Enhanced forest interior estimations utilizing lidar-assisted 3D forest cover map

Abstract

Accurate estimation of forest interior is essential for large scale sustainable forest management. Conventional 2D forest cover maps only indicate the presence and absence of forests and may introduce quantity and allocation bias in the delineation of forest interior due to the inclusion of early successional/recently disturbed forests. We addressed this issue by utilizing both lidar data and National Land Cover Database (NLCD) to generate a 3D forest cover map in the forested region of western North Carolina, USA. We first classified forest cover as either early or late successional based on vertical structural information (using either a single-height threshold or a variable-height threshold by forest type). We then estimated forest interior based on these different forest cover classes and assessed disagreement in quantity and allocation of interior forests as determined using 2D forest cover maps. In addition, we determined spatial relationships between developed areas and forest interiors to evaluate the impact of landscape settings on the estimation of forest interior. Our results indicated that using a single-height threshold, the 3D forest cover map was able to distinguish early vs. late successional forests with a classification accuracy of 96.6%. A similar classification accuracy (94.6%) was achieved when applying variable-height thresholds by forest type. Based on the single-height threshold method, excluding early successional forests (6.4% of all forested area) reduced estimates of forest interior by 10.3%, 9.6% and 10.4% at spatial resolutions of 4.4 ha, 39.7 ha and 234 ha, respectively. Using variable-height threshold by forest type, the estimates of forest interior were approximately 1% less than the estimates using the single-height threshold method due to the slightly decrease of the early successional forest classification (5.7% of all forest area). Our results indicated the forest interior may be overestimated without accounting for successional stage. Moreover, geospatial distance analysis revealed the overestimation of forest interior to be most pronounced in highly-fragmented areas. Our study demonstrated the advantage of considering 3D structural information for the accurate estimation of forest interior, particularly in areas already having fragmented forests. This information can be relatively easily obtained from lidar data. This 3D method will allow for the creation of high-accuracy forest interior maps that can help to answer a variety of ecological questions and improve forest management and conservation.