Airborne laser scanning (ALS) acquisitions provide piecemeal coverage across the western US, as collections are organized by local managers of individual project areas. In this study, we analyze different factors that can contribute to developing a regional strategy to use information from completed ALS data acquisitions and develop maps of multiple forest attributes in new ALS project areas in a rapid manner. This study is located in Oregon, USA, and analyzes six forest structural attributes for differences between: (1) synthetic (i.e., not-calibrated), and calibrated predictions, (2) parametric linear and semiparametric models, and (3) models developed with predictors computed for point clouds enclosed in the areas where field measurements were taken, i.e., “point-cloud predictors”, and models developed using predictors extracted from pre-rasterized layers, i.e., “rasterized predictors”. Forest structural attributes under consideration are aboveground biomass, downed woody biomass, canopy bulk density, canopy height, canopy base height, and canopy fuel load. Results from our study indicate that semiparametric models perform better than parametric models if no calibration is performed. However, the effect of the calibration is substantial in reducing the bias of parametric models but minimal for the semiparametric models and, once calibrations are performed, differences between parametric and semiparametric models become negligible for all responses. In addition, minimal differences between models using point-cloud predictors and models using rasterized predictors were found. We conclude that the approach that applies semiparametric models and rasterized predictors, which represents the easiest workflow and leads to the most rapid results, is justified with little loss in accuracy or precision even if no calibration is performed.

Selected modeling methods are compared for predicting 5 forest attributes, basal area (BA), stem volume (VOL), Lorey's height (LOR), quadratic mean diameter (QMD), and tree density (DEN), from airborne LiDAR metrics in southwestern Oregon, in the United States. The selected methods included most similar neighbor (MSN) imputation, gradient nearest neighbor (GNN) imputation, Random Forest (RF)–based imputation, BestNN imputation, ordinary least square (OLS) regression, spatial linear model (SLM), and geographically weighted regression (GWR). Several performances of each method were assessed by 500 simulations with different numbers of training data. No single modeling method was always superior to the others in prediction of the forest attributes. The best method varied according to response variable, prediction type, and performance measures, even though there was a leading group (SLM, OLS, BestNN, and GWR) that always outperformed the other methods in root mean squared prediction error (RMSPE). The model's performance was quite affected when the number of training data used in the modeling procedure was small. The optimal sizes of training data were 100–150 for point prediction and 200–250 for total prediction. SLM showed its applicability to wider conditions in that it produced better performance in most cases. RF imputation produced poorer performances than the other methods, particularly with lower prediction interval coverage. This might be because RF imputation had some bias and a smaller prediction standard error; the poor performance by RF did not stem from the smaller number of predictor variables. Résumé. Des méthodes de modélisation choisies sont comparées pour prédire 5 attributs de la forêt, la surface terrière (BA), le volume de la tige (VOL), la hauteur de Lorey (LOR), le diamètre de la tige de surface terrière moyenne (QMD) et la densité des arbres (DEN), à partir de mesures LiDAR aéroporté dans le sud-ouest de l'Oregon, aux États-Unis. Les méthodes choisies ont inclus l'imputation du voisin le plus similaire (most similar neighbor; MSN), l'imputation du gradient du plus proche voisin (gradient nearest neighbor; GNN), l'imputation basée sur les forêts d'arbres décisionnels (Random Forest; RF), l'imputation BestNN, la régression de la méthode des moindres carrés ordinaire (ordinary least square; OLS), le modèle spatial linéaire (spatial linear model; SLM), et la régression géographiquement pondérée (geographically weighted regression; GWR). Plusieurs essais de chaque méthode ont été évalués par 500 simulations avec des nombres différents de données d'apprentissage. Aucune méthode de modélisation ne fut toujours supérieure aux autres dans la prédiction des attributs forestiers. La meilleure méthode a varié selon la variable de réponse, le type de prédiction et les mesures de performance, même s'il y avait un meilleur groupe de méthodes (SLM, OLS, BestNN, et GWR) qui a toujours mieux performé que les autres en ce qui concerne l'erreur quadratique moyenne de prévision (RMSPE). La performance du modèle a été fortement affectée quand seulement un petit nombre de données d'apprentissage a été utilisé dans la procédure de modélisation. Les nombres optimaux de données d'apprentissage étaient de 100–150 pour des prévisions ponctuelles et 200–250 pour la prédiction totale. Le SLM a montré son applicabilité à des conditions plus étendues, car il a produit une meilleure performance dans la plupart des cas. L'imputation RF a produit des performances inférieures aux autres méthodes, notamment à cause d'une couverture inférieure de l'intervalle de prévision. Ce pourrait être parce que l'imputation RF avait un certain biais et une plus petite erreur standard de prévision. La mauvaise performance de l'imputation RF ne provient pas du plus petit nombre de variables prédictives.

Stands are the primary unit for tactical and operational forest planning. Forest managers can use remote-sensing-based forest inventories to precisely estimate attributes of interest at the stand scale. However, remote-sensing-based inventories typically rely on models relating remote-sensing information to forest attributes for fixed area plots with accurate coordinates. The collection of that kind of ground data is expensive and time-consuming. Furthermore, remote-sensing-based inventories provide precise descriptions of the forest when the remote-sensing data were collected, but they inevitably become outdated as the forest evolves. Fay–Herriot (FH), models can be used with ground information from variable radius plots even if the plot coordinates are unknown. Thus, they provide an efficient way to update old remote-sensing-based inventories or develop new ones when fixed radius plots are unavailable. In addition, FH models are well described in the small-area estimation literature and allow reporting estimation uncertainties, which is key to incorporating quality controls to remote-sensing inventories. We compared two scenarios developed in the Willamette National Forest, OR, United States, to produce stand-level estimates of above-ground biomass (AGB), and Volume (V) for natural and managed stands. The first, Case 1, was developed using auxiliary data from a recent lidar acquisition. The second, Case 2, was developed to update an old remote-sensing-based inventory. Results showed that FH models allowed for improvements in efficiency with respect to direct stand-level estimates obtained using only field data for both case scenarios and both typologies of stands. Average improvements in efficiency in natural stands were 37.36% for AGB and 33.10% for Volume for FH models from Case 1 and 20.19% for AGB and 19.25 for V for Case 2. For managed stands, average improvements for Case 1 were 2.29 and 19.92% for AGB and V, respectively, and for Case 2, improvements were 15.55% for AGB and 16.05% for V.

The use of individual tree detection methods to support forest management inventories has been a research topic for over two decades, but a formal assessment of these methods to produce stand-level and region-level predictions of forest attributes and measures of error is lacking. We employed model-based estimation methods in conjunction with the semi-individual tree crown approach (s-ITC) to produce predictions and measures of error for tree volume (VOL), basal area (BA), stem density (DEN), and quadratic mean diameter (QMD) at the scale of forest stands and the entire study region. We compared the s-ITC approach against the area-based approach (ABA) for predictions of region-level and stand-level attributes via model-based root mean squared errors (RMSEs). The study was conducted at the Panther Creek watershed in Oregon, USA using a set of 78 field plots and aerial lidar information. For region-level attributes, s-ITC RMSEs demonstrated changes between −31% and 17% relative to ABA models. At the stand level, median s-ITC RMSEs generally increased, with changes between −29% and 414% relative to ABA models, but demonstrated important reductions in stands where segmentation provided large increases in sample size and was less prone to extrapolation than ABA models. The ABA demonstrated smaller RMSEs in stands without sampled population units for all variables. Our findings motivate further research into niche applications where s-ITC models may consistently outperform ABA models.