A semi-variogram approach for estimating stems per hectare in Eucalyptus grandis plantations using discrete-return lidar height data

Estimating stems per hectare (SPHA) for a given forest area from high spatial resolution remotely sensed data usually follows the identification of individual trees. A common method of tree identification is through local maxima filtering, which in the context of a lidar canopy height model (CHM), seeks to locate the highest value within a specified neighbourhood of pixels. Hence, specifying an appropriate window size is a critical consideration. This study investigated the potential of the semi-variogram range towards defining an average window size for a given plot within Eucalyptus species plantations. The analysis also included comparisons of CHMs with three pixel sizes (spatial resolutions) (0.2 m, 0.5 m, and 1 m) at lidar point density of 5 points/m² and three lidar point densities (1 point/m², 3 points/m², and 5 points/m²). These variations were introduced to study the effect of interpolated height surface resolution and lidar point density, respectively, on the identification of trees. Semi-variogram analysis yielded range values that varied distinctly with spatial resolution and point density. Computation of SPHA based on the semi-variogram range values resulted in overall accuracies of 73%, 56%, and 41% for 0.2 m, 0.5 m, and 1 m resolutions, respectively. A comparative approach, that defines window size based on pre-determined tree spacing, yielded corresponding accuracies of 82%, 82%, and 68% at the respective CHM resolutions. Point density comparisons based on interpolated CHM of 0.2 m resolution and the semi-variogram approach resulted in similar results between 5 points/m² (73%) and 3 points/m² (70%), whereas 1 point/m² returned the lowest accuracy (56%). Similar trends with superior accuracies were observed using the pre-determined tree spacing approach from the same resolution CHM: 82% (5 points/m²), 80% (3 points/m²), and 74% (1 point/m²). While all estimates were negatively biased, the CHM with a 0.2 m spatial resolution at a point density of 3 points/m² resulted in a reasonable level of accuracy, negating the need for high density (>3 points/m²) lidar surveys for this purpose. It was concluded that the semi-variogram approach showed promise for estimation of SPHA, particularly due to its independence from a priori knowledge regarding the tree stocking of the plantation.