Ilkka Korpela

Simulated Multispectral Imagery for Tree Species Classification Using Support Vector Machines

The information content of remotely sensed data depends primarily on the spatial and spectral properties of the imaging device. This paper focuses on the classification performance of the different spectral features (hyper- and multispectral measurements) with respect to three tree species. The Support Vector Machine was chosen as the classification algorithm for these features. A simulated optical radiation model was constructed to evaluate the identification performance of the given multispectral system for the tree species, and the effects of spectral-band selection and data preprocessing were studied in this setting. Simulations were based on the reflectance measurements of the pine (Pinus sylvestris L.), spruce [ Picea abies (L.) H. Karst.], and birch trees (Betula pubescens Ehrh. and Betula pendula Roth). Leica ADS80 airborne sensor with four spectral bands (channels) was used as a fixed multispectral sensor system that leads to response values for the at-sensor radiance signal. Results suggest that this four-band system has inadequate classification performance for the three tree species. The simulations demonstrate on average a 5-15 percentage points improvement in classification performance when the Leica system is combined with one additional spectral band. It is also demonstrated for the Leica data that feature mapping through a Mahalanobis kernel leads to a 5-10 percentage points improvement in classification performance when compared with other kernels.

Mapping of snow-damaged trees based on bitemporal airborne LiDAR data

The use of multitemporal LiDAR data in forest-monitoring applications has been so far largely unexplored. In this work, we aimed to develop and test a simple method for the detection of snow-induced canopy changes by employing bitemporal LiDAR data acquired in 2006–2010. Our study area was located in southern Finland (62°N, 24°E), where snow-induced damage occurred in 10 permanent Scots pine (Pinus sylvestris)-dominated plots in winter 2009–2010. For the detection of snow-damaged crowns, we developed a ∆CHM method by contrasting bitemporal LiDAR canopy height models (CHMs) and analyzing the resulting difference image, using binary image operations to extract the damaged crowns. Furthermore, we examined the structural and spatial factors that could explain snow damage at the individual tree level. The ∆CHM method developed is based on two threshold parameters, i.e., the required height difference in the contrasted CHMs and the minimum plausible area of damage. When testing the performance of ∆CHM method, we found that the plot-level omission error rates were 19–75%, while the commission error rates were 0–21%. Furthermore, the relative estimation accuracy of the damaged crown projection area (DCPA) ranged from −16.4 to 5.4%. The observed damage could be explained at tree level by stem tapering, relative tree size, and local stand density. To conclude, ∆CHM method developed constitutes a potential tool for the monitoring of structural canopy changes in the dominant tree layer if dense multitemporal LiDAR data are available.

Relasphone—Mobile and Participative In Situ Forest Biomass Measurements Supporting Satellite Image Mapping

Due to the high cost of traditional forest plot measurements, the availability of up-to-date in situ forest inventory data has been a bottleneck for remote sensing image analysis in support of the important global forest biomass mapping. Capitalizing on the proliferation of smartphones, citizen science is a promising approach to increase spatial and temporal coverages of in situ forest observations in a cost-effective way. Digital cameras can be used as a relascope device to measure basal area, a forest density variable that is closely related to biomass. In this paper, we present the Relasphone mobile application with extensive accuracy assessment in two mixed forest sites from different biomes. Basal area measurements in Finland (boreal zone) were in good agreement with reference forest inventory plot data on pine ( R 2 = 0 . 75 , R M S E = 5 . 33 m 2 /ha), spruce ( R 2 = 0 . 75 , R M S E = 6 . 73 m 2 /ha) and birch ( R 2 = 0 . 71 , R M S E = 4 . 98 m 2 /ha), with total relative R M S E ( % ) = 29 . 66 % . In Durango, Mexico (temperate zone), Relasphone stem volume measurements were best for pine ( R 2 = 0 . 88 , R M S E = 32 . 46 m 3 /ha) and total stem volume ( R 2 = 0 . 87 , R M S E = 35 . 21 m 3 /ha). Relasphone data were then successfully utilized as the only reference data in combination with optical satellite images to produce biomass maps. The Relasphone concept has been validated for future use by citizens in other locations.

Modelling lidar-derived boreal forest canopy cover with SPOT 4 HRVIR data

Forest canopy cover (C) is needed in forest area monitoring and for many ecological models. Airborne scanning lidar sensors can produce fairly accurate C estimates even without field training data. However, optical satellite images are more cost-efficient for large area inventories. Our objective was to use airborne lidar data to obtain accurate estimates of C for a set of sample plots in a boreal forest and to generalize C for a large area using a satellite image. The normalized difference vegetation index (NDVI) and reduced simple ratio (RSR) were calculated from the satellite image and used as predictors in the regressions. RSR, which combines information from the red, near-infrared, and shortwave infrared bands, provided the best performance in terms of absolute root mean square error (RMSE) (7.3%) in the training data. NDVI produced a markedly larger RMSE (10.0%). However, in an independent validation data set, RMSE increased (13.0–17.1%) because the systematic sample of validation plots contained more variation than the training plots. Our results are better than those reported earlier, which is probably explained by more consistent C estimates derived from the lidar. Our approach provides an efficient method for creating C maps for large areas.

