Testing the usability of truncated angle count sample plots as ground truth in airborne laser scanning-based forest inventories

Airborne laser scanner (ALS)-based forest inventory method usuallyadopt a laser canopy height distribution approach in which forestcharacteristics are predicted using measures such as percentilesof the distribution of laser canopy heights across a fixed area.The method requires a ground-truth sample of accurately measuredfield plots. One possibility for reducing the costs lies inthe use of existing field plots for ground-truth purposes. Themost obvious alternative in Finland would be to use truncatedangle count sample plots of the National Forest Inventory ormore locally data of checking of inventory by compartments.Due to the lack of suitable angle count ground-truth data andcorresponding laser data, we tested this possibility using dataon fixed-area sample plots, in which tree locations were simulated.The trees for a truncated angle count sample plot were thenchosen and the resulting data together with the characteristicsof an ALS-based canopy height distribution were used to constructregression models to predict stem volume, basal area, stem number,basal area median diameter and height. The accuracy of the standattributes was found to be almost as good as in the case ofmodels of fixed-area plots.     

Estimating percentile-based diameter distributions in uneven-sized Norway spruce stands using airborne laser scanner data

Abstract

A model for prediction of stand basal area and diameters at 10 percentiles of a basal area distribution was estimated from small-footprint laser scanner data from primeval conifer forest using partial least squares regression. The regression explained 44–80% and 67% of the variability of the 10 percentiles and stand basal area, respectively. The predicted percentiles, scaled by the predicted stand basal area, were used to compute diameter distributions. A cross-validation showed that the mean differences between the predicted and observed number of stems by diameter class were non-significant (p>0.05) for 22 of 29 diameter classes. Moreover, plot volume was calculated from the predicted diameter distribution and cross-validation revealed a non-significant deviation between predicted and observed volume of ?3.3% (of observed volume). An independent validation showed non-significant mean differences for 20 of 21 diameter classes for data corresponding to the model calibration data. Plot volumes calculated from the predicted diameter distributions deviated from observed volume by ?4.4%. The model reproduced diameter distributions corresponding to the model calibration data (uneven-sized forest) well. However, the model is not flexible enough to reproduce normal and uniform diameter distributions. Volume estimates derived from predicted diameter distributions were generally well determined, irrespective of the observed distribution.     

Estimation of stem volume using laser scanning-based canopy height metrics

The aim of this study was to test different stem volume predictorsthat are capable of utilizing laser scanning-based canopy heightmetrics as independent variables. The three laser scanning-basedmethods compared were (1) a direct prediction model for thestem volume at plot level, (2) a volume prediction system basedon the modelled percentiles of the basal area diameter distribution,and (3) a parameter prediction method used to determinate Weibull-basedbasal area diameter distributions for the plot-level stem volumeprediction. The predicted volumes were also compared with field-measuredvolumes obtained with the Finnish conventional inventory bycompartments. The best results were obtained with the firstmethod, i.e. the model that predicts plot-level stem volumesdirectly, which is logical. Furthermore, the simulated reductionof point density of laser data had no effect on the accuracyof stem volume predictions. The percentile-based modelling ofdiameter distributions was applied, in particular, to the determinationof non-homogenous stand structure; using this method, it iseven possible to fit multimodal distributions. In terms of theaccuracy of the predicted plot-level stem volumes, the volumeprediction method based on modelled percentiles of basal areadiameter distributions was the second best, whereas the volumeprediction method based on the parameter prediction of the Weibull-basedbasal area diameter distributions resulted in slightly worseresults. However, the accuracies of the three laser-based volumeprediction methods tested were superior to the published resultsof spectral value-based remote sensing studies implemented usingdata collected from Finland. Furthermore, the accuracy of plot-levelstem volume estimates calculated from field assessments wasconsiderably weaker than the accuracy of the three volume predictionmethods that utilized measures obtained with laser scanning.     

