Estimating forest characteristics in scanned aerial photographs with respect to requirements for economic forest management planning

The objective of forest management planning is often expressed as maximum sustainable economic yield. Methods used to collect information for forestry planning should, therefore, include variables significant for economic evaluations of management alternatives. It is important to be able to differentiate mature stands with respect to timber volumes and species mixture. In this study, digital high‐altitude aerial photographs are tested as a data source for planning. Circular plot data from a forest estate in northern Sweden were used as reference material. Global positioning system (GPS) measurements, with differential correction, were used to georeference the plots. Harvesting priorities were calculated for each plot using the Forest Management Planning Package. Volumes, species mixture and harvest priorities were estimated using regression analysis based on textural and spectral information from aerial photographs. The results show that the dependent variables could be estimated fairly well using only spectral information, e.g., R ² = 0.44 when estimating timber volume at reference plot (10 m radius) level. Aggregated to stand level, the precision was comparable with customary field survey methods (e.g., RMSE= 13.4% for timber volume).     

Spatial prediction of forest stand variables

This study aims at the development of a model to predict forest stand variables in management units (stands) from sample plot inventory data. For this purpose we apply a non-parametric most similar neighbour (MSN) approach. The study area is the municipal forest of Waldkirch, 13 km north-east of Freiburg, Germany, which comprises 328 forest stands and 834 sample plots. Low-resolution laser scanning data, classification variables as well rough estimations from the forest management planning serve as auxiliary variables. In order to avoid common problems of k-NN-approaches caused by asymmetry at the boundaries of the regression spaces and distorted distributions, forest stands are tessellated into subunits with an area approximately equivalent to an inventory sample plot. For each subunit only the one nearest neighbour is consulted. Predictions for target variables in stands are obtained by averaging the predictions for all subunits. After formulating a random parameter model with variance components, we calibrate the prior predictions by means of sample plot data within the forest stands via BLUPs (best linear unbiased predictors). Based on bootstrap simulations, prediction errors for most management units finally prove to be smaller than the design-based sampling error of the mean. The calibration approach shows superiority compared with pure non-parametric MSN predictions.     

Change detection using bi-temporal aerial photographs and registration at the stand level

There is a trend to continuously update forest data in forest management planning systems. Thus, changes in forest stands caused by, e.g. operations and storm damages should be detected in order to ensure the accuracy of forest data and beneficial decisions related to the treatments of the stands. This justifies the application of aerial photographs in change detection as being reasonable because they are already used in forest management planning. This study presents a semi-automatic method based on bi-temporal aerial photographs and registration at the stand and segment levels for the detection of changes in boreal forests. Linear stepwise discriminant analysis and the non-linear k-nearest neighbour (k-NN) method were tested and statistically compared in classification. The classification results at the stand level were found to be better than at the segment level. Compared to previous studies, the results of this study demonstrate remarkable improvement in the classification accuracy of moderate changes. The results showed that change detection substantially improved when the registration at the stand level was used, especially in the detection of thinned stands. To some extent, the method can be already applied operationally.     

The Ingrowth Bag Method in Measuring Root Production on Peatland Sites

The aim of this study was to develop a method for segment-based forest inventory and determine whether segment-level inventories can be used in forest management planning. The study area covered 76 ha located in two different aerial photographs in eastern Finland. The study area was segmented into 220 segments with the aid of aerial photographs and the segment-level forest characteristics were assessed in the field using relascope sample plots and a field computer which displayed the aerial photographs, segment borders and surveyor's location on the screen. The segment estimates were calculated as weighted averages of k nearest neighbours (kNN) for the segments and the sample plots. The estimates were tested with a cross-validation technique. The averages and the standard deviations of the spectral values of aerial images extracted for the segments and the sample plots were used in the kNN estimation. The relative root mean square error of the mean volume was 58.1% (bias –6.4%) at the segment level and 57.9% (bias –0.9%) at the sample plot level. The segment-based approach studied here needs further research and improvement before it can be applied to forest management planning.     

