We contribute to a better understanding of different remote sensing techniques for tree height estimation by comparing several techniques to both direct and indirect field measurements. From these comparisons, factors influencing the accuracy of reliable tree height measurements were identified. Different remote sensing methods were applied on the same test site, varying the factors sensor type, platform, and flight parameters. We implemented light detection and ranging (LiDAR) and photogrammetric aerial images received from unmanned aerial vehicles (UAV), gyrocopter, and aircraft. Field measurements were carried out indirectly using a Vertex clinometer and directly after felling using a tape measure on tree trunks. Indirect measurements resulted in an RMSE of 1.02 m and tend to underestimate tree height with a systematic error of −0.66 m. For the derivation of tree height, the results varied from an RMSE of 0.36 m for UAV-LiDAR data to 2.89 m for photogrammetric data acquired by an aircraft. Measurements derived from LiDAR data resulted in higher tree heights, while measurements from photogrammetric data tended to be lower than field measurements. When absolute orientation was appropriate, measurements from UAV-Camera were as reliable as those from UAV-LiDAR. With low flight altitudes, small camera lens angles, and an accurate orientation, higher accuracies for the estimation of individual tree heights could be achieved. The study showed that remote sensing measurements of tree height can be more accurate than traditional triangulation techniques if the aforementioned conditions are fulfilled.

Forest gaps are important structural elements in forest ecology to which various conservation-relevant, photophilic species are associated. To automatically map forest gaps and detect their changes over time, we developed a method based on Digital Surface Models (DSM) derived from stereoscopic aerial imagery and a LiDAR-based Digital Elevation Model (LiDAR DEM). Gaps were detected and delineated in relation to height and cover of the surrounding forest comparing data from two public flight campaigns (2009 and 2012) in a 1023-ha model region in the Northern Black Forest, Southwest Germany. The method was evaluated using an independent validation dataset obtained by visual stereo-interpretation. Gaps were automatically detected with an overall accuracy of 0.90 (2009) and 0.82 (2012). However, a very high users’ accuracy of more than 0.95 (both years) was counterbalanced by a producer’s accuracy of 0.84 (2009) and 0.73 (2012) as some gaps were not automatically detected. Accuracy was mainly dependent on the shadow occurrence and height of the surrounding forest with user’s accuracies dropping to 0.70 (2009) and 0.52 (2012) in high stands (>8 m tree height). As one important step in the workflow, the class of open forest, an important feature for many forest species, was delineated with a very good overall accuracy of 0.92 (both years) with uncertainties occurring mostly in areas with intermediate canopy cover. Presence of complete or partial shadow and geometric limitations of stereo image matching were identified as the main sources of errors in the method performance, suggesting that images with a higher overlap and resolution and ameliorated image-matching algorithms provide the greatest potential for improvement.

Research Highlights: This study developed the first remote sensing-based forest cover map of Baden-Württemberg, Germany, in a very high level of detail. Background and Objectives: As available global or pan-European forest maps have a low level of detail and the forest definition is not considered, administrative data are often oversimplified or out of date. Consequently, there is an important need for spatio-temporally explicit forest maps. The main objective of the present study was to generate a forest cover map of Baden-Württemberg, taking the German forest definition into account. Furthermore, we compared the results to NFI data; incongruences were categorized and quantified. Materials and Methods: We used a multisensory approach involving both aerial images and Sentinel-2 data. The applied methods are almost completely automated and therefore suitable for area-wide forest mapping. Results: According to our results, approximately 37.12% of the state is covered by forest, which agrees very well with the results of the NFI report (37.26% ± 0.44%). We showed that the forest cover map could be derived by aerial images and Sentinel-2 data including various data acquisition conditions and settings. Comparisons between the forest cover map and 34,429 NFI plots resulted in a spatial agreement of 95.21% overall. We identified four reasons for incongruences: (a) edge effects at forest borders (2.08%), (b) different forest definitions since NFI does not specify minimum tree height (2.04%), (c) land cover does not match land use (0.66%) and (d) errors in the forest cover layer (0.01%). Conclusions: The introduced approach is a valuable technique for mapping forest cover in a high level of detail. The developed forest cover map is frequently updated and thus can be used for monitoring purposes and for assisting a wide range of forest science, biodiversity or climate change-related studies.

