Current issue: 55(5)
Under compilation: 56(1)
Current remote sensing methods can provide detailed tree species classification in boreal forests. However, classification studies have so far focused on the dominant tree species, with few studies on less frequent but ecologically important species. We aimed to separate European aspen (Populus tremula L.), a biodiversity-supporting tree species, from the more common species in European boreal forests (Pinus sylvestris L., Picea abies [L.] Karst., Betula spp.). Using multispectral drone images collected on five dates throughout one thermal growing season (May–September), we tested the optimal season for the acquisition of mono-temporal data. These images were collected from a mature, unmanaged forest. After conversion into photogrammetric point clouds, we segmented crowns manually and automatically and classified the species by linear discriminant analysis. The highest overall classification accuracy (95%) for the four species as well as the highest classification accuracy for aspen specifically (user’s accuracy of 97% and a producer’s accuracy of 96%) were obtained at the beginning of the thermal growing season (13 May) by manual segmentation. On 13 May, aspen had no leaves yet, unlike birches. In contrast, the lowest classification accuracy was achieved on 27 September during the autumn senescence period. This is potentially caused by high intraspecific variation in aspen autumn coloration but may also be related to our date of acquisition. Our findings indicate that multispectral drone images collected in spring can be used to locate and classify less frequent tree species highly accurately. The temporal variation in leaf and canopy appearance can alter the detection accuracy considerably.
Photogrammetric point clouds obtained with unmanned aircraft systems (UAS) have emerged as an alternative source of remotely sensed data for small area forest management inventories (FMI). Nonetheless, it is often overlooked that small area FMI require considerable field data in addition to UAS data, to support the modelling of forest attributes. In this study, we propose a method whereby tree volumes by species are predicted with photogrammetric UAS data and Sentinel-2 images, using models fitted with airborne laser scanning data. The study area is in a managed boreal forest area in Eastern Finland. First, we predicted total volume with UAS point cloud metrics using a prior regression model fitted in another area with ALS data. Tree species proportions were then predicted by k nearest neighbor (k-NN) imputation based on bi-seasonal Sentinel-2 images without measuring new field plot data. Species-specific volumes were then obtained by multiplying the total volume by species proportions. The relative root mean square error (RMSE) values for total and species-specific volume predictions at the validation plot level (30 m × 30 m) were 9.0%, and 33.4–62.6%, respectively. Our approach appears promising for species-specific small area FMI in Finland and in comparable forest conditions in which suitable field plots are available.
Airborne laser scanning (ALS) data is nowadays often available for forest inventory purposes, but adequate field data for constructing new forest attribute models for each area may be lacking. Thus there is a need to study the transferability of existing ALS-based models among different inventory areas. The objective of our study was to apply ALS-based mixed models to estimate the diameter, height and crown base height of individual sawlog sized Scots pines (Pinus sylvestris L.) at three different inventory sites in eastern Finland. Different ALS sensors and acquisition parameters were used at each site. Multivariate mixed-effects models were fitted at one site and the models were validated at two independent test sites. Validation was carried out by applying the fixed parts of the mixed models as such, and by calibrating them using 1–3 sample trees per plot. The results showed that the relative RMSEs of the predictions were 1.2–6.5 percent points larger at the test sites compared to the training site. Systematic errors of 2.4–6.2 percent points also emerged at the test sites. However, both the RMSEs and the systematic errors decreased with calibration. The results showed that mixed-effects models of individual tree attributes can be successfully transferred and calibrated to other ALS inventory areas in a level of accuracy that appears suitable for practical applications.
The aim of this study was to examine how well stem volume, above-ground biomass and dominant height can be predicted using nationwide airborne laser scanning (ALS) based regression models. The study material consisted of nine practical ALS inventory projects taken from different parts of Finland. We used field sample plots and airborne laser scanning data to create nationwide and regional models for each response variable. The final models had one or two ALS predictors, which were chosen based on the root mean square error (RMSE), and cross-validated. Finally, we tested how much predictions would improve if the nationwide models were calibrated with a small number of regional sample plots. Although forest structures differ among different parts of Finland, the nationwide volume and biomass models performed quite well (leave-inventory-area-out RMSE 22.3% to 33.8%, mean difference [MD] –13.8% to 18.7%) compared with regional models (leave-plot-out RMSE 20.2% to 26.8%). However, the nationwide dominant height model (RMSE 5.4% to 7.7%, MD –2.0% to 2.8%, with the exception of the Tornio region – RMSE 11.4%, MD –9.1%) performed nearly as well as the regional models (RMSE 5.2% to 6.7%). The results show that the nationwide volume and biomass models provided different means than real means at regional level, because forest structure and ALS device have a considerable effect on the predictions. Large MDs appeared especially in northern Finland. Local calibration decreased the MD and RMSE of volume and biomass models. However, the nationwide dominant height model did not benefit much from calibration.
The fusion of optical satellite imagery, strips of lidar data and field plots is a promising approach for the inventory of tropical forests. Airborne lidars also enable an accurate direct estimation of the forest canopy cover (CC), and thus a sample of lidar strips can be used as reference data for creating CC maps which are based on satellite images. In this study, our objective was to validate CC maps obtained from an ALOS AVNIR-2 satellite image wall-to-wall, against a lidar-based CC map of a tropical forest area located in Laos. The reference CC values which were needed for model training were obtained from a sample of four lidar strips. Zero-and-one inflated beta regression (ZOINBR) models were applied to link the spectral vegetation indices derived from the ALOS image with the lidar-based CC estimates. In addition, we compared ZOINBR and logistic regression models in the forest area estimation by using >20% CC as a forest definition. Using a total of 409 217 30 × 30 m population units as validation, our model showed a strong correlation between lidar-based CC and spectral satellite features (root mean square error = 12.8%, R2 = 0.82). In the forest area estimation, a direct classification using logistic regression provided better accuracy than the estimation of CC values as an intermediate step (kappa = 0.61 vs. 0.53). It is important to obtain sufficient training data from both ends of the CC range. The forest area estimation should be done before the CC estimation, rather than vice versa.