Mapping Coarse Woody Debris with Random Forest Classification of Centimetric Aerial Imagery

Abstract

Coarse woody debris (CWD; large parts of dead trees) is a vital element of forest ecosystems, playing an important role in nutrient cycling, carbon storage, fire fuel, microhabitats, and overall forest structure. However, there is a lack of effective tools for identifying and mapping both standing (snags) and downed (logs) CWD in complex natural settings. We applied a random forest machine learning classifier to detect CWD in centimetric aerial imagery acquired over a 270-hectare study area in the boreal forest of Alberta, Canada. We used a geographic object-based image analysis (GEOBIA) approach in the classification with spectral, spatial, and LiDAR (light detection and ranging)-derived height predictor variables. We found CWD to be detected with great accuracy (93.4 ± 4.2% completeness and 94.5 ± 3.2% correctness) when training samples were located within the application area, and with very good accuracy (84.2 ± 5.2% completeness and 92.2 ± 3.2% correctness) when training samples were located outside the application area. The addition of LiDAR-derived variables did not increase the accuracy of CWD detection overall (<2%), but aided significantly (p < 0.001) in the distinction between logs and snags. Foresters and researchers interested in CWD can take advantage of these novel methods to produce accurate maps of logs and snags, which will contribute to the understanding and management of forest ecosystems.

Estimating Coarse Woody Debris Volume Using Image Analysis and Multispectral LiDAR

“…We created a visible-CWD (vCWD) raster image over the study area by processing aerial images as well as LiDAR data via a GEOBIA workflow, involving machine-learning "random forest" supervised classification. This process was described in detail by Lopes Queiroz et al [12]. All spectral bands from the orthomosaic and the NDVI layer, generated with the 2017 mission data, were used to segment the study area into image-objects using eCognition [35].…”

“…The spatial, spectral and height attributes of these objects were used as predictor variables to train a RF classifier, which was applied to the whole study area in three processing chunks that were later merged together. More details of the reference dataset, as well as about the process used to test and apply the classifier can be found in Lopes Queiroz et al [12], who reported completeness between 76.9% and 81.5% and correctness between 75.8% and 87.0% for logs in the study area. Completeness is defined as the percentage of true positives relative to all positives in reference data and correctness is defined as a percentage of true positives relative to all positives in testing data.…”

“…Since the advent of high-speed, large-data computing, a variety of automated remote-sensing solutions have been implemented to identify and measure forest structure and components through the analysis of imagery collected with aircraft and satellites. Remote sensing of CWD has been achieved with variable degrees of success on satellite data [8,9] and high-resolution piloted and unpiloted aircraft imagery [10][11][12]. These studies commonly use pixel-based or geographical object-based image analysis (GEOBIA) solutions to process aerial images using linear models, classification trees, and machine learning to extract information regarding CWD objects.…”

“…If human interpreters could identify at most 90% of relatively large CWD during leaf-off, which represents good visibility conditions, one would expect automated detection to perform worse on more difficult conditions or smaller CWD. Some of the most successful recent studies of automated detection of CWD on forests, such as those of Duan et al [11], Stereńczak et al [14], and Lopes Queiroz et al [12], have developed sophisticated methods for the detection of visible CWD on aerial images, but have based their accuracy metrics of what is visible in aerial photos and not from field data, and therefore their efficacy based on ground truth remains unknown.…”

Archaeological documentation of wood caribou fences using unmanned aerial vehicle and very high-resolution satellite imagery in the Mackenzie Mountains, Northwest Territories

“…Challenges in CWD field sampling techniques (Ståhl et al 2001;Woldendorp et al 2004), as well as the relevance of dead wood in a host of forest ecosystem functions (Woodall et al 2019), have sparked an interest in applying remotely sensed data for this purpose. For example, CWD objects have recently been detected with encouraging results using LiDAR point clouds (Polewski et al 2015;Joyce et al 2019), aerial photography by means of Unmanned Aerial Vehicles (UAV) (Inoue et al 2014;Duan et al 2017;Panagiotidis et al 2019), or using LiDAR combined with multispectral aerial photography (Richardon and Moskal 2016;Stereńczak et al 2017;Lopes Queiroz et al 2019). Alternatively, models can be established leveraging spectral and structural relationships between plot-level observations of CWD volume and LiDAR or winter Landsat scenes (Sumnall et al 2016;Wolter et al 2017).…”