Deadwood mapping is of high relevance for studies on forest biodiversity, forest disturbance, and dynamics. As deadwood predominantly occurs in forests characterized by a high structural complexity and rugged terrain, the use of remote sensing offers numerous advantages over terrestrial inventory. However, deadwood misclassifications can occur in the presence of bare ground, displaying a similar spectral signature. In this study, we tested the potential to detect standing deadwood (h > 5 m) using orthophotos (0.5 m resolution) and digital surface models (DSM) (1 m resolution), both derived from stereo aerial image matching (0.2 m resolution and 60%/30% overlap (end/side lap)). Models were calibrated in a 600 ha mountain forest area that was rich in deadwood in various stages of decay. We employed random forest (RF) classification, followed by two approaches for addressing the deadwood-bare ground misclassification issue: (1) post-processing, with a mean neighborhood filter for “deadwood”-pixels and filtering out isolated pixels and (2) a “deadwood-uncertainty” filter, quantifying the probability of a “deadwood”-pixel to be correctly classified as a function of the environmental and spectral conditions in its neighborhood. RF model validation based on data partitioning delivered high user’s (UA) and producer’s (PA) accuracies (both > 0.9). Independent validation, however, revealed a high commission error for deadwood, mainly in areas with bare ground (UA = 0.60, PA = 0.87). Post-processing (1) and the application of the uncertainty filter (2) improved the distinction between deadwood and bare ground and led to a more balanced relation between UA and PA (UA of 0.69 and 0.74, PA of 0.79 and 0.80, under (1) and (2), respectively). Deadwood-pixels showed 90% location agreement with manually delineated reference to deadwood objects. With both alternative solutions, deadwood mapping achieved reliable results and the highest accuracies were obtained with deadwood-uncertainty filter. Since the information on surface heights was crucial for correct classification, enhancing DSM quality could substantially improve the results.

Information pertaining to forest timber volume is crucial for sustainable forest management. Remotely-sensed data have been incorporated into operational forest inventories to serve the need for ever more diverse and detailed forest statistics and to produce spatially explicit data products. In this study, data derived from airborne laser scanning and image-based point clouds were compared using three volume estimation methods to aid wall-to-wall mapping of forest timber volume. Estimates of forest height and tree density metrics derived from remotely-sensed data are used as explanatory variables, and forest timber volumes based on sample field plots are used as response variables. When compared to data derived from image-based point clouds, airborne laser scanning produced slightly more accurate estimates of timber volume, with a root mean square error (RMSE) of 26.3% using multiple linear regression. In comparison, RMSEs for volume estimates derived from image-based point clouds were 28.3% and 29.0%, respectively, using Semi-Global Matching and enhanced Automatic Terrain Extraction methods. Multiple linear regression was the best-performing parameter estimation method when compared to k-Nearest Neighbour and Support Vector Machine. In many countries, aerial imagery is acquired and updated on regular cycles of 1–5 years when compared to more costly, once-off airborne laser scanning surveys. This study demonstrates point clouds generated from such aerial imagery can be used to enhance the estimation of forest parameters at a stand and forest compartment level-scale using small area estimation methods while at the same time achieving sampling error reduction and improving accuracy at the forest enterprise-level scale.

Multi-temporal Sentinel 2 optical images and 3D photogrammetric point clouds can be combined to enhance the accuracy of timber volume models on large spatial scale. Information on the proportion of broadleaf and conifer trees improves timber volume models obtained from 3D photogrammetric point clouds. However, the broadleaf-conifer information cannot be obtained from photogrammetric point clouds alone. Furthermore, spectral information of aerial images is too inconsistent to be used for automatic broadleaf-conifer classification over larger areas. In this study we combined multi-temporal Sentinel 2 optical satellite images, 3D photogrammetric point clouds from digital aerial stereo photographs, and forest inventory plots representing an area of 35,751 km² in south-west Germany for (1) modelling the percentage of broadleaf tree volume (BL%) using Sentinel 2 time series and (2) modelling timber volume per hectare using 3D photogrammetric point clouds. Forest inventory plots were surveyed in the same years and regions as stereo photographs were acquired (2013–2017), resulting in 11,554 plots. Sentinel 2 images from 2016 and 2017 were corrected for topographic and atmospheric influences and combined with the same forest inventory plots. Spectral variables from corrected multi-temporal Sentinel 2 images were calculated, and Support Vector Machine (SVM) regressions were fitted for each Sentinel 2 scene estimating the BL% for corresponding inventory plots. Variables from the photogrammetric point clouds were calculated for each inventory plot and a non-linear regression model predicting timber volume per hectare was fitted. Each SVM regression and the timber volume model were evaluated using ten-fold cross-validation (CV). The SVM regression models estimating the BL% per Sentinel 2 scene achieved overall accuracies of 68%–75% and a Root Mean Squared Error (RMSE) of 21.5–26.1. The timber volume model showed a RMSE% of 31.7%, a mean bias of 0.2%, and a pseudo-R² of 0.64. Application of the SVM regressions on Sentinel 2 scenes covering the state of Baden-Württemberg resulted in predictions of broadleaf tree percentages for the entire state. These predicted values were used as additional predictor in the timber volume model, allowing for predictions of timber volume for the same area. Spatially high-resolution information about growing stock is of great practical relevance for forest management planning, especially when the timber volume of a smaller unit is of interest, for example of a forest stand or a forest district where not enough terrestrial inventory plots are available to make reliable estimations. Here, predictions from remote-sensing based models can be used. Furthermore, information about broadleaf and conifer trees improves timber volume models and reduces model errors and, thereby, prediction uncertainties.