Augmenting forest inventories: Using drones to derive individual tree metrics
Uncrewed Aerial System (UAS) drone pilots demonstrate the use of UAS for forest monitoring.
Over the last two decades, the restoration of dry conifer forests has increasingly prioritized the reintroduction of horizontal and vertical tree structural complexity. This emphasis has come from research showing that increased spatial complexity in forest structures promotes resilience to disturbance and is necessary to restore ecological functions and wildlife habitat.
However, most forest inventory and monitoring approaches lack the resolution, extent, or spatial explicitness required to describe within stand heterogeneity at a level adequate to inform forest management. Recently, unmanned aerial system (UAS) remote sensing has emerged with potential methods for bridging this gap.
Specifically, photogrammetric Structure from Motion (SfM) algorithms have been shown as a cost-efficient way to characterize forest structure in 3-dimensions. This set of studies tested how decisions about data collection, processing, and modeling impact the quality of forest inventory metrics derived from unmanned aerial system data. Working in ponderosa pine forests of Colorado, Arizona, and South Dakota, we compared unmanned aerial system estimates against sites with 100 percent mapped trees. We found that under the right processing settings we could identify the location and height of more than 90 percent of trees, with most of the missed trees being shorter and often beneath another tree. Tree heights were precise in all tests. The quality of extracted diameter at breast height (DBH) values was improved by filtering them against regional data from the Forest Service Forest Inventory and Analysis program.
Diameter at breast height values missed by the unmanned aerial system could be reliably modeled from the subset that was extracted. Stand level estimates of trees per hectare, basal area, and quadratic mean diameter was achieved within 5-15 percent error of field values. These results demonstrate the potential of a unmanned aerial system-based inventory strategy for estimating individual tree structural attributes (such as location, height, and diameter at breast height) in ponderosa pine forests, without the need for in situ field observations. Such data has potential to unlock the next generation of precision forest management tools and to effectively inform spatially explicit management objectives.
Greater than 90 percent of trees could be extracted from unmanned aerial system data. Tree height and diameter at breast height had mean errors less than 0.5 m and 3 cm, respectively. Stand-level basal area estimates were within 4–15 percent of field observed values. Future work will track management project from pre- to post-treatment to evaluate the ability of unmanned aerial system to inform treatment effectiveness.