Once you create this directory, extract (unzip) the "Fire_events.zip" there. This will (should) create a new subdirectory called Fire_events. Per each fire id you run the optical_scenes.py script for, a new subdirectory will be created within "Fire_events". Also, depending on what the -s parameter is, the corresponding subdirectory will be created within the fire_id folder where the footprints and images will be saved.

Q2. Optical Signal Evaluation

To perform the optical signal evaluation and visualize the results, you can use the Jupyter notebooks provided here: One for Sentinel-2 (Sentinel Image Analysis.ipynb) images and one for Landsat-8 images (Landsat Image Analysis.ipynb). These notebooks are designed modularly, meaning that the functions used in these notebooks can be used outside the notebooks for further investigation and analysis. They should be self-explanatory given the docstrings and comments provided. If you have any questions, please contact Behzad.

Assuming that you have already downloaded the pre- and post-fire images for that fire in the previous step, to run these notebooks, you should:

1. Change fire_id in cell #3 to the fire ID of your interest
2. Change image_dir in cell #4 to the directory where your images are located.
3. Change fire_events to reflect the directory where your fire shapefile is saved (assuming you have followed the directory structure explained above).
4. The MASK_IMAGES variable in these notebooks determines whether the calculations should be masked to the perimeter of the fire or not. Recommended values: True when calculating dNBR or applying BRDF, False when you want to create and RGB or false-color composite image.
5. These notebooks assume exactly one image before the fire and exactly one image after the fire. If you have more images, you should modify the notebooks.

You should run the notebooks once per fire event.