Assessment of Radiometric Correction Methods for ADS40 Imagery

This article presents the results of an assessment of radiometric correction methods of images taken by the large-format aerial, photogrammetric, multispectral pushbroom camera Leica Geosystems ADS40. The investigation was carried out in the context of the multi-site EuroSDR project “Radiometric aspects of digital photogrammetric images”. Images were collected at the forestry research test site Hyytiala, Finland in August, 2008. Two processing workflows were evaluated: one based on the photogrammetric software Leica XPro, which in radiometric processes relies on physical modeling and information collected from the imagery only, and one based on ATCOR-4, which is software dedicated to physical atmospheric correction of airborne multi-, hyperspectral and thermal scanner data, and can be operated either with or without in-situ reflectance and atmospheric observations. Outputs of these processes are reflectance images. Three participants processed the data with several processing options which resulted in a total of 12 different radiometrically corrected reflectance images. The data analysis was based on field and laboratory reflectance measurements of reference reflectance targets and field measurements of permanent targets (asphalt, grass, gravel). Leica XPro provided up to 5 % reflectance accuracy without any ground reference and ATCOR-4 provided reflectance accuracy better than 5 % with vicarious inflight radiometric calibration of the sensor. The results show that the radiometric correction of multispectral aerial images is possible in an efficient way in the photogrammetric production environment.

The performance of a local maxima method for detecting individual tree tops in aerial photographs

Tree locations are needed in image‐based single tree forest inventories. Accurate tree‐top image positions would also be useful in image matching (IM) for the estimation of a canopy surface model. We explored the performance of a local maxima (LM)‐based method for detecting image positions of individual tree tops by using digitized aerial photographs in recently thinned stands in Southern Finland. The accuracy of tree detection at the single tree level was assessed using a novel 3D approach. The results indicate that the LM method works most reliably in the central parts of the aerial images. Small trees are mostly missed by the LM detector and commission errors seem unavoidable. We propose further work that would assess the applicability of the LM method as a feature detector for use in IM.

Estimation of tree crown volume from airborne lidar data using computational geometry

Although simple geometrical shapes are commonly used to describe tree crowns, computational geometry enables calculation of the individual crown properties directly from airborne lidar point clouds. Our objective was to calculate crown volumes (CVs) using this technique and validate the results by comparing them with field-measured values and modelled ellipsoidal crowns. The CVs of standing trees were obtained by measuring the crown radii at different heights, integrating the obtained crown profiles as solids of revolution, and finally averaging the volumes obtained from the four separate profiles. With the lidar data, the CVs were extracted using 3D alpha shape and 3D convex hull techniques. Crown base heights (CBHs) were also estimated from the lidar data and used to exclude echoes from the understory, which was also done using field-based CBHs to exclude this error source. The results show that the field-measured CVs had a high correlation with lidar-based estimates (best R 2 = 0.83), but the lidar-based estimates were generally smaller than the field values. The best correspondence (root mean square difference (RMSD) = 45.0%, average difference = –24.7%) was obtained using the convex hull of the point data and field-measured CBH. The CBHs were consistently overestimated (RMSD = 37.3%; average difference = –20.0%), especially in spruces with long crowns. Thus using lidar-based CBH also increased the inaccuracy of the CV estimates. While the underestimation of CV is mainly explained by the inadequate number of echoes from the lower regions of the crowns, the CVs obtained from the lidar were better than those obtained with ellipsoids fitted by using general models for crown dimensions. The utility of the estimated CVs in the prediction of stem diameter is also demonstrated.

Logistic Regression-Based Spectral Band Selection for Tree Species Classification: Effects of Spatial Scale and Balance in Training Samples

In this letter, we evaluated the pixel-level and plot-level tree species classification of Scots Pine, Norway Spruce, and deciduous birch in a boreal forest using 64-band AisaEAGLE II hyperspectral data in a wavelength range of 400-1000 nm. First, band selection was performed using a sparse logistic regression-based feature selection algorithm with pixel-level and plot-level data in case of balanced and imbalanced training data. This resulted in 8-11 selected hyperspectral bands, depending on the properties of the data used. We evaluated a tree species classification with 8-11 selected hyperspectral bands directly for a least squares support vector machine (LS-SVM)-based pixel-level classification with a relatively small training set size (0.5%-1.5% of the total data) and obtained an accuracy and kappa of around 93.50% and 0.90, respectively. These results are around 0.53%-0.94% points lower than those obtained using all of the hyperspectral bands. Second, one important wavelength region highlight by the selected bands was used to modify the sensor sensitivity configuration in the Leica Airborne Digital Sensor 40 (ADS40) multispectral sensor. Using a simulation model and the hyperspectral data, the modified and standard Leica ADS40 sensor responses were simulated and compared, and the modified system simulated response indicates a 3%-5% point improvement in the pixel-level and plot-level LS-SVM classification accuracy compared with the simulated responses of the standard Leica ADS40 band configuration.

Laserkeilainperustainen puulajiluokitus - puu- ja metsikkötekijöiden, opetusaineiston koon, laserintesiteetin normalisoinnin sekä keilainmallin vaikutus

Seloste artikkelista: Korpela, I., Orka, H. O., Maltamo, M., Tokola, T. & Hyyppa, J. / Tree species classification using airborne LiDAR - effects of stand and tree parameters, downsizing of training set, intensity normalization, and sensor type. Silva Fennica 44 (2010) : 2, 319-339.

Inventering av plantskog med laserscanning och digitalt flygfoto

Referat av artikeln: Korpela, I., Tuomola, T., Tokola, T. & Dahlin, B. 2008. Appraisal of seedling stand vegetation with airborne imagery and discrete-return LiDAR - an exploratory analysis. Silva Fennica 42 (5) : 753-772.