Terje Gobakken

Laser scanning of forest resources: the nordic experience

This article reviews the research and application of airborne laser scanning for forest inventory in Finland, Norway and Sweden. The first experiments with scanning lasers for forest inventory were conducted in 1991 using the FLASH system, a full-waveform experimental laser developed by the Swedish Defence Research Institute. In Finland at the same time, the HUTSCAT profiling radar provided experiences that inspired the following laser scanning research. Since 1995, data from commercially operated time-of-flight scanning lasers (e.g. TopEye, Optech ALTM and TopoSys) have been used. Especially in Norway, the main objective has been to develop methods that are directly suited for practical forest inventory at the stand level. Mean tree height, stand volume and basal area have been the most important forest mensurational parameters of interest. Laser data have been related to field training plot measurements using regression techniques, and these relationships have been used to predict corresponding properties in all forest stands in an area. Experiences from Finland, Norway and Sweden show that retrieval of stem volume and mean tree height on a stand level from laser scanner data performs as well as, or better than, photogrammetric methods, and better than other remote sensing methods. Laser scanning is, therefore, now beginning to be used operationally in large-area forest inventories. In Finland and Sweden, research has also been done into the identification of single trees and estimation of single-tree properties, such as tree position, tree height, crown width, stem diameter and tree species. In coniferous stands, up to 90% of the trees represented by stem volume have been correctly identified from canopy height models, and the tree height has been estimated with a root mean square error of around 0.6 m. It is significantly more difficult to identify suppressed trees than dominant trees. Spruce and pine have been discriminated on a single-tree level with 95% accuracy. The application of densely sampled laser scanner data to change detection, such as growth and cutting, has also been demonstrated.

Model-based inference for biomass estimation in a LiDAR sample survey in Hedmark County, Norway

In forest inventories, regression models are often applied to predict quantities such as biomass at the level of sampling units. In this paper, we propose a model-based inference framework for combining sampling and model errors in the variance estimation. It was applied to airborne laser (LiDAR) data sets from Hedmark County, Norway, where the model error proportion of the total variance was found to be large for both scanning (airborne laser scanning) and profiling LiDAR when biomass was estimated. With profiling LiDAR, the model error variance component for the entire county was as large as 71% whereas for airborne laser scanning, it was 43% of the total variance. Partly, this reflects the better accuracy of the pixel-based regression models estimated from scanner data as compared with the models estimated from profiler data. The framework proposed in our study can be applied in all types of sample surveys where model-based predictions are made at the level of individual sampling units. Especially, it should be useful in cases where model-assisted inference cannot be applied due to the lack of a probability sample from the target population or due to problems of correctly matching observations of auxiliary and target variables.

Comparing stand inventories for large areas based on photo-interpretation and laser scanning by means of cost-plus-loss analyses

Evaluations of inventory methods usually end when precision and bias are quantified. Additional information on the appropriateness of a method may be provided through cost-plus-loss analyses, where the total costs are calculated as the sum of net present value (NPV) losses, i.e. expected economic losses as a result of future incorrect decisions due to errors in measurements, and inventory costs. The aim of the study was to compare inventories of basal area, dominant height and number of trees per hectare based on photo-interpretation and laser scanning from two sites in Norway by means of cost-plus-loss analyses. In general, more precise estimates were found for laser scanning than for photo-interpretation, while the biases were about equally distributed between the two methods. On average for the two sites, the inventory costs, NPV losses and total costs for photo-interpretation were about 6, 49 and 54 euros ha−1, respectively, while they were 11, 13 and 25 euros ha−1 for laser scanning. The data used for the comparison were limited to two sites and 77 stands, and certain simplifying assumptions were made in the cost-plus-loss analyses. Still, there is reason to believe that the results of the study are of general validity with respect to the main conclusion when comparing the two methods.