Practical large-scale forest stand inventory using a small-footprint airborne scanning laser

Mean tree height, dominant height, mean diameter, stem number, basal area and timber volume of 116 georeferenced field sample plots were estimated from various canopy height and canopy density metrics derived by means of a small-footprint laser scanner over young and mature forest stands using regression analysis. The sample plots were distributed systematically throughout a 6500 ha study area, and the size of each plot was 232.9 m². Regressions for coniferous forest explained 60–97% of the variability in ground reference values of the six studied characteristics. A proposed practical two-phase procedure for prediction of corresponding characteristics of entire forest stands was tested. Fifty-seven test plots within the study area with a size of approximately 3740 m² each were divided into 232.9 m² regular grid cells. The six examined characteristics were predicted for each grid cell from the corresponding laser data using the estimated regression equations. Average values for each test plot were computed and compared with ground-based estimates measured over the entire plot. The bias and standard deviations of the differences between predicted and ground reference values (in parentheses) of mean height, dominant height, mean diameter, stem number, basal area and volume were ?0.58 to ?0.85 m (0.64–1.01 m), ?0.60 to ?0.99 m (0.67–0.84 m), 0.15–0.74 cm (1.33–2.42 cm), 34–108 ha^?1 (97–466 ha^?1), 0.43–2.51 m² ha^?1 (1.83–3.94 m² ha^?1) and 5.9–16.1 m³ ha^?1 (15.1–35.1 m³ ha^?1), respectively.     

Airborne laser scanning as a method in operational forest inventory: Status of accuracy assessments accomplished in Scandinavia

Abstract

This research reports the major evaluation results from an operational stand-based forest inventory using airborne laser scanner data carried out in Norway. This is the first operational inventory in which data from two separate districts are combined. Laser data from two forest areas of 65 and 110 km² were used to predict six biophysical stand variables used in forest planning. The predictions were based on regression equations estimated from 250 m² field training plots distributed systematically throughout the two forest areas. Test plots with a size of 0.1 ha were used for validation. The testing revealed standard deviations between ground-truth values and predicted values of 0.58–0.85 m (3.4–5.6%) for mean and dominant heights, 2.62–2.87 m² ha^?1 (9.3–14.3%) for basal area, and 18.7–25.1 m³ ha^?1 (10.8–12.8%) for stand volume. No serious bias was detected. For 10 of the 12 estimated regression models there were no significant effects of district.     

Accuracy of forest inventory using airborne laser scanning: evaluating the first nordic full-scale operational project

This research reports the major results from an evaluation of the first Nordic operational stand-based forest inventory using airborne laser scanner data. Laser data from a forest area of 250 km² were used to predict six biophysical stand variables used in forest planning. The predictions were based on regression equations estimated from 250 m² field training plots distributed systematically throughout the forest area. Test plots with an approximate size of 0.1–0.4 ha were used for validation. The testing revealed standard deviations between ground-truth values and predicted values of 0.36–1.37 m (1.9–7.6%) for mean height, 0.70–1.55 m (3.0–7.6%) for dominant height, 2.38–4.88 m² ha^?1 (7.8–14.2%) for basal area and 13.9–45.9 m³ ha^?1 (6.5–13.4%) for stand volume. No serious bias was detected.     

Assessment of bias due to random measurement errors in stem volume growth estimation by the Swedish National Forest Inventory

