Improving predictions of post-fire forest regeneration by mapping individual surviving trees

Background/Question/Methods

Wildfires in semiarid conifer forests are increasingly burning at high severity across unusually large areas. The resulting large patches of near-complete tree mortality challenge natural regeneration processes, which for most non-serotinous conifer species rely on dispersal of seeds from nearby surviving trees. While existing work shows a clear positive relationship between seed source proximity and seedling density, substantial unexplained variation remains. We built on existing efforts by mapping individual surviving trees using low-cost drone surveys combined with structure-from-motion processing. We performed these surveys across 150- to 300-ha areas that burned at moderate to high severity in five different wildfires in California mixed-conifer forests, and we paired the aerial surveys with ground-based surveys of natural seedling establishment in 851 plots in dense grids (30- to 50-m spacing) across seed source gradients. We predicted seedling density at a given point (plot) as the sum of the dispersal kernel-derived seed density estimates of all mapped trees. We optimized dispersal kernel parameters along with offsets for tree height and relative topographic position to minimize seedling density prediction error.

Results/Conclusions

Across our five focal areas, preliminary drone-based individual-tree mapping detected over 100,000 surviving trees in spatial arrangements ranging from individual trees to clusters of thousands. In comparison against ground-based tree maps, drone-based mapping detected overstory trees > 10 m tall with sensitivity = 0.69 and precision = 0.90. Preliminary dispersal models explained plot-measured regeneration density from seed source data alone with R2 = 0.31, improving on previous models and with opportunity for ongoing refinement. Patterns were much stronger for the larger-seeded, shorter-dispersing pines (Pinus spp.) than for smaller-seeded, longer-dispersing species. In 11% of plots > 200 m from surviving trees (confirmed by manual imagery inspection and plot-based observation) and where the dispersal model predicted near-zero regeneration, pine seedling densities exceeded 100 ha-1 and were positively correlated with downed cone density, suggesting the presence of an ephemeral post-mortality canopy seed bank in these non-serotinuous species given conducive fire seasonality. Our work demonstrates that by mapping individual trees and summing their dispersal kernels, the accuracy of post-fire forest regeneration predictions can be improved. As availability of individual-tree mapping data including lidar expands, individual tree-based dispersal predictions will become practical across large wildfires and may help to guide post-fire reforestation efforts.