In this study, for five sites around the world, we look at the effects of different model types and variable selection approaches on forest yield modelling performances in an area-based approach (ABA). We compared ordinary least squares regression (OLS), k-nearest neighbours (kNN) and random forest (RF). Our objective was to test if there are systematic differences in accuracy between OLS, kNN and RF in ABA predictions of growing stock volume. The analyses are based on a 5-fold cross-validation at five study sites: an eucalyptus plantation, a temperate forest and three different boreal forests. Two completely independent validation datasets were also available for two of the boreal sites. For the kNN, we evaluated multiple measures of distance including Euclidean, Mahalanobis, most similar neighbour (MSN) and an RF-based distance metric. The variable selection approaches we examined included a heuristic approach (for OLS, kNN and RF), exhaustive search among all combinations (OLS only) and all variables together (RF only). Performances varied by model type and variable selection approaches among sites. OLS and RF had similar accuracies and were more efficient than any of the kNN variants. Variable selection did not affect RF performance. Heuristic and exhaustive variable selection performed similarly for OLS. kNN fared the poorest amongst model types, and kNN with RF distance was prone to overfitting when compared with a validation dataset. Additional caution is therefore required when building kNN models for volume prediction though ABA, being preferable instead to opt for models based on OLS with some variable selection, or RF with all variables together.

Airborne Laser Scanners (ALS) and Terrestrial Laser Scanners (TLS) are two lidar systems frequently used for remote sensing forested ecosystems. The aim of this study was to compare crown metrics derived from TLS, ALS, and a combination of both for describing the crown structure and fuel attributes of longleaf pine (Pinus palustris Mill.) dominated forest located at Eglin Air Force Base (AFB), Florida, USA. The study landscape was characterized by an ALS and TLS data collection along with field measurements within three large (1963 m² each) plots in total, each one representing a distinct stand condition at Eglin AFB. Tree-level measurements included bole diameter at breast height (DBH), total height (HT), crown base height (CBH), and crown width (CW). In addition, the crown structure and fuel metrics foliage biomass (FB), stem branches biomass (SB), crown biomass (CB), and crown bulk density (CBD) were calculated using allometric equations. Canopy Height Models (CHM) were created from ALS and TLS point clouds separately and by combining them (ALS + TLS). Individual trees were extracted, and crown-level metrics were computed from the three lidar-derived datasets and used to train random forest (RF) models. The results of the individual tree detection showed successful estimation of tree count from all lidar-derived datasets, with marginal errors ranging from −4 to 3%. For all three lidar-derived datasets, the RF models accurately predicted all tree-level attributes. Overall, we found strong positive correlations between model predictions and observed values (R² between 0.80 and 0.98), low to moderate errors (RMSE% between 4.56 and 50.99%), and low biases (between 0.03% and −2.86%). The highest R² using ALS data was achieved predicting CBH (R² = 0.98), while for TLS and ALS + TLS, the highest R² was observed predicting HT, CW, and CBD (R² = 0.94) and HT (R² = 0.98), respectively. Relative RMSE was lowest for HT using three lidar datasets (ALS = 4.83%, TLS = 7.22%, and ALS + TLS = 4.56%). All models and datasets had similar accuracies in terms of bias (<2.0%), except for CB in ALS (−2.53%) and ALS + TLS (−2.86%), and SB in ALS + TLS data (−2.22%). These results demonstrate the usefulness of all three lidar-related methodologies and lidar modeling overall, along with lidar applicability in the estimation of crown structure and fuel attributes of longleaf pine forest ecosystems. Given that TLS measurements are less practical and more expensive, our comparison suggests that ALS measurements are still reasonable for many applications, and its usefulness is justified. This novel tree-level analysis and its respective results contribute to lidar-based planning of forest structure and fuel management.