Abstract

We evaluated the performance of two methods for estimating stem volume increment at individual tree level with respect to bias due to random measurement errors. Here, growth is either predicted as the difference between two consecutive volume estimates where single-tree volume functions are applied to data from repeated measurements or by a regression model that is applied to data from a single survey and includes radial increment. In national forest inventories (NFIs), the first method is typically used for permanent plots, the second for temporary plots. The Swedish NFI combines estimates from both plot types to assess growth at national and regional scales and it is, therefore, important that the two methods provide similar results. The accuracy of these estimates is affected by random measurement errors in the independent variables, which may lead to systematic errors in predicted variables due to model non-linearity. Using Taylor series expansion and empirical data from the Swedish NFI we compared the expected bias in stem volume growth estimates for different diameter classes of Scots pine (Pinus sylvestris L.) and Norway spruce (Picea abies (L.) Karst.). Our results indicate that both methods are fairly insensitive to random measurement errors of the size that occur in the Swedish NFI. The empirical comparison between the two methods showed greater differences for large diameter trees of both pine and spruce. A likely explanation is that the regressions are uncertain because few large trees were available for developing the models.     

Effects of field plot configurations on the uncertainties of ALS-assisted forest resource estimates

The effects of field plot configurations on the uncertainties of plot-level forest resource estimates were analyzed using airborne laser scanner data, aerial photographs and field measurements. The aim was to select a field sample plot configuration that can be used for both large area and management inventories. Error estimates were evaluated at the plot level using six different training plot configurations. Additionally, separate plots with two different sizes were used for evaluation. Stem volume and five other forest resource characteristics were considered. The field measurement costs of the different plot configurations were also studied. RMSEs and mean deviations for airborne laser scanning ALS-assisted estimates were practically the same for the fixed radius plot, the two concentric plots and the angle count plot with a basal area factor of q = 1 for all three evaluation plot sizes. Angle count plots with basal area factors of q = 1.5 and 2 increased the RMSEs. For the former plot configurations, the RMSEs for the ALS-assisted estimates could be attributed to inaccuracy in the predicted relationships between the field data and ALS data, not to the training plot configuration. Tree measurements and costs can, therefore, be reduced from those of the Finnish management inventories without increasing RMSEs.     

Effect of forest structure and health on the relative surface temperature captured by airborne thermal imagery – Case study in Norway Spruce-dominated stands in Southern Finland

The effect of forest structure and health on the relative surface temperature captured by airborne thermal imagery was investigated in Norway Spruce-dominated stands in Southern Finland. Airborne thermal imagery, airborne scanning light detection and ranging (LiDAR) data and 92 field-measured sample plots were acquired at the area of interest. The surface temperature correlated most negatively with the logarithm of stem volume, Lorey’s height and the logarithm of basal area at a resolution of 254?m² (9?m radius). LiDAR-derived metrics: the standard deviations of the canopy heights, canopy height (upper percentiles and maximum height) and canopy cover percentage were most strongly negatively correlated with the surface temperature. Although forest structure has an effect on the detected surface temperature, higher temperatures were detected in severely defoliated canopies and the difference was statistically significant. We also found that the surface temperature differences between the segmented canopy and the entire plot were greater in the defoliated plots, indicating that thermal images may also provide some additional information for classifying forests health status. Based on our results, the effects of forest structure on the surface temperature captured by airborne thermal imagery should be taken into account when developing forest health mapping applications using thermal imagery.     

Combining ALS and NFI training data for forest management planning: a case study in Kuortane,Western Finland

Forest inventories based on airborne laser scanning (ALS) have already become common practice in the Nordic countries. One possibility for improving their cost effectiveness is to use existing field data sets as training data. One alternative in Finland would be the use of National Forest Inventory (NFI) sample plots, which are truncated angle count (relascope) plots. This possibility is tested here by using a training data set based on measurements similar to the Finnish NFI. Tree species-specific stand attributes were predicted by the non-parametric k most similar neighbour (k-MSN) approach, utilising both ALS and aerial photograph data. The stand attributes considered were volume, basal area, stem number, mean age of the tree stock, diameter and height of the basal area median tree, determined separately for Scots pine, Norway spruce and deciduous trees. The results obtained were compared with those obtained when using training data based on observations from fixed area plots with the same centre point location as the NFI plots. The results indicated that the accuracy of the estimates of stand attributes derived by using NFI training data was close to that of the fixed area plot training data but that the NFI sampling scheme and the georeferencing of the plots can cause problems in practical applications.     

