Wildfire Event Tracking and Evolution for Operational Satellite Pipelines

Adapt NASA FEDS-style VIIRS fire-object tracking into the current pipeline and stack, producing time-resolved perimeters and spread metrics beyond proximity clustering.

Requirements

  • B.Sc. in Machine Learning, Data Science, Computer Science, Mathematics
  • Knowledge of Python and software development (Git)
  • Basic geospatial stack (GeoPandas, PostGIS concepts)
  • Familiarity with clustering (e.g. DBSCAN, spatial proximity grouping)
  • Basic concepts of data analysis and time series processing
  • Nice-to-have: Airflow familiarity, reading scientific papers

Description

Wildfires are among the most destructive natural hazards in Europe and worldwide. For civil protection and emergency mapping, it is not enough to know that active fire pixels were detected: operators need to understand where the fire was, how its footprint evolved between satellite overpasses, and where the active front is moving. Satellite-based active fire products (VIIRS, MTG, Sentinel-3 SLSTR) provide frequent point observations, but turning them into coherent fire objects with perimeters and spread metrics can be challenging.

At LINKS Foundation we have developed an operational service for near-real-time wildfire monitoring: ingesting active-fire detections from geostationary and polar-orbiting satellites (including MTG, VIIRS, and Sentinel-3), grouping them into fire events for situational awareness, and enriching those events with burned-area delineation from Sentinel-2 and, where needed, physics-based spread modelling. This internship extends that monitoring line of work rather than starting from scratch.

NASA’s Fire Event Explorer on the VEDA Dashboard visualizes how individual wildfires progress over time—perimeters, active fire lines, and context such as wind, using vector data derived from VIIRS. The underlying approach is described in Chen et al. (2022), California wildfire spread derived using VIIRS satellite observations and an object-based tracking system: at each time step, active fire pixels are clustered in space; each cluster is either linked to an existing fire object or starts a new one; perimeters and fire fronts are updated shortly after data acquisition.

Gap and objective. Our current monitoring groups detections into events but does not yet produce NASA-like time-resolved perimeters and active fire lines based on VIIRS. Modelled spread and optical delineation products exist, but they are not tied into a single, detection-driven evolution of the fire footprint over time. The internship will design and implement a prototype object-based tracker on VIIRS detections and integrate the results into our operational data services.

Main Activities

  • Literature and data survey: FEDS (Chen et al.), Fire Event Explorer, FIRMS/VIIRS product documentation, other papers and works
  • Baseline analysis: Characterize how the current event-grouping approach behaves on historical fire cases
  • Tracker implementation: Prototype event evolution using VIIRS: cluster → associate → update perimeter and active fire line
  • Integration: Design how to store outputs and serve them in the existing product
  • Validation: Quantitative comparison against NASA FEDS on selected fires
  • BONUS: Minimal timeline visualization (notebook or lightweight map view) inspired by Fire Event Explorer starting from currently stored detections

Expected outcome: a working prototype/evaluation notebook, and a short technical note comparing the new tracker with the current event grouping and the NASA one.

Contacts