Predicting stand-thinning maturity from airborne laser scanning data

Abstract

Airborne laser scanning (ALS) has been used in recent years to acquire accurate remote-sensing material for carrying out practical forest inventories. Still, much of the information needed in forest management planning must be collected in the field. For example, forest management proposals are often determined in the field by an expert. In the present study, statistical features extracted from ALS data were used in logistic regression models and in nonparametric k-MSN estimation to predict the thinning maturity of stands. The research material consisted of 381 treewise measured circular plots in young and advanced thinning stands from the vicinity of Evo, in southern Finland. Timing of thinning was determined in the field by an expert and coded as a binary variable. Models were developed (1) to locate stands that will reach thinning maturity within the next 10-year period and (2) for stands in which thinning should be done immediately. For comparison purposes, logistic regression models were formulated from accurately field-measured stand characteristics. Logistic regression models based on ALS features predicted the thinning maturity with a classification accuracy of 79% (1) and 83% (2). The respective percentages were 66% and 83% with models based on field-measured stand characteristics and 70% and 86% with k-MSN. The study showed that ALS data can be used to predict stand-thinning maturity in a practical way.     

Estimation of stand volumes using the <Emphasis Type="Italic">k</Emphasis>-nearest neighbors method in Kyushu,Japan

The k-nearest neighbors (kNN) method is widely employed in national forest inventory applications using remote sensing data. The objective of this study was to evaluate the kNN method for stand volume estimation by combining LANDSAT/ETM+ data with 622 field sample plots from the Japanese National Forest Inventory (NFI) in Kyushu, Japan. The root mean square error (RMSE) and relative RMSE of the volume estimates rapidly decreased as the number of nearest neighbors (k) increased up to five, and then it slightly declined. They were consistently smaller for the Euclidean distance than for the Mahalanobis distance. The estimation errors (RMSE and relative RMSE) were 169.2 m³/ha and 66.2%, respectively (k = 10). The relative RMSE was similar to the previous studies. The estimated values were more accurate towards the mean value of the total volume, with an overestimation of the low volumes and an underestimation of the high volumes. We found a significant linear relationship between the observed stand volumes and estimated errors, which suggests that systematic errors may be reduced using this linearity. This research concluded that the kNN method is suitable for estimating stand volumes in Kyushu.     

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.     

Estimating forest data for analyses of forest production and utilization possibilities at local level by means of multi-source National Forest Inventory

The sample plot data of National Forest Inventories (NFI) are widely used in the analysis of forest production and utilization possibilities to support national and regional forest policy. However, there is an increasing interest for similar impact and scenario analyses for strategic planning at the local level. As the fairly sparse network of field plots only provides calculations for large areas, satellite image data have been applied to produce forest information for smaller areas. The aim of this study was to test the feasibility of generating forest data for a Finnish forest analysis tool, the MELA system, by means of the Landsat satellite imagery and the NFI sample plot data. The study was part of the preparation of a local forestry programme, where a strategic scenario analysis for the forest area of two villages (ca 8000 ha) was carried out. Management units that approximate forest stands were delineated by image segmentation. Stand volume and other parameters for each forest segment were estimated from weighted means of the NFI sample plots, where the individual sample plot weights were estimated by the k nearest neighbour (kNN) method. Two different spectral features were tested: single pixel values and average pixel values within a segment. The estimated forest data were compared with the forest data based on independent stand-level field assessments in two subareas, a national park and an area of forest managed for timber production.In the national park, the estimated mean volume of the growing stock from both spectral feature sets (about 160 m³ ha⁻¹) was clearly lower than that obtained from stand-level field assessment (186 m³ ha⁻¹). Using average pixel values within a segment resulted in a higher proportion of pine and a lower proportion of spruce volume than using single pixel values. It also resulted in an estimated felling potential nearly 10% higher over the first 10-year period in the scenario analysis of the area dedicated to timber production. However, the maximum long-term sustainable removal was at the same level (about 30,000 m³ year⁻¹) for both feature sets over the simulated 30-year period. The resulting annual felling area in the first 10-year period was 12% lower when the segment averages were applied, but the difference subsequently levelled off. The kNN approach in estimating initial forest data for scenario analyses at the local level was found promising.     

