B4h: Novel remote sensing approaches to quantify carbon stocks, structure and functional diversity of forests - 1

Assessing Crown projection area (CPA) is crucial for estimating inter-tree competition and tree biomass volume. An approach to the fully automated assessment of CPA, using a multi-layer seeded region growing algorithm and high-resolution 3D point cloud data is presented. Data were collected via multi-scan TLS in a 4.08 ha mixed-species stand with high vertical structure, located in Lower-Austria. The new approach is tested against independently repeated manual CPA-measurements by different observers that serve as the reference. The inter-observer bias of the manual CPA-measurements is quantified, and allometric models are used to predict CPA from diameter at breast height (DBH) measurements. These predictions are then compared to the TLS-based estimates on the single tree- and stand-level. 

It is shown, that for single trees, there is a remarkable inter-observer bias of manual CPA measurements. The deviation between CPA measurements derived from TLS data and manual measurements is on par with the deviations between manual measurements by different observers. The inter-observer bias of the manual measurements propagates into the allometric models, resulting in a high uncertainty of the derived estimates at tree-level. At the stand level, TLS measurements reveal the high influence of crown morphology on the CPA, which only can be taken into account by the TLS measurements and not by the allometric models.

Fire-tolerant eucalypt forests of south-eastern Australia are assumed to fully recover from even the most intense fires but surprisingly very few studies have assessed that recovery. Accurate assessment of horizontal and vertical attributes of tree crowns after fire is essential to understand the fire’s legacy effects on tree growth and on forest structure. This study quantitatively assessed individual tree crowns nearly a decade after a 2009 wildfire that burnt extensive areas of temperate eucalypt forests. We used airborne lidar data validated with field measurements to estimate multiple individual tree crown metrics in 12 to 13 plots (0.05 ha) per each of four wildfire severities (unburnt, low, moderate, high). Linear mixed-effects models indicated persistent effects of both moderate- and high-severity wildfire on tree crown architecture. Trees at high-severity sites in two size classes (20 – 50 cm, >50 cm diameter) had significantly less crown projection area and live crown width than those at unburnt and low-severity sites. Significant differences in lidar-based metrics (crown cover, evenness, leaf area density profiles) indicated that tree crowns at moderate- and high-severity sites were comparatively narrow and more evenly distributed down the tree stem. These conical-shaped crowns contrasted sharply with the rounded crowns of trees at unburnt and low-severity sites. Our data provide clear evidence that moderate- and high-severity wildfire can modify the structure of fire-tolerant eucalypts’ tree for nearly a decade after the wildfire. The implications of these legacy effects on, for example, tree productivity and the accuracy of biomass allometric equations warrants further study.

In contrast to traditional inventories, modern terrestrial laser scanning (TLS) systems are available, which became likewise easy-to-handle in survey practice. However, before TLS systems can substitute traditional measurement instruments, computer algorithms have to be developed so that the novel laser-based technology is able to fulfill very basic tasks, prominently the automatic detection of tree positions and the measurement of tree diameters.

In this talk, a multi‐stage density-based clustering approach is presented for the automatic mapping of tree positions, and a subsequently applied algorithm is shown for the automatic measurement of tree diameters. The algorithms were tested in different settings with respect to the number and the spatial alignment of scanner positions under manifold forest conditions, covering different age classes and mixture scenarios and representing a broad gradient of structural complexity.

If circular sample plots were applied with a fixed truncation distance of 20 m, the tree mapping algorithm showed a detection rate of 89.3% with seven scanner positions (hexagon vertices plus center coordinates) and achieves a detection rate of 78.2% with four scanner positions aligned in a triangle plus center. Detection rates were significantly increased with smaller truncation distances. Thus, with a truncation distance of 10 m the hexagon setting yielded a detection rate of 92% and the triangle 90.4%. Other alignments of scanner positions were also tested, but proved to be either unfavorable or too labor intensive. The commission rates were on average less than 5%. The algorithms were successfully tested by means of an international benchmark dataset. 

Forest structural types (FSTs) assessment is important for a sustainable forest management. We used field and airborne laser scanning (ALS) data from Boreal, Mediterranean and Atlantic biogeographical regions and developed a simple methodology for FSTs identification. Four forest structural variables – quadratic mean diameter (QMD) , Gini coefficient (GC) , basal area larger than mean (BALM) and density of stems (N) – were calculated from forest inventory data. Hierarchal clustering analysis (HCA) was applied using these four structural variables and potential clusters (FSTs) were determined in the coniferous and deciduous forests. Then, the empirical threshold values for discriminating those clusters were extracted using classification and regression tree analysis (CART). Single storey FSTs, multi-layered FSTs and exponentially decreasing size distributions (reversed J) were separated by lower, medium and high values of GC and BALM, respectively, and these two were found to be the most important variables in the identification of FSTs. We also identified young/mature and sparse/dense subtypes within each of these main FST groups using QMD and N. Furthermore, similar structural predictors derived from ALS – maximum height (Max), L-coefficient of variation (Lcv), L-skewness (Lskew), and percentage of penetration (Cover), – were used in the nearest neighbour method to predict the FSTs. The overall accuracy achieved in the deciduous forest (0.87) was higher than the coniferous forest (0.72). Our methodology proves that ALS data is useful for structural heterogeneity assessment of forests and paves the way toward transnational assessments of forest structure across bioregions.

