FIRECYCLE Vision

Turning before-and-after satellite imagery into traceable post-fire assessments, helping land managers identify changed areas and prioritize human review.

Comment

Inspiration

After a wildfire, land managers need to understand where the landscape has changed and which areas require closer inspection. Satellite imagery provides a valuable starting point, but a useful assessment must also explain its sources, processing steps, and limitations.

FIRECYCLE Vision grows from our work on territorial monitoring and post-fire recovery in Extremadura, Spain. Its guiding principle is simple: every map should be accompanied by evidence that people can inspect.

What it does

FIRECYCLE Vision focuses on comparing before-and-after satellite observations to support post-fire assessment.

The existing processing code retrieves Sentinel-2 L2A spectral bands, aligns them to a common geographic grid, calculates the Normalized Burn Ratio (NBR) and its temporal difference (dNBR), and exports georeferenced rasters and summary statistics.

It also records source identifiers, processing parameters, and SHA-256 hashes. Outputs are marked as derived results pending human review.

The intended users are forestry technicians, local authorities, and restoration teams. The purpose is to help them identify areas for further investigation.

How we built it

The current foundation uses Python, Requests, NumPy, and Rasterio. A defined area of interest and target grid keep the before-and-after calculations spatially consistent.

The processing code includes reflectance conversion, nodata handling, spectral-index calculations, and exports of results and provenance records.

For the competition version, the next implementation step is substantive OpenCV 5 processing to extract and characterize regions of change, together with execution of a meaningful processing component on AWS. These integrations remain to be completed and demonstrated.

Challenges we ran into

Reliable comparison requires more than subtracting two images. Different acquisition conditions, cloud contamination, invalid pixels, radiometric conventions, and spatial alignment can affect the result.

Another challenge is interpretation: spectral change alone does not establish ecological damage, restoration priority, or field conditions. Those conclusions require additional evidence and expert review.

The submission must therefore distinguish what the software calculates from what a qualified person can conclude.

Accomplishments that we're proud of

The repository contains a concrete processing foundation for before-and-after satellite analysis, with georeferenced outputs and explicit provenance.

An important design choice is keeping derived measurements separate from verified conclusions. Human review remains visible in the output status.

These are implementation foundations; accuracy, runtime, and operational benefits still require reproducible evaluation.

What we learned

Traceability is part of the product. A useful result should explain which observations produced it, how it was calculated, and where uncertainty remains.

A focused demonstration on one documented area is also easier to evaluate rigorously than a broad platform with many unvalidated capabilities.

What's next for FIRECYCLE Vision

Complete and verify the OpenCV 5 integration and AWS execution.

Evaluate change masks against a documented reference, report detection errors and processing time, and demonstrate difficult cases such as cloud contamination or incomplete coverage.

Finally, provide reproducible instructions and a working demonstration showing the source imagery, processing workflow, results, and human-review boundary.

Built With

Share this project:

Updates

Submission history