Airborne laser scanner (ALS) measurements from two test sites in Estonia were used to estimate forest canopy-base height (HVL). The ALS data was collected by Estonian Land Board using Leica ALS50-II scanner. The HVL was estimated by using mode value and standard deviation of the ALS pulse reflection position height distribution. The pulse reflections which had height less than 0.5 m over the estimated digital terrain model were excluded from the analysis. In situ measurements of canopy base height (HVA) were carried out in 20 mesotrophic Norway spruce and silver birch forest stands in Järvselja and in 45, mostly Scots pine dominant, mesotrophic forest stands in Aegviidu. Determination coefficients of linear regression between HVL and HVA for both test sites were over 0.8 and the residual standard errors of the models were less than two meters. The influence of forest understory vegetation to the estimation of HVL was tested by excluding the near-to-ground vegetation reflections which had height less than 1.5 m. The test results revealed no significant impact of forest understory to the HVL models. The cross validation showed that the HVL models were independent of test sites and tree species composition. The Järvselja data based HVL model had 1.3 m negative bias if applied to Aegviidu forests and the Aegviidu data based HVL model had 1.4 m positive bias if applied to Järvselja forests. In the Aegviidu test site, difference of HVL models of coniferous and deciduous stands was tested and the difference was found not to be significant
Airborne laser scanning (ALS) based standing wood volume models were analysed in two contrasting test sites with different forest types in Estonia. In Aegviidu test site main tree species are Scots pine and Norway spruce and Laeva test site is mainly dominated by deciduous species. ALS data measurements were carried out for Aegviidu in 2008 and for Laeva in 2013. Approximately 450 sample plots were established additionally to the forest inventory dataset in both test sites. Exclusive to the sample plots, 46 stands were measured in 2012 in Aegviidu for stand level model. The sample plot-based model standard error in Aegviidu was Se = 59.8 m3/ha (22%) and in Laeva Se = 69.2 m3/ha (29%). The stand-level model based on 46 measured stands from Aegviidu had Se = 38.4 m3/ha. Based on the models a cross-validation between the two test sites was carried out and systematic differences between the two test sites were found. The reasons are related to differences in optical properties of trees, crown shapes, flight configuration and canopy cover even though the sample plot based models included ALS-based canopy cover variable. The ALS-based wood volume estimate was also compared to forest inventory (FI) data and systematically larger estimates compared to FI dataset in both test sites were found. This average systematic error increased substantially (by 100 m3/ha) for stands with volume over 250 m3/ha. It was also detected that a model developed on small point clouds drawn for sample plots may produce systematic errors when applied to stand-level point clouds.
Thinning cuttings create moderate disturbances in forest stands. Thinning intensity indicates the amount of felled wood relative to the initial standing volume. We used sparse point clouds from airborne lidar measurements carried out in 2008 and 2012 at Aegviidu test site, Estonia, to study stand level relationships of thinning intensity to the changes in canopy cover and ALS-based wood volume estimates. Thinning intensity (Kr, HRV) was estimated from forest inventory data and harvester measurements of removed wood volume. The thinning intensity ranged from 17% to 56%. By raising threshold from 1.3 m to 8.0 m over ground surface we observed less canopy cover change, but stronger correlation with thinning intensity. Correlation between ALS-based and harvester-based thinning intensity was moderate. The ALS-based thinning intensity estimate was systematically smaller than Kr, HRV. Forest height growth compensates for a small decrease in canopy cover and intensity estimates for weak thinnings are not reliable using sparse point clouds and a four-year measurement interval.
Field measurements from 450 sample plots, airborne lidar data and spectral images from Aegviidu, Estonia, 15 by 15 km test site were used to analyse options to estimate main forest inventory variables using remote sensing data. Up to 7 m random error in location of 15 m radius sample plots within homogeneous stands causes usually about 0.5 m standard deviation in lidar pulse return height distribution percentiles. Forest mean height can be predicted with linear relationship from 80th percentile of lidar pulse return height distribution. Upper percentiles of pulse return height distribution are not significantly affected by omitting returns from ground and forest understorey vegetation. Total stem volume in forest can be predicted by using 80th percentile, 25th percentile and canopy cover as model arguments with less than 70 m3 ha-1 standard error. Best species specific stem volume models had 10 m3 ha-1 smaller standard error.
