Forest disturbances caused by snow result in considerable economic losses associated with timber production in Finland. Snow disturbances reduce the future growth of a forest and potentially increase the risk of insect attacks and fungal colonizations. Information on snow disturbances and their severities is important for decision making so that the negative impacts on timber production and carbon sequestration can be minimized.

We used a random forest (RF) classifier to detect tree-level snow disturbances using high-density airborne lidar data and aerial images in Eastern Finland. A set of 35 field plots with a radius of 10 m was collected from forests where on average 8.6 trees per plot were broken by snow. In each field plot, a subset of broken trees (on average 5 trees) was accurately positioned. A set of healthy trees were taken from the field plots of operational forest inventories organized by Finnish Forest Centre. The principles of a canopy height model-based individual tree detection (ITD) were used to prepare tree-level training data for the RF classifier. The training data of 161 broken and 848 healthy trees were balanced using a Synthetic Minority Oversampling Technique (SMOTE) in every iteration of the leave-plot-out cross validation. The classification performance of the RF model was evaluated using an F1 score at the tree-level, and the mean difference associated with the predicted number of broken trees at the plot-level.

While the RF classifier produced an F1 score of 0.67 in the classification of healthy and broken trees, the number of broken trees was underestimated (mean difference 5.6 trees). The findings suggest that the detection of snow disturbances is challenging, yet feasible using high-density airborne lidar data and aerial images.

This project was funded by the Government of Yukon, Canada

Communities in the wildland urban interface in western Canada are exposed annually to a risk of wildfires. Assessing wildfire risks in advance allows communities to undertake mitigation actions such as prescribing fuel treatments.

In this study, LiDAR data in combination with satellite is used to perform advanced fire risk assessment by generating fuel types and characterizing fire hazard at different spatial scales. Fuel attributes key to managing fire behaviour, including Canopy Bulk Density, Canopy Fuel Load, Stem Density, Canopy Base Height, and Canopy Height, were estimated using area based and individual tree based techniques on LiDAR data at 5, 10 and 20m² resolutions. Next, attributes such as Stem Density, Canopy Height, Canopy Base Height, ladder fuel density, and vegetation types were computed based on 2014 and 2020 data. Fuel type data based on the ITI and Sentinel-2 Satellite were assessed using existing field plots. Using these outputs, several different fire models based on the generated fuel type data and fuel attributes were produced using Canadian Forest Fire Danger Rating System models. Finally, risk estimates derived from these fire models were generated for all buildings in the community based on radiation and ember density exposure, as well as proximity to other structures.

The results show that a much more detailed map of fuel attributes and fuel types can be generated using LiDAR and satellite with R² of 0.74 for lidar fuel types vs. an R² of 0.56 for traditional inventory polygon approach. These maps can be used to identify structures with the highest wildfire hazard exposure as well forest areas containing the most hazardous fuel structures. Fuel management prescriptions can then be generated to target specific areas identified, enabling better use of funds to enhance the safety of the community.

Fuel consumption relates to fuel loads, fuel composition and quality, fuel moisture, active fire behavior especially energy flux, plume development, the quantity and quality of smoke emissions, and post-fire effects. Because airborne laser scanning (ALS) is sensitive to forest biomass density and fuel loads, our objective was to estimate consumption by differencing maps of pre- and post-fire fuel loads predicted from pre- and post-fire field plot and ALS collections on the Monroe Mountain unit of the Fishlake National Forest (NF) in central Utah, USA. We focused on 3 stand-replacing, prescribed crown fires applied by Fishlake NF managers to restore subalpine forests having very high accumulated fuel loads after decades of fire exclusion, resulting in bark beetle kill of encroached subalpine fir and Engelmann spruce and overmature, dying or dead aspen; the management goal was to regenerate aspen stands for improved habitat for elk and other wildlife. Trees were tallied and fuel loads measured at 61 field plots across 3 burns, all of which were characterized both before and after burning. We used a Random Forests model to predict total fuel loads fitted to the full distribution of pre- and post-fire field plot measures of fuel loads, using ALS height and density metrics as predictor variables. We found biomass fuel loads up to and exceeding 400 Mg/ha, and local maxima in the consumption maps sometimes exceeded 200 Mg/ha. At a fall burn that produced the highest smoke plume, our mapped consumption patterns were found to relate well to Ka-band Doppler radar and two Doppler lidars observations of plume development. At a spring burn in a driving wind, our consumption estimates were used to upscale microbial biomass sampled from the smoke with Unmanned Autonomous Systems to the entire burn. Across all 3 burns, aspen regeneration has been robust across all burn severities.

Litter is the major contributor to fine surface fuel loads that drive fire behavior, fuel consumption, emissions, and fire effects. However, high spatial resolution maps of litter are missing for most forested sites due to inherent complexity. Proposed approaches, including data-driven techniques and ecological models, don’t often account for its spatial and temporal variability that is largely constrained by overstory canopy cover, which provides opportunities and challenges for the direct quantification of canopy fine fuels that contribute to litter inputs to the surface fuel bed.

Overstory foliar biomass—which can be characterized by airborne lidar scanning (ALS) data—drives annual litterfall and accumulation as larger loads are expected under and near trees than in canopy gaps and on forest edges. We mapped litter loads at high spatial resolution (5 m) by characterizing spatial patterns of annual litterfall using ALS data to segment tree crowns and estimate foliage biomass. Annual litterfall was then estimated from foliage biomass based on the expected leaf longevity of the dominant species. Outputs were rasterized at 5 m, and a convolution filter was applied to simulate litter dispersion on neighboring areas. Finally, we quantified litter accumulation using fire history records through a spatially explicit implementation of the traditional Olson accumulation model. The methodology has been tested in several longleaf pine dominated forests of the southeastern US and is being expanded to western coniferous forest sites. RMSD and BIAS between estimates and reference field measurements show good agreement (0.21 kg m^-2 and 0.01 kg m^-2 respectively). This methodology to map patterns of litter production and litter accumulation provides a means to characterize the discontinuity of the litter layer. This is a novel yet simple approach that holds great potential to map fine surface fuel loads at high spatial resolution across relatively large areas where ALS is available.

The mountain pine beetle (MPB) (Dendroctonus ponderosae Hopkins) is a forest insect beetle native to the pine forests of western North America that spreads from host tree to host tree. With changing climates and warming temperatures, the population and geography of the MPB are expected to expand north and east and up to higher elevations as the warmer climates that MPB thrives in shifts. In Canada, the most commonly used fire behaviour modelling system is the Canadian Forest Fire Behavior Prediction (FBP) system, which utilizes weather, topography, and fuel types to predict fire behaviour. A significant drawback of the FBP system, however, is the limited range of fuel types, with only 16 types being defined for all of Canada and none considering MPB infestation status, which can significantly alter the forest structure of attacked trees and plots. To characterize changes in forest structure and corresponding fuels, light detection and ranging (LiDAR) is both an effective and accurate tool for capturing these structural changes. We acquired LiDAR data from remotely piloted aircraft systems (RPAS), as well as a handheld mobile laser scanner (MLS) across a range of MPB-impacted stands in Alberta, Canada. MLS and RPAS point clouds were then combined to create a high point density fused dataset. MPB-attacked plots were classified into 3 levels based on the number of trees attacked. Analysis of variance (ANOVA) tests were conducted and found trends of forest fuel as well as LiDAR structure across MPB attack levels. Given the ability to separate between MPB attack classes, a linear regression model was developed to predict fuel loads across the study areas. This model, and more advanced ones that will be developed, will help the study and understanding of MPB attack effect on fine-scale fuel advance for better management.