Assessing stand and data variability using airborne laser scanner

Abstract

An efficient forest management requires accurate and cost-effective measurements of forest inventory parameters. The cost of LiDAR surveys are directly dependent on the number and size of validation plots as well as the sampling density of points needed to adequately estimate forest inventory parameters. In this study we investigate (i) the spatial variability of the forest stand, i.e., the effect of the area chosen on the prediction of forest parameters and (ii) the relation between prediction accuracy and sampling point density for the estimation of top height, basal area and volume at plot level. Assessment of the stand’s spatial variability was carried out by comparing the accuracy of the top height estimations, using the 99th percentile of a normalised distribution of points, over areas of different size. Original sampling density was synthetically reduced to 10, 5, 4, 3, 2, 1, 0.50, 0.33, 0.25 and 0.20 returns per m2. Forest parameters were subsequently estimated for each point density by means of 99th percentile (top height) and linear regression models (basal area and volume). Predictions were validated using 11 stands, each containing one 50 50 m plot. Results show that the optimum area for forest parameters prediction is 1600 m2 with an average top height accuracy of 95.05% and a standard deviation of 3.41%. Larger sizes will merely increase the cost of field data collection without improving the accuracy. Interestingly, top height predictions were slightly more accurate for lower point densities. Linear equations yielded RMSEs of 3.28-5.28 m2/ha and 29.41-36.04 m3/ha for basal area and volume respectively. There were therefore small differences in terms of accuracy of predicted parameters for different point densities, which indicates that once a good DTM is created, future LiDAR surveys can be accomplished over the same area at lower sampling densities, and thus reducing the costs but without disregarding estimation accuracy.