Canopy gap fraction has been estimated from hemispherical images using a thresholding method to separate sky and canopy pixels. The optimal objective thresholding rule has been searched by many authors without satisfactory results due to long list of reasons. Some recent studies have shown that unprocessed readings of camera CCD or CMOS sensor (raw data) have linear relationship with incident radiation. This allows a pair of cameras used in similar to a pair of plant canopy analyzers and canopy gap fraction can be calculated as the ratio of below canopy image and above canopy image. We tested new freeware program HemiSpherical Project Manager (HSP) for the restoration of the above canopy image from below canopy image which allows making field measurements with single below canopy operated camera. Results of perforated panel image analysis and comparison of plant area index (PAI) estimated independently by three operators from real canopy hemispherical images showed high degree of reliability of the new approach. Determination coefficients of linear regression of the PAI estimations of the three operators were 0.9962, 0.9875 and 0.9825. The canopy gap fraction data obtained from HSP were used to validate Nobis-Hunziker automatic thresholding algorithm. The results indicated that the Nobis-Hunziker algorithm underestimated PAI from out of camera JPEG images and overestimated PAI from raw data.
Forest height increment rate is related to the forest growth conditions. Data bases of previous forest inventories contain information about forest heightage relationship on large number of forest stands while repeated measurements of permanent sample plots provide an excellent reference for comparison. Repeated airborne laser scanning of forest stands is an additional source for the estimation of change in forest structure. In this study, height growth of middle-aged and older forest stands for about 10 year period was compared to an algebraic difference model on permanent sample plots (66) and for a sample of forest stands with repeated airborne laser scanning data (61). The model was based on a large dataset of forest inventory records from the period of 1984–1993. Statistically significant increased forest height growth was found in permanent sample plots based on tree height measurements (9 cm yr−1) as well in stands with repeated laser scanning data (4.5 cm yr−1) in South-East Estonia compared to the algebraic difference model. The difference between the two data sets was explained by their mean age and site class, but the increased forest height growth compared to the old forest inventory data indicates improved growth conditions of forests in the test area. The results hint also that empirical data-based forest growth models need to be updated to avoid biased growth estimates.
The aim of this study was to compile the leaf area index (LAI) map of a 3×3 km VALERI test site in Järvselja, Estonia. Canopy transmittance measurements of LAI by LAI-2000 Plant Canopy Analyzers and digital cameras supplied with fisheye converters were carried out on 42 elementary sampling units at ground level and at breast height level. The vegetation LAI was estimated as a sum of the tree canopy true green LAI obtained from the inversion of canopy gap fraction data and of the ground vegetation effective LAI. Red channel from SPOT-4 HRV-IR image and airborne lidar data based canopy transmittance were used for up-scaling. The LAI map was compared with standard LAI products from Terra MODIS and ENVISAT MERIS. We found that lidar data based LAI estimate on up-scaled map saturated at high values (LAI > 4.5) compared to the LAI estimates based on SPOT-4 HRV-IR red channel. Validation of MODIS LAI product revealed substantial underestimates of LAI compared to the up-scaled field measurements and rather large random noise. ENVISAT MERIS LAI product was more similar to up-scaled field measurements; however, rather large unexpected random variations exist in its time series.
Nearest neighbour techniques are useful for constructing maps of forest inventory variables based on sample plot and auxiliary data from remote sensing. The most problematic issue of the nearest neighbour technique is possible systematic bias in the estimated values. In this study a 15 by 15 km test site in nemoral multilayer mixed forests was established in Laeva, Estonia. A set of 444 circular sample plots was used as reference set. Airborne lidar data and Landsat-8 Operational Land Imager image were used to construct five different feature variable sets consisting of original variables and alternatively principal components. The response variables were wood volume of first tree layer, wood volume of the second tree layer and main species code. A special test was carried out where a substantial amount of Silver birch dominated plots were removed from the reference set. The wood volume prediction validation was carried out on 89 forest growth sample plots and on 2290 forest stands. Species composition prediction was validated on 986 forest stands. As in many previous studies the results confirmed superiority of airborne lidar variables over spectral variables for wood volume estimation. The first three principal components of airborne lidar variables and first five principal components of all possible original feature variables contained over 99% of the information and performed well in imputations. The imputed wood volume at small values was overestimated and underestimated at large values regardless of used reference set. The feature variable sets containing spectral data performed better for species composition imputation. There was a forest age dependent discrepancy in predicted species proportions: birch and spruce proportions were underestimated in young stands and overestimated in older stands while proportion of aspen had exactly the opposite errors. The lack of fit depended slightly on the feature variable sets. The birch dominated plot partial removal from the reference set changed the predicted proportion of species but did not remove the forest age dependent lack of fit. The result can be important for the studies in which bootstrap samples are used to estimate error statistics for nearest neighbour technique based forest inventory variable maps.