Estimating Stem Volume and Basal Area in Forest Compartments by Combining Satellite Image Data with Field Data

The weighted k - nearest neighbour (kNN) method was used for estimating stem volume (m3 ha?¹) and basal area (m2 ha?¹) on a compartment level (average 19 ha) by combining satellite image data with measurements from Swedish National Forest (NFI) inventory plots. In the kNN method each estimation location (target plot) is assigned a value that is an average, which is weighted, of the attribute data from the k closest reference plots (NFI plots). The distance between target and reference plot was measured on different scales, which were transforms of spectral values and/or ancillary data. The standard error (assuming bias with no trend) of stem volume estimates in the compartments was 36% using only spectral data. This estimation accuracy improved to 17% if site index, age of the forest and mean tree height (ancillary data) were known for the compartments. Low volumes were overestimated and high volumes underestimated. This bias was reduced if ancillary data were added but was also dependent on the transform of the original scales.     

Determination of Mean Tree Height of Forest Stands by Digital Photogrammetry

The mean tree height of 73 forest stands in a 1000 ha forest area was determined from canopy heights generated by automatic image matching using a digital photogrammetric workstation and digitized panchromatic aerial photographs with a scale of 1:15 000. First, the mean height of each stand was computed as the arithmetic mean of the quantile corresponding to the 75th percentile of the distribution of the canopy heights from the image matching within square grid cells with cell sizes of 236-400 m². The mean heights from the image matching underestimated the true heights by 5.42 m. Secondly, field-measured mean tree heights of 165 georeferenced sample plots distributed systematically throughout the 1000 ha forest area were regressed against the mean heights derived from the image matching. The regression equations were used to predict the mean heights of the 73 stands. In very young forest stands, the predicted mean heights overestimated the true heights by 0.4 m and the precision was 0.9-1.0 m. In young and mature stands, the average difference between predicted height and ground-truth ranged between -1.6 and 0.5 m, and the precision ranged from 1.1 to 2.1 m.     

Comparison of nearest neighbour approaches for small area estimation of tree species-specific forest inventory attributes in central Europe using airborne laser scanner data

The three nonparametric k nearest neighbour (kNN) approaches, most similar neighbour inference (MSN), random forests (RF) and random forests based on conditional inference trees (CF) were compared for spatial predictions of standing timber volume with respect to tree species compositions and for predictions of stem number distributions over diameter classes. Various metrics derived from airborne laser scanning (ALS) data and the characteristics of tree species composition obtained from coarse stand level ground surveys were applied as auxiliary variables. Due to the results of iterative variable selections, only the ALS data proved to be a relevant predictor variable set. The three applied NN approaches were tested in terms of bias and root mean squared difference (RMSD) at the plot level and standard errors at the stand level. Spatial correlations were considered in the statistical models. While CF and MSN performed almost similarly well, large biases were observed for RF. The obtained results suggest that biases in the RF predictions were caused by inherent problems of the RF approach. Maps for Norway spruce and European beech timber volume were exemplarily created. The RMSD values of CF at the plot level for total volume and the species-specific volumes for European beech, Norway spruce, European silver fir and Douglas fir were 32.8, 80.5, 99.0, 137.0 and 261.1%. These RMSD values were smaller than the standard deviation, although Douglas fir volume did not belong to the actual response variables. All three non-parametric approaches were also capable of predicting diameter distributions. The standard errors of the nearest neighbour predictions on the stand level were generally smaller than the standard error of the sample plot inventory. In addition, the employed model-based approach allowed kNN predictions of means and standard errors for stands without sample plots.     