Forest resource inventories corresponded with models according to the purpose of survey equipment and management. With the advancement of science and technology, manual survey mode can be changed into a survey mode of high technology equipment. In this paper, the practical operation is used to explore the application of Handheld Mobile Laser Scanning (HMLS) and Unmanned Aerial Vehicle (UAV) in forest resource inventories. Further, we use Terrestrial Laser Scanning (TLS) as the valid data to compare the accuracy. With advantage of handheld mobile laser scanners (HMLS) and UAV, we discuss with estimating limitations and potentials of forest plot surveys. Using TLS data as ground true data, we combine the HMLS point cloud and UAV point cloud data to estimate the tree height of 43 trees. As One-way ANOVA results displayed, the tree height results between HMLS (11.8±2.1 m) and TLS (15.9±3.2 m) has significant differences (P<0.01), but after we combined the UAV point cloud and HMLS point cloud, the results shows no differences (P>0.05) with combined point cloud data (16.7±3.7 m) and TLS (15.9±3.2 m) point cloud data. For the TLS method, it has much more limitations for our investigation in Taiwan including time consuming and few conveniences. Combined point cloud data has more flexible procedure with high accuracy instead of time-consuming TLS method.

Developing countries that intend to implement the United Nations REDD-plus framework and obtain economic incentives are required to estimate changes in forest carbon stocks based on the IPCC guidelines. In this study, we developed a method to support REDD-plus implementation by estimating tropical forest aboveground biomass (AGB) by combining airborne LiDAR with high-spatial-resolution (HSR) satellite data, and by mapping forest carbon stocks and their changes at a sub-national level using time-series Landsat data. The results of AGB estimation from airborne LiDAR and HSR satellite data were used as training data for classification of forest carbon stock classes of Landsat data. We used a two-step method to estimate AGB and map it in a tropical environment in Cambodia. First, we created a multiple-regression model to estimate AGB from the LiDAR data and plotted field-surveyed AGB values against AGB values predicted by the LiDAR-based model, and calculated reflectance values in each band of the satellite data for the analyzed objects. Then, we created a multiple-regression model using AGB predicted by the LiDAR-based model as the dependent variable and the mean and standard deviation of the reflectance values in each band of the satellite data as the explanatory variables. Then, we conducted object-based classification of forest carbon stocks for time-series Landsat data using AGB derived from high-resolution satellite data as training data. Our results suggest that this approach can provide the forest carbon stock map required to support REDD-plus.

Climate change is causing novel stress to forests that is difficult to predict. Information on forest health has been identified as one of the key information gaps in evaluating the effects of climate change. The mapping of small-scale forest disturbance events is largely based on visual observations which is time-consuming and prone to error especially in the early stages of tree decline. Thus, new methods for objective estimation of tree decline are required. Multispectral lidar can provide highly detailed measurements of tree structure and reflectance simultaneously enabling novel approaches for the detection of tree stress. Over the last four years, we have developed a novel remote sensing method for detecting and evaluating tree decline using multispectral terrestrial lidar. The developed methods are based on utilizing lidar intensity data and the sensitivity of the used wavelengths to varying leaf water content (LWC). Thus, we have also investigated the relationship between LWC and various disturbance symptoms in different environments to determine the optimal LWC metrics for detecting tree stress. We found that multispectral lidar can detect and assess tree decline of single trees with high accuracy in our test forest that was infested by Ips typographus (L.) and its fungal symbionts. The developed methods showed potential in discriminating between healthy and stressed trees already in the early stages of tree decline when the foliage did not show visual symptoms. Therefore, the methods could provide new means for objective assessment of early tree decline allowing improved estimation, prediction and mitigation of forest damages.

Maps of stand-level forest disturbance history (Time Since Disturbance, TSD) are fundamental for modeling forest ecosystem processes. Recent research shows that airborne LiDAR data analysis can reconstruct the long-term (>100 years) disturbance history of a forest; this is a significantly longer time frame than what can be observed with the established methods of change detection with optical data. However, the use of airborne LiDAR is limited by its high cost compared to satellite data. These limitations could largely overcome by NASA’s spaceborne LiDAR Global Ecosystem Dynamics Investigations (GEDI) instrument, launched in December 2018. The sampling configuration of GEDI (footprints separated by 600m across-track and 60m along-track) imposes, however, some challenges for continuous spatial analysis compared to the common discrete-return airborne LiDAR. 

We propose an object-oriented data fusion methodology where Landsat data are used to identify forest stands, and GEDI waveforms are used to estimate TSD. The methodology involves three main steps: (1) forest stand delineation using object-based techniques of image segmentation on Landsat derived data; (2) estimation of TSD at the GEDI footprint level using random forest analysis; (3) estimation of TSD at the stand level, by combining TSD estimated at the GEDI footprint locations with the forest stand map. The methodology is demonstrated on a 52,000ha study area located in the Nez-Perce Clearwater National Forest (Idaho, USA), where reference maps of TSD are available for accuracy assessment. The results indicate that the GEDI-Landsat fusion has the potential of estimating TSD with accuracy comparable to airborne LiDAR data.