A comparison of different methods for forest resource estimation using information from airborne laser scanning and CIR orthophotos

This article compares three methods for forest resource estimation based on remote sensing features extracted from Airborne laser scanning and CIR orthophotos. The estimation was made exemplarily for the total stem volume of trees for a given area, measured in cubic metres per hectare [m3 ha⁻¹] (as one of the most important quantitative parameters to characterise a forest stand). The following methods were compared: Regression Analysis (RA), k-NN (nearest neighbour) method and a method that utilises regional yield tables, referred to as the yield table method (YT-method). The estimation of stem volume was examined in a mixed forest in Southern Germany using 300 circular inventory plots, each with a size of 452 m2. Remote sensing features relating to vegetation height and structures were extracted and used as input variables in the different approaches. The accuracy of the estimation was analysed using scatter plots and quantified using absolute and relative root mean square errors (RMSE). The comparison was made for all plots, as well as for averaged plot values located within forest stands that have the same age class. On “plot level” the RMSE yielded 79.79 m3 ha⁻¹ (RA), 81.93 m3 ha⁻¹ (k-NN) and 81.78 m3 ha⁻¹ (YT-method) and for the averaged values 35.75 m3 ha⁻¹ (RA), 35.06 m3 ha⁻¹ (k-NN) and 42.98 m3 ha⁻¹ (YT-method). Advantages and disadvantages, as well as requirements, of the methods are discussed.     

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%.     

The k-nearest neighbor technique with local linear regression

In a standard k-nearest neighbor (kNN) technique, imputations of unit-level values in the variables of interest (Y) are based on the k-nearest neighbors in a set of reference units. Nearest is defined with respect to a distance metric in the space of auxiliary variables (X). This study evaluates kNN imputations of Y with a selection, by the same distance metric, of k-nearest locally weighted regression models. Imputations are obtained as predictions using the X values of the k-nearest neighbors in the population. In simulated random sampling from three artificial multivariate populations and two actual univariate populations and sampling units composed of a single population element or a cluster of four elements, the new kNN technique: (1) improved the correlation between an imputation and its actual value; (2) lowered the root mean square error (RMSE) of imputations; (3) increased the slope in regressions of actual y values regressed against their imputed values; (4) performed relatively best with k values of 4 and sample sizes of 200 or greater; (5) compared favorably with a recently proposed kNN calibration procedure; and (6) had a higher (15–28%) RMSE than with a simple local linear regression. Distribution matching had a consistent negative effect (+10%) on RMSE.     

Two-compartment models for describing temporal variation in the effective leaf area index of Robinia pseudoacacia stands on the Loess Plateau

Leaf area index (LAI) is a key parameter in many ecological models and its phenology significantly affects on net ecosystem production in deciduous forests. We examined trends in LAIe (effective leaf area index) using two-compartment models to test effects of stand age and slope aspect on seasonal rates of LAIe increase (k₁) and decrease (k₂). LAIe measurements were acquired from 24 sample plots over five months and used to develop robust predictive models for modeling LAIe dynamics. Generally, the stands’ k₁ values were substantially greater than their k₂ values. The LAIe increased rapidly between leaf emergence and saturation, and then lower slowly from leaf saturation to senescence. The strongest increase in LAIe was observed in 11- to 15-year-old stands on shady slopes (k₁ = 4.76) and in young stands (<10 years) on sunny slopes (k₁ = 3.94), but k₂ values were largest in mature stands. The trees on shady slopes generally had higher k₁ and lower k₂ values than those on sunny slopes, except in the youngest stand. The results showed that two-compartment models can robustly describe leaf growth and loss rates in black locust stands, and that both stand age and aspect both strongly affect the rate and magnitude of changes in LAIe during the growing season. Hence, the two-compartment model is recommended as an effective method when monitoring the LAIe quantitative dynamics of broad-leaved forest.     