Estimation of above ground forest biomass from airborne discrete return laser scanner data using canopy-based quantile estimators

A conceptual model describing why laser height metrics derived from airborne discrete return laser scanner data are highly correlated with above ground biomass is proposed. Following from this conceptual model, the concept of canopy-based quantile estimators of above ground forest biomass is introduced and applied to an uneven-aged, mature to overmature, tolerant hardwood forest. Results from using the 0th, 25th, 50th, 75th and 100th percentiles of the distributions of laser canopy heights to estimate above ground biomass are reported. A comparison of the five models for each dependent variable group did not reveal any overt differences between models with respect to their predictive capabilities. The coefficient of determination (r ² ) for each model is greater than 0.80 and any two models may differ at most by up to 9%. Differences in root-mean-square error (RMSE) between models for above ground total, stem wood, stem bark, live branch and foliage biomass were 8.1, 5.1, 2.9, 2.1 and 1.1 Mg ha^?1, respectively.     

Overstorey density influence on the height of Picea abies regeneration in northern Sweden

The study aimed specifically at investigating if canopy opennesswas a better predictor of the height growth of Norway spruce(Picea abies (L.) Karst.) advance regeneration than overstoreybasal area or overstorey standing volume. In 1990, a field experimentwith 3 x 2 factorial design and two replications (blocks) wasestablished in an uneven-aged Norway spruce forest. Plots hada net plot area of 30 x 30 m, each with a 10-m-wide treatedbuffer zone. Three overstorey density levels retained approximately15, 40 and 70 per cent of the pre-harvest overstorey standingvolume and were allotted to the plots. Two types of thinningthat harvested smaller trees or harvested larger trees wererandomly allocated to each pair of overstorey density plots.In mid-June 2000, canopy openness was estimated from hemisphericalphotographs taken at five marked points in the centre of eachof the plots at 0.9 m from ground to the top of the ‘fish-eye’camera lens. Regression results showed that canopy opennesswas a better predictor of height increments of spruce seedlings(0.1< height < 0.5 m), saplings (0.5 height < 2.0m), and small trees (height 2.0 m, diameter at 1.3 m height< 5 cm) than with overstorey basal area (m² ha^–1) oroverstorey standing volume (m³ ha^–1). The height incrementof the spruce advance regeneration was not significantly correlatedto stand basal area or to standing volume. Overstorey basalarea in the net plots was significantly negative (P 0.05) withmean canopy openness estimates, and the r² value was 0.40. Resultsindicated that basal area was not linearly related to canopyopenness as it increased, which might explain the lack of predictivepower of retained basal area on spruce regeneration height indense stands in boreal Sweden.     

Prediction of tree height,basal area and stem volume in forest stands using airborne laser scanning

The aim of this study was to develop prediction models using laser scanning for estimation of forest variables at plot level, validate the estimations at stand level (area 0.64 ha) and test the effect of different laser measurement densities on the estimation errors. The predictions were validated using 29 forest stands (80×80 m²), each containing 16 field plots with a 10 m radius. For the best tested case, mean tree height, basal area and stem volume were predicted with a root mean square error of 0.59 m (3% of average value), 2.7 m² ha^?1 (10% of average value) and 31 m³ ha^?1 (11% of average value), respectively, at stand level. There were small differences in terms of prediction errors for different measuring densities. The results indicate that mean tree height, basal area and stem volume can be estimated in small stands with low laser measurement densities producing accuracies similar to traditional field inventories.     

Predicting the population dynamics of three understory broad-leaved evergreen species under the influence of Sika deer in primary and secondary forests of mid-western Japan

We examined the population dynamics of three broad-leaved tree species with different susceptibilities to deer predation. Simulation analysis was conducted using a size-structured matrix model for a primary forest plot (PP) and a secondary forest plot (SP) with 56% and 12% evergreen conifer composition in the canopy, respectively. In both plots, populations of Neolitsea sericea, a species that is susceptible to deer predation, initially declined significantly but eventually leveled off. The number of small stems decreased, while that of larger stems increased, indicating that the population dynamics of N. sericea are strongly affected by browsing pressure and that the number of large trees is important for population maintenance. When we examined two deer-resistant species, Pieris japonica and Illicium anisatum, the population of P. japonica increased in the SP and decreased in the PP, whereas that of I. anisatum increased in both plots, likely because mortality tends to increase in persistently dark environments. No significant difference was observed between the present and predicted size distributions of resistant species in the PP. Competition for resources is expected to intensify in the SP as a result of the predicted increase in large stems of the resistant species I. anisatum. Therefore, a specific conservation and management strategy for tree species should be considered for each forest type under the influence of Sika deer.     

