Fine scale heterogeneity in surface fuel components is best characterized with point cloud data captured by terrestrial laser scanning (TLS) at plot scale, but these data are impractical to collect across an entire prescribed (Rx) burn unit. Here, we exploited multi-source hierarchical point cloud datasets collected pre- and post-fire in a Rx burn unit at The Nature Conservancy’s (TNC) Sycan Marsh preserve in southern Oregon to test approaches for 3D fuel mapping for input to advanced fire models such as QUIC-Fire. The TLS analytical tool 3DFin was applied to estimate tree heights and diameters at breast height (dbh), and then another tool (Superpoint transformer) was used to classify the ponderosa pine forest into boles, branches, shrubs, and herbaceous fuel components. This supervised classification of TLS point clouds was then used to inform a subsequent classification of UAS photogrammetric point clouds collected synoptically across the burn unit before and after the Rx Fire. The TLS-derived tree height and dbh estimates served as reference data for imputing dbh to tree objects segmented from UAS and ALS point clouds, based on independently estimated tree heights. Ponderosa pine allometrics and reasonable assumptions were used to partition tree biomass estimates into bole, branches and foliage components. Finally, loadings of surface fuel components were predicted using surface fuel accumulations predicted at tree object scales from ALS-derived crown metrics and reasonable estimates of deposition rate, decomposition rate, and time since fire. The resulting map at 1m resolution more realistically captures heterogeneity in surface fuel distributions, as is observable in the field but with insufficient precision to inform physics-based fire behavior models. To close this gap, FastFuels was used to voxelize aggregate bulk density of trees and surface fuels at 1m from the classified point cloud data across the entire burn unit (~10 ha) as inputs for QUIC-Fire; subsequent QUIC-Fire simulations compared Rx fire behavior arising with point cloud mapped heterogeneity versus more homogeneous surface fuel loadings typically used and available from LANDFIRE. This comparison tested whether heterogeneous surface fuel beds have as much influence on surface fire behavior as a more clumped (vs homogeneous) distribution of trees. Our results highlight the merit in sampling fuel conditions at fine scale using TLS to provide training data, the value of ALS and UAS data for mapping tree objects at synoptic scales, and the need to account for surface fuel heterogeneity for improved fire behavior models and Rx fire planning.
- Tags
-