Logicality of expert evaluations of berry yields by the experience and interests of forest planners

Abstract

In Finland, expert knowledge has been widely utilized when developing models that facilitate predicting the impacts of alternative forest management options on non-wood forest products. However, expert modelling has been criticized because of the uncertainty and inconsistencies it includes. In this study, expert evaluations regarding bilberry (Vaccinium myrtillus L.) and cowberry (Vaccinium vitis-idaea L.) yields of different imaginary forest stands were analysed to find out whether the assessments were logically consistent. In particular, the consistency of the evaluations concerning berry production capacity of different stand densities and site fertilities was examined. The evaluations were made by 266 forest planners and other people whose work was related to field work in forest planning. The study also assessed whether the level of expertise (measured by two different variables, i.e. interest in berry picking and forest planning experience) affected the logic of the evaluations. It was found that on average both bilberry and cowberry yield assessments were in line with empirical research results found in literature. In addition, interest in berry picking was a more reasonable measure of berry yields in different forest stands than planning experience. The results of this study encourage the use of expert knowledge as a tool for forest planning and management.     

Estimation of premature forests in Georgia (USA) using U.S. Forest Service FIA data and Landsat imagery

We used geographic information system applications and statistical analyses to classify young, premature forest areas in southeastern Georgia using combined data from Landsat TM 5 satellite imagery and ground inventory data. We defined premature stands as forests with trees up to 15 years old. We estimated the premature forest areas using three methods: maximum likelihood classification (MLC), regression analysis, and k-nearest neighbor (kNN) modeling. Overall accuracy (OA) of classifying the premature forest using MLC was 82% and the Kappa coefficient of agreement was 0.63, which was the highest among the methods that we have tested. The kNN approach ranked second in accuracy with OA of 61% and a Kappa coefficient of agreement of 0.22. Regression analysis yielded an OA of 57% and a Kappa coefficient of 0.14. We conclude that Landsat imagery can be effectively used for estimating premature forest areas in combination with image processing classifiers such as MLC.     

Potentials and limits of the <Emphasis Type="Italic">k</Emphasis>-nearest-neighbour method for regionalising sample-based data in forestry

The k-nearest-neighbour (knn) method is known as a robust nonparametric method. It is used to estimate unknown values of data sets by means of similarity to reference data sets with known values. The spectral information of satellite remote sensing data can be used to provide the common characteristics in the knn estimation process. In forest sciences, the knn method is studied for its application potential. Some application examples are: (1) the estimation of parameters such as basal area, stem volume, number of trees per diameter class and tree species; (2) the estimation of forest debris and non-wood goods and services; (3) the production of wall-to-wall information for modelling, risk management and logistics. On the other hand, different limitations with respect to methodological characteristics as well as the selection of suitable parameters must be taken into consideration. The scope of this article concentrates on the discussion of the application potential and limits of the knn method in forestry with particular emphasis on management planning needs. The study is based on data taken from a forest inventory (FI) covering a test site near Rottenburg, in southwest Germany. Analysis results are compared with the traditional outcome of inventory data analysis and partly presented in thematic maps, which show identical spatial distribution patterns. For the map of six tree species, a map accuracy of 52.2% was found. The user’s accuracy for the prevailing tree species was between 52.6% for Picea abies and 69.4% for Quercus sp. A timber volume map for Quercus sp. clearly visualises the bias at the extreme ends of the volume distribution. The root mean square error (RMSE) for the total timber volume estimate was 30.9% for k = 5 and could be reduced to 22.6% for k = 20. For Quercus sp., however, the respective RMSE values were between 106.5 and 84.8%. Significant differences between FI and knn estimates were mainly found for rare classes with minor representation in the reference data.     

Developing an airborne laser scanning dominant height model from a countrywide scanning survey and national forest inventory data

Abstract

An airborne laser scanning (ALS) dominant height model was developed based on data from a national scanning survey with the aim of developing a digital terrain model (DTM) for Denmark. Data obtained in the ongoing Danish national forest inventory (NFI) were used as reference data. The data comprised a total of 2072 measurements of dominant height on NFI sample plots inventoried in 2006–2007 and their corresponding ALS data. The dominant height model included four variables derived from the ALS point cloud distribution. The variables were related to canopy height, canopy density and species composition on individual plots. The RMSE of the final model was 2.25 m and the model explained 93.9% of the variation (R ²). The model was successful in predicting dominant height across a wide range of forest tree species, stand heights, stand densities, canopy cover and growing conditions. The study demonstrated how low-density ALS data obtained in a survey not specifically aimed at forest applications may be used for obtaining biophysical forest properties such as dominant height, thereby reducing the overall forest inventory costs.     

