National Forest Attribute Maps Using ALS andSample Plot Data for the Netherlands

Abstract

Forest inventory is essential for sustainable forest and nature management, as well as for policymaking at various levels. Traditional methods rely on plot sampling and provide reliable estimates of variables such as stem volume, basal area, and dominant height. These methods, however, lack spatial specificity. Countries such as Sweden, Finland, and Norway have successfully integrated airborne laser scanning (ALS) data with linking models to generate spatially explicit forest variable estimates. In the Netherlands, however, these approaches remained unexplored.

This study evaluates the suitability of four modelling methods for constructing detailed forest attribute maps using sample plot data from NBI7 (7th Dutch national forest inventory) and 25 ALS metrics from AHN4 (4th Dutch national ALS dataset). An area-based approach (ABA) was used, with exploratory data analysis (EDA) and four modelling techniques: least squares regression, generalized additive models (GAMs), k-nearest neighbours (k-NN), and random forest. The bestperforming models, based on RMSE (Root Mean Squared Error) from leave-one-out crossvalidation (LOOCV), achieved RMSE values of 1.34 m (~ 6%) for dominant height, 57.64 m³/ha (~ 26%) for stem volume, and 6.85 m²/ha (~ 27%) for basal area. For the number of trees, no satisfactory model was found with a best RMSE of 418 n/ha (~ 57%). All selected models were linear models derived from least squares regression. Stand-level validation showed RMSE values of 5.95 m (~30%) for dominant height, 32.84 m³/ha (~20%) for stem volume, and 5.5 m²/ha (~37%) for basal area. However, discrepancies between the ALS-based predictions and field measurements resulted from overhanging vegetation.