Tree Species Classification in Boreal Forests With Hyperspectral Data

Tree species mapping in forest areas is an important topic in forest inventory. In recent years, several studies have been carried out using different types of hyperspectral sensors under various forest conditions. The aim of this work was to evaluate the potential of two high spectral and spatial resolution hyperspectral sensors (HySpex-VNIR 1600 and HySpex-SWIR 320i), operating at different wavelengths, for tree species classification of boreal forests. To address this objective, many experiments were carried out, taking into consideration: 1) three classifiers (support vector machines (SVM), random forest (RF), and Gaussian maximum likelihood); 2) two spatial resolutions (1.5 m and 0.4 m pixel sizes); 3) two subsets of spectral bands (all and a selection); and 4) two spatial levels (pixel and tree levels). The study area is characterized by the presence of four classes 1) Norway spruce, 2) Scots pine, together with 3) scattered Birch and 4) other broadleaves. Our results showed that: 1) the HySpex VNIR 1600 sensor is effective in boreal tree species classification with kappa accuracies over 0.8 (with Pine and Spruce reaching producers accuracies higher than 95%); 2) the role of the HySpex-SWIR 320i is limited, and its bands alone are able to properly separate only Pine and Spruce species; 3) the spatial resolution has a strong effect on the classification accuracy (an overall decrease of more than 20% between 0.4 m and 1.5 m spatial resolution); and 4) there is no significant difference between SVM or RF classifiers.

Estimation of diameter and basal area distributions in coniferous forest by means of airborne laser scanner data

Diameter and basal area distributions are used in many forest management planning packages for predicting stand volume and growth. The distribution parameters and the 24 and 93 percentiles for parameter recovery of a two-parameter Weibull were derived for empirical diameter and basal area distributions of 54 plots of 3740 m2 each. Regression analysis was used to relate the distribution parameters and percentiles to various canopy height and canopy density metrics derived from airborne laser scanner data over young and mature coniferous forest. On average, the distance between transmitted laser pulses was 1.0 m on the ground. Aerial photo-interpretation was used to divide the plots into three strata according to age class and site quality. The stratum-specific regressions explained 20–93% of the variability in the observed percentiles. Total plot volume predicted from the estimated distributions was used to assess the accuracy of the regressions. Cross-validation of the regressions revealed a bias of −4.8 to 2.7% between predicted and ground-truth values of plot volume when the predicted frequencies of the diameter and basal area distributions were scaled to ground-truth stem number (N) and basal area (G), respectively. The standard deviations (SD) of the differences between predicted and ground-truth values of plot volume were 5.6–29.1%. However, when the scaling variables (N and G) were predicted from the laser data, the bias of plot volume determined by cross-validation was −4.7 to 6.6% and the SD was 11.4–24.2%.

Inventory of Small Forest Areas Using an Unmanned Aerial System

Acquiring high spatial and temporal resolution imagery from small unmanned aerial systems (sUAS) provides new opportunities for inventorying forests at small scales. Only a few studies have investigated the use of UASs in forest inventories, and the results are inconsistent and incomplete. The present study used three-dimensional (3D) variables derived from UAS imagery in combination with ground reference data to fit linear models for Lorey’s mean height (hL), dominant height (hdom), stem number (N), basal area (G), and stem volume (V). Plot-level cross validation revealed adjusted R2 values of 0.71, 0.97, 0.60, 0.60, and 0.85 for hL, hdom, N, G, and V, respectively, with corresponding RMSE values of 1.4 m, 0.7 m, 538.2 ha−1, 4.5 m2∙ha−1, and 38.3 m3∙ha−1. The respective relative RMSE values were 13.3%, 3.5%, 39.2%, 15.4%, and 14.5% of the mean ground reference values. The mean predicted values did not differ significantly from the reference values. The results revealed that the use of UAS imagery can provide relatively accurate and timely forest inventory information at a local scale. In addition, the present study highlights the practical advantages of UAS-assisted forest inventories, including adaptive planning, high project customization, and rapid implementation, even under challenging weather conditions.

Weibull and percentile models for lidar-based estimation of basal area distribution

Abstract The aim of this study was to assess the accuracy of basal area distributions of sample plots in coniferous forest derived from small-footprint airborne laser scanner data, and to compare the accuracy of two methods for derivation of such distributions based on: (1) two percentiles of a two-parameter Weibull and parameter recovery, and (2) a system of 10 percentiles defined across the range of observed diameters. The 12 percentiles were derived from the empirical basal area distributions of 141 plots with size 300–400 m2. Regression analysis was used to relate the percentiles to various canopy height and canopy density metrics derived from the laser data. On average, the distance between transmitted laser pulses was 0.9 m on the ground. The plots were divided into three strata according to age class and site quality. The stratum-specific regressions explained 7–91% of the variability. Total plot volume predicted from the estimated distributions was used to assess the accuracy of the regressions. Cross-validation of the regressions revealed a bias of −1.2 to 2.1% between predicted and ground-truth values of plot volume. The standard deviations of the differences between predicted and ground-truth values of plot volume were 15.1–16.4%. Neither bias nor standard deviation differed significantly between the two validated methods.