The level of spatial co-registration between airborne laser scanning (ALS) and ground data can determine the goodness of the statistical inference used in forest inventories. The importance of positioning methods in the field can increase, depending on the structural complexity of forests. An area-based approach was followed to conduct forest inventory over seven National Forest Inventory (NFI) forest strata in Spain. The benefit of improving the co-registration goodness was assessed through model transferability using low- and high-accuracy positioning methods. Through the inoptimality losses approach, we evaluated the value of good co-registered data, while assessing the influence of forest structural complexity. When using good co-registered data in the 4th NFI, the mean tree height (HT_mean), stand basal area (G) and growing stock volume (V) models were 2.6%, 10.6% and 14.7% (in terms of root mean squared error, RMSE %), lower than when using the coordinates from the 3rd NFI. Transferring models built under poor co-registration conditions using more precise data improved the models, on average, 0.3%, 6.0% and 8.8%, while the worsening effect of using low-accuracy data with models built in optimal conditions reached 4.0%, 16.1% and 16.2%. The value of enhanced data co-registration varied between forests. The usability of current NFI data under modern forest inventory approaches can be restricted when combining with ALS data. As this research showed, investing in improving co-registration goodness over a set of samples in NFI projects enhanced model performance, depending on the type of forest and on the assessed forest attributes.

Estimating forest inventory variables is important in monitoring forest resources and mitigating climate change. In this respect, forest managers require flexible, non-destructive methods for estimating volume and biomass. High-resolution and low-cost remote sensing data are increasingly available to measure three-dimensional (3D) canopy structure and to model forest structural attributes. The main objective of this study was to evaluate and compare the individual tree volume estimates derived from high-density point clouds obtained from airborne laser scanning (ALS) and digital aerial photogrammetry (DAP) in Eucalyptus spp. plantations. Object-based image analysis (OBIA) techniques were applied for individual tree crown (ITC) delineation. The ITC algorithm applied correctly detected and delineated 199 trees from ALS-derived data, while 192 trees were correctly identified using DAP-based point clouds acquired from Unmanned Aerial Vehicles (UAV), representing accuracy levels of respectively 62% and 60%. Addressing volume modelling, non-linear regression fit based on individual tree height and individual crown area derived from the ITC provided the following results: Model Efficiency (Mef) = 0.43 and 0.46, Root Mean Square Error (RMSE) = 0.030 m³ and 0.026 m³, rRMSE = 20.31% and 19.97%, and an approximately unbiased results (0.025 m³ and 0.0004 m³) using DAP and ALS-based estimations, respectively. No significant difference was found between the observed value (field data) and volume estimation from ALS and DAP (p-value from t-test statistic = 0.99 and 0.98, respectively). The proposed approaches could also be used to estimate basal area or biomass stocks in Eucalyptus spp. plantations.

Semi- and nonparametric models are popular in the area-based approach (ABA) using airborne laser scanning. It is unclear, however, how many predictors and training plots are needed to provide accurate predictions without overfitting. This work aims to explore these limits for various approaches: ordinary least squares regression (OLS), generalized additive models (GAM), least absolute shrinkage and selection operator (LASSO), random forest (RF), support vector machine (SVM), and Gaussian process regression (GPR). We modeled timber volume (m³·ha^–1) for four boreal sites using ABA with 2–39 predictors and 20–500 training plots. OLS, GAM, LASSO, and SVM overfitted as the number of predictors approached the number of training plots. They required ≥15 plots per predictor to provide accurate predictions (RMSE ≤30%). GAM required ≥250 plots regardless of the number of predictors. The number of predictors only mildly affected RF and GPR, but they required ≥200 and ≥250 training plots, respectively. RF did not overfit in any circumstances, whereas GPR overfit even with 500 training plots. Overall, using up to 39 predictors did not generally result in overfit, and for most model types, it resulted in better accuracy for sufficiently large datasets (≥250 plots).

Several methods have been proposed to perform site classification for timber production. However, there is frequent need to assess site productive capacity before forest establishment. This has motivated the application of Artificial Neural Networks (ANN) for site classification. Hereby, the traditional guide curve (GC) procedure was compared to the ANN with no stand measures as input. In addition, different ANN settings were tested to assess the best setting. The variables used to train the ANN were: climatic variables, soil types, spacing and genetic material. The results from the ANN and the GC methods were compared to the observed classes, which were defined using the observed dominant high at the age of seven years. The comparison was performed using the Kappa coefficient (K) and descriptive analysis. The results showed that the cost function “Cross Entropy” and the output activation function “Softmax” were the best for this purpose. The ANN classification resulted in substantial agreement with the observed indices against a moderate agreement of the GC procedure. The change in growth patterns throughout the rotation may have hindered the proper classification by the CG method, which does not happen with the ANN. Moreover, the GC method shows efficiency on classification in cases which data from stands at the age close to the reference age are available. Also, it could be possible to improve its accuracy if another advanced regression techniques were applied. However, the ANN method presented here is not sensible to growth instability and allows classifying sites with no plantation history.