Seasonal Landsat summaries improve cross-fire prediction of post-fire conifer establishment

After a wildfire sweeps through a forest, the landscape is often left scarred and silent. For forest managers and ecologists, the most pressing question is not just how much damage occurred, but whether the forest will heal itself. Will new pine and fir trees take root, or will the land remain dominated by shrubs and grasses? This decision determines whether a region needs expensive human intervention to replant trees or if nature will do the work on its own. To answer this, scientists have long turned to satellites, which can see vast areas of burned land that are too dangerous or remote for people to visit. However, a satellite image is just a picture of light bouncing off the ground; it cannot tell the difference between a new pine sapling, a sprouting shrub, or a surviving tree branch. The challenge has always been to figure out if the patterns seen from space actually predict where new conifers are growing, and if those patterns hold true when looking at a completely different fire than the one the computer was taught on.

A new study by independent researcher Joshua Puliafico tackles this problem by testing how well satellite data can predict forest recovery across different wildfires. The research focuses on using data from the Landsat satellite program, which has been photographing the Earth for decades. The team wanted to know if summarizing the satellite data into broad seasonal trends would be more useful for predicting forest recovery than trying to capture every single detail of the vegetation's growth cycle. They built computer models to learn from field data collected after twenty-five different fires in the western United States. These models were then tested on thirteen completely new fires that the computer had never seen before, simulating a real-world scenario where a manager needs to assess a fresh burn site using lessons learned from past events.

The researchers compared three different ways of feeding information into their computer models. The first was a baseline approach that used only ecological context, such as how hot the fire burned, how far the site was from existing trees that could drop seeds, and the local climate. The second approach added a summary of the satellite data, condensing months of observations into a single seasonal picture for each year after the fire. The third approach was the most detailed, adding hundreds of extra variables that tried to capture the specific timing of green-up, the peak of growth, and the decline of vegetation within each season. The goal was to see which method helped the computer best guess whether a specific plot of land would have new trees or not.

The results showed that the middle approach, using seasonal summaries, was the clear winner. When the models were tested on the thirteen new fires, adding the seasonal satellite data significantly improved their ability to distinguish between areas that would regrow and those that would not. This improvement was consistent across almost all the new fires, suggesting that the method works well even when moving from one geographic area to another. The models using seasonal summaries were able to rank the fire sites more accurately, helping managers prioritize which areas to visit first. Specifically, when the models were asked to select the top twenty percent of plots that were most likely to lack new trees, the seasonal approach helped them find a much higher proportion of those empty spots compared to using only the basic ecological data.

Interestingly, the most detailed approach did not provide any extra benefit. Adding the hundreds of extra variables that tracked the precise timing of growth within the seasons did not make the predictions more accurate. In fact, for the new fires, the detailed model performed slightly worse than the simpler seasonal summary. This suggests that for the purpose of predicting forest recovery across different fires, a broad view of the seasonal cycle is more reliable than a high-resolution view of the daily or weekly changes. The extra detail may have introduced too much noise or relied too heavily on local conditions that did not apply to the new fires.

The study also revealed a crucial nuance about what the satellite data is actually measuring. The researchers tested a version of the model that only used satellite data from before the fire occurred. Surprisingly, this pre-fire information also improved the predictions compared to the basic ecological baseline. This means that the success of the seasonal model is not solely due to seeing new trees grow after the fire. Instead, the satellite data likely captures a mix of factors: the condition of the land before the fire, the severity of the burn, and the subsequent recovery. Because the pre-fire data was already helpful, the researchers concluded that the full seasonal model cannot be credited entirely with measuring post-fire recovery in isolation. It is a powerful tool that combines history and current conditions to make a better guess.

Ultimately, this work offers a practical path forward for forest management. It demonstrates that a relatively simple summary of satellite observations can help managers identify which burned areas are struggling to recover and which are likely to bounce back on their own. By using these seasonal summaries, agencies can focus their limited field survey teams on the sites that truly need attention, rather than wasting resources on areas that are already recovering. While the models are not perfect and cannot yet replace a physical visit to the forest, they provide a reliable way to screen thousands of acres and make smarter decisions about where to invest in reforestation efforts. The study confirms that looking at the forest through the lens of seasonal change, rather than trying to capture every fleeting moment of growth, is the most effective way to predict the future of these recovering landscapes.