Participated in a hackathon in Stockholm to detect forest storm damages
From Crisis Hackathon to Forest Storm Damage Detection
Over two intense days, I participated in a hackathon focused on the future of crisis management in Sweden. The event brought together people working with Earth observation, AI, GIS, data analysis, infrastructure, and emergency preparedness to explore how satellite data can support faster decision-making during crises.
I joined Team 7, together with David Scholz and Arnab Barua, and we worked on the forest track. Our challenge was simple to frame but difficult to solve:
If a major storm hits Sweden tonight, how quickly can we know where the forest damage has occurred?
This is not only a forestry question. Forests cover a large part of Sweden and are deeply connected to the economy, rural livelihoods, carbon storage, energy, construction, and total defence preparedness. When storms cause windthrow, rapid information becomes critical. Damaged timber can quickly lose value, increase bark beetle risk, and create logistical pressure for inspection and salvage operations. The earlier authorities and forest managers know where the damage is, the better they can prioritise field checks and response.
Our goal was to build a prototype that could detect storm-related forest damage using open satellite data.
We used Sentinel-1 radar, Sentinel-2 optical imagery, and Sweden’s Nationella Marktäckedata 2018 land-cover layer. The first step was to define where damage could reasonably occur. We masked out non-forest areas and removed known clear-cuts so that the analysis focused on stable productive forest. This helped reduce obvious false positives before applying satellite-based change detection.
For the Sentinel-2 workflow, we compared pre- and post-storm optical imagery using vegetation and moisture-related indices: NDVI, NBR, and NDMI. We used a deliberately sensitive threshold across all three indices and then filtered small patches to identify candidate damaged areas. This gave a clear candidate layer of approximately 428 hectares across 287 patches, which we then checked visually against before-and-after imagery. The visual comparison showed that many detected areas corresponded well with visible forest canopy loss.
The Sentinel-1 workflow was more challenging. Radar is attractive for crisis response because it works through clouds and darkness, which is exactly what is needed during winter storms. However, winter conditions in Sweden introduce another problem: snow and freeze effects strongly influence C-band SAR backscatter. Our first rapid before/after approach, comparing December 2018 with January–February 2019, produced sparse and unreliable anomalies. Even stable forests showed a roughly 1 dB shift, indicating that the signal was strongly affected by winter snow/freeze conditions rather than damage alone.
So we changed strategy. Instead of forcing an immediate winter comparison, we tested a broader seasonal approach: autumn 2018 versus spring 2019. This reduced the snow-related issue and produced a much cleaner Sentinel-1 result, with around 163–175 hectares of SAR disturbance candidates. It still did not match Sentinel-2 as clearly, but it showed that radar can provide complementary information when framed carefully.
The main lesson for me was that crisis mapping is not just about applying an algorithm. The hardest part is often understanding the sensing conditions, choosing the right temporal window, defining what should be excluded, and being honest about what the data can and cannot tell us. In this case, Sentinel-2 gave the clearest damage extent, while Sentinel-1 showed potential but also reminded us that radar is not automatically a perfect solution in winter forest environments.
By the end of the hackathon, our team had a working prototype demonstrating that open satellite data can support faster, landscape-wide awareness of storm damage. It is not a replacement for field inspection, but it could help agencies such as Skogsstyrelsen prioritise where to look first and respond faster than relying only on ground reports.
For me, the hackathon was a valuable opportunity to work on a practical Earth observation problem under real time pressure. It connected remote sensing, crisis management, forestry, and public-sector decision-making in a very concrete way. It also reinforced something I keep coming back to in my own research: satellite data becomes most useful when it is translated into information that can support real decisions.