Estimating Quebec provincial forest resources using ICESat/GLAS

Ground plots, airborne profiling and space lidar (light detection and ranging) measurements of canopy height and crown closure, space radar topographic data, a Landsat cover type map, and a vegetation zone map were used in a model-assisted, two-phase sampling design to estimate the aboveground biomass and carbon resources of Quebec. It was determined that a simple random sampling estimator, with covariance terms added, could be used to quantify the variabil- ity of regional Geoscience Laser Altimeter System (GLAS) biomass estimates where interorbit distances are, on average, ‡15 km apart. Prediction error increased standard errors, on average, 24.4%, 4.6%, and 2.8% at the cover type, vegetation zone, and provincial levels, respectively. Inclusion of the covariance term in the calculation of grouped cover type variances increased the vegetation zone standard errors up to 3.7 times and the provincial standard errors 15.6 times. In the southern commercial forests of Quebec, GLAS underestimated ground-based biomass values by 7.3% (stratified lin- ear model) and 10.2% (nonstratified linear model). Quebec forests support 2.57 ± 0.33 gigatonnes of carbon (nonstratified linear model). Approximately 25% of that carbon was found to be located in two southern vegetation zones (northern hard- wood and mixedwood), another 25% in two northern vegetation zones (taiga and treed tundra), and the remaining 50% in the boreal zone.

Single tree detection in heterogeneous boreal forests using airborne laser scanning and area-based stem number estimates

Adaptive single tree detection methods using airborne laser scanning (ALS) data were investigated and validated on 40 large plots sampled from a structurally heterogeneous boreal forest dominated by Norway spruce and Scots pine. Under the working assumption of having uniformly distributed tree locations, area-based stem number estimates were used to guide tree crown delineation from rasterized laser data in two ways: (1) by controlling the amount of smoothing of the canopy height model and (2) by obtaining an appropriate spatial resolution for representing the forest canopy. Single tree crowns were delineated from the canopy height models (CHMs) using a marker-based watershed algorithm, and the delineation results were assessed using a simple tree crown delineation algorithm as a reference method (‘RefMeth’). Using the proposed methods, approximately 46–50% of the total number of trees were detected, while approximately 5–6% false positives were found. The detection rate was, in general, higher for Scots pine than for Norway spruce. The accuracy of individual tree variables (total height and crown width) extracted from the laser data was compared with field-measured data. The individual tree heights were better estimated for deciduous tree species than for the coniferous species Norway spruce and Scots pine. The estimation of crown diameters for Scots pine and deciduous species achieved comparable accuracy, being better than for Norway spruce. The proposed methodology has the potential for easy integration with operational laser scanner-based stand inventories.

Non-parametric prediction of diameter distributions using airborne laser scanner data

Abstract The aim of this study was to apply the non-parametric k-most similar neighbour (MSN) method and airborne laser scanner data to predict stand diameter distributions in a 960 km2 forest district in south-eastern Norway. The specific objectives of the study were (1) to examine the use of different dependent and independent variables in the canonical correlation analysis of MSN, and (2) to examine the influence of reduced number of training data plots by means of simulations. The reliability of the constructed diameter distributions was analysed using error indices and the accuracy of stand attributes derived from predicted diameter distributions. The study material included a total of 201 plots and they were reduced to 181, 161, … , 41 plots in the simulations. The results indicated that when selecting dependent variables in the canonical correlation analysis it is sufficient to have variables reflecting stand means and aggregated variables (sums) to obtain accurate predictions of diameter distributions. Furthermore, the prediction models should not to be too detailed, i.e. they should not include a great number of independent variables since cross-validation always tends to give too optimistic results. Validation on independent data will often show considerably poorer reliability figures. Finally, the results indicated that even such a low number of training plots as about 100 can produce accurate enough predictions of stand attributes and diameter distributions.