Types of bark beetle (Coleoptera: Scolytidae) infestation in spruce forest stands affected by air pollution, bark beetle outbreak and honey fungus (Armillaria mellea)

583 spruce stands in an area affected by air pollution and bark beetle outbreak in Eastern Slovakia were studied in 1996. According to bark beetle infestation of dominant and codominant trees, stands were classified into following types of spruce stand decline:Ips typographus-A,Ips typographus-B,Polygraphus poligraphus, I. typographus/P. poligraphus—A,I. typographus/P. poligraphus—B. The presence of attacked trees in forest edges, bark beetle spots and forest interior was the key important factor for the classification. Data from forest inventory and forest management evidence together with data on types of spruce stands decline were used in further analyses. Results shows that the distribution of forest stands classified into different types or uninfested stands is related mainly to host size and site quality. The percentage of spruce, exposition of stands and stand density showed significant effects. The mechanisms of spreading of studied bark beetle outbreak could be explained by direct effects of stress of trees caused by an abrupt increase of level of solar irradiation and by weakening of trees by the honey fungus.     

Towards a DNA marker assisted seed source identification: a pilot study in European beech (<Emphasis Type="Italic">Fagus sylvatica</Emphasis> L.)

The widespread European forest tree Fagus sylvatica L. is of great importance for forest management. However, information about seed dispersal is still very rare, though important for harvesting strategies and later on seed source identification. We refined a DNA fingerprinting method for beech nut shells in order to directly assign dispersed seeds to their mother trees. A pilot study was conducted in two beech stands in Germany where leaves of the adult trees and the exocarp of dispersed seeds were fingerprinted at six nSSR loci. While one stand was randomly analysed for adults and dispersed seeds the other was systematically investigated following common harvesting procedures. Imitating the typical net harvesting strategy, seeds were collected beneath 19 adult trees. Exocarp genotyping revealed that on average three different mother trees contributed to a sample of five or six seeds collected beneath a single adult tree. Of the identified mother trees most were located within a radius of 15 m from the sampling point. The repeated pattern of seed dispersal within a short distance constitutes the basis for a straightforward strategy for the assignment of seed lots to a seed source stand. This strategy is based on the matching of individual genotypes without the necessity for a full inventory of the putative source stand. Additionally, we provide allelic ladders of five nSSR loci for standardization among laboratories.     

Spatial patterns and genetic structures within beech populations (<Emphasis Type="Italic">Fagus sylvatica</Emphasis> L.) of forked and non-forked individuals

Analyses of distribution patterns and genetic structures of forest stands can address distinct family structures and provide insights into the association of genetic and phenotypic variation patterns. In this study, point pattern analysis and spatial autocorrelation were used to examine the spatial and genetic structures in two naturally generated beech stands, which differ in age, trunk morphology, and stand management. Significant tree clumping was observed at distances up to 20 m in the young forest stand, whereas dispersion at distances under 10 m was observed in the old stand. The spatial analysis based on Ripley’s k function of the two different groups of trees showed that the non-forked trees match in both stands the spatial pattern of all trees while the forked were randomly distributed. Additionally, according to the bivariate analysis, forked trees in both stands were randomly distributed as related to non-forked tree positions. Finally, Moran’s I values were not very high, though significant genetic autocorrelation was identified at distances up to 20 m in the young stand, suggesting the existence of distinct family structures. However, no significant genetic structuring was observed in the old stand. Our findings suggest that spatial genetic patterns are impacted by stand age, environmental factors and human activities. The spatial distribution of forked trees was not clearly associated to family structures. Random effects and also micro-environmental variation could be additional factors explaining forking of beech individuals.