Remote sensing characterization of vegetation recovery after fire

Another way to characterize vegetation dynamics is to examine how classified vegetation products change over time. The team developed a process that enables us to sample every undisturbed 30-meter pixel of the LANDFIRE Existing Vegetation Type (EVT) product within 3 km of the fire boundary. Pixel locations are separated into separate “test” and “training” data pools, and independent geospatial data (e.g., Landsat Band 2-7 30-meter Analysis Ready Data (ARD)) for the year prior to the fire (peak green period) is then extracted. An ML EVT model can then be built using the training data for each independent variable, with accuracy metrics provided using the held-out test data. This ML model can then be applied forward and backward in time to ARD data acquired at the same time as the original model was developed. The resulting EVT output maps can then be examined to see changes in vegetation classification as a result of the fire, as well as changes over time to reflect classification changes that occur during recovery.

Long-term landscape recovery is also being spectrally assessed using a time series of synthetic Landsat imagery (1984-2022). While Normalized Burn Ratio (NBR) assessments are a commonly used burn severity assessment method, such as used in the Monitoring Burn Severity Trends program, this work aims to apply a similar approach over a longer timeframe. In terms of NBR, it is known that areas with higher burn severity take longer to recover to pre-fire conditions than areas with lower burn severity. This work aims to establish a method for predicting landscape-scale recovery times using differential NBR assessments as a starting point. Considering the effects of meteorological/climatic as well as topographic influences on this recovery will allow for a range of recovery scenarios to be generated. Currently, a working recovery model is being used to generate prototype forecasts of recovery times from historical fires near the Dixie Fire to establish a long-term post-fire recovery record (Figure 2). A more robust working definition of spectral recovery is being developed to include climate variables such as seasonal precipitation and temperature anomalies as well as topographic (slope, aspect, and elevation) inputs. More robust modeling methods (e.g., ML) are being considered to determine the impacts of these additional inputs and determine whether they should be incorporated into other forecasts.

Next steps will include integrating what we have learned from using different types of sensors to characterize vegetation condition and post-fire recovery processes and developing an approach that leverages their respective strengths to develop products or tools that can be used for post-fire decision support. Further exploration of the relationships and connections between these medium-resolution products and data collected at significantly finer resolutions is also needed, for example, to understand the loss of fidelity when comparing vegetation structure assessments from ground-based lidar, airborne lidar, or GEDI.

funds: Funding for this project is provided by the Robert T. Stafford Disaster Relief and Emergency Assistance Act (42 U.S.C. 5121 et seq.) and Federal Disaster Relief Supplemental Funding ActThrough this funding, the USGS supports recovery efforts in declared natural disaster areas to aid recovery from massive wildfires, devastating hurricanes, prolonged volcanic eruptions, and destructive earthquakes. This enables the USGS to repair and replace equipment and facilities, collect high-resolution elevation data, and conduct scientific research and assessments to support recovery and reconstruction decisions.