Using pre-classification to improve the accuracy of species-specific forest attribute estimates from airborne laser scanner data and aerial images

The aim of this study was to examine whether pre-classification (stratification) of training data according to main tree species and stand development stage could improve the accuracy of species-specific forest attribute estimates compared to estimates without stratification using k-nearest neighbors (k-NN) imputations. The study included training data of 509 training plots and 80 validation plots from a conifer forest area in southeastern Norway. The results showed that stratification carried out by interpretation of aerial images did not improve the accuracy of the species-specific estimates due to stratification errors. The training data can of course be correctly stratified using field observations, but in the application phase the stratification entirely relies on auxiliary information with complete coverage over the entire area of interest which cannot be corrected. We therefore tried to improve the stratification using canopy height information from airborne laser scanning to discriminate between young and mature stands. The results showed that this approach slightly improved the accuracy of the k-NN predictions, especially for the main tree species (2.6% for spruce volume). Furthermore, if metrics from aerial images were used to discriminate between pine and spruce dominance in the mature plots, the accuracy of volume of pine was improved by 73.2% in pine-dominated stands while for spruce an adverse effect of 12.6% was observed.     

Deciduous forest mapping using change detection of multi-temporal canopy height models from aerial images acquired at leaf-on and leaf-off conditions

Discrimination of deciduous trees using spectral information from aerial images has only been partly successfully due to the complexity of the reflectance at different view angles, times of acquisition, phenology of the trees and inter-tree radiance. Therefore, the objective was to evaluate the accuracy of estimating the proportion of deciduous stem volume (P) utilizing change detection between canopy height models (CHMs) generated by digital photogrammetry from leaf-on and leaf-off aerial images instead of using spectral information. The study was conducted at a hemi-boreal study area in Sweden. Using aerial images from three seasons, CHMs with a resolution of approximately 0.5?m were generated using semi-global matching. For training plots, metrics describing the change between leaf-on and leaf-off conditions were calculated and used to model the continuous variable P, using the Random Forest approach. Validated at sub-stands, the estimation accuracy of P in terms of root mean square error and bias was found to be 18% and ?6%, respectively. The overall classification accuracy, using four equally wide classes, was 83% with a kappa value of 0.68. The validation plots in classes of high proportion of coniferous or deciduous stem volume were well classified, whereas the mixed forest classes showed lower classification accuracies.     

Determination of number of stems in coniferous forest stands by means of aerial photo‐interpretation

Two models for determination of the number of stems per hectare in forest stands (N) from attributes derived by aerial photo‐interpretation were developed. The models relied on the assumption that N could be determined by dividing the total stand volume per hectare with the volume of the “average tree”; defined by stand mean height and the diameter corresponding to mean basal area of a stand. Input variables of the models were stand mean height, crown closure and site quality. Additionally, model II required input of average stand volume per hectare and average mean diameter derived from stratified field sample plot inventories. Material for 143 coniferous stands was used for the testing of the models. The stands were recorded by intensive field measurements. Aerial photographs at the approximate scale of 1:15 000 were used for photo‐interpretation. The N value was underestimated in model I by 5.4–47.0%. The standard deviation for the differences was 15.2–26.2% for mature stands and 41.4–44.2% for young thinning phase stands. For model II, the mean difference between the predicted and observed N value was in the range ‐16.1% to 12.2%.