Tree Species Classification and Health Status Assessment for a Mixed Broadleaf-Conifer Forest with UAS Multispectral Imaging

Abstract

Automatic discrimination of tree species and identification of physiological stress imposed on forest trees by biotic factors from unmanned aerial systems (UAS) offers substantial advantages in forest management practices. In this study, we aimed to develop a novel workflow for facilitating tree species classification and the detection of healthy, unhealthy, and dead trees caused by bark beetle infestation using ultra-high resolution 5-band UAS bi-temporal aerial imagery in the Czech Republic. The study is divided into two steps. We initially classified the tree type, either as broadleaf or conifer, and we then classified trees according to the tree type and health status, and subgroups were created to further classify trees (detailed classification). Photogrammetric processed datasets achieved by the use of structure-from-motion (SfM) imaging technique, where resulting digital terrain models (DTMs), digital surface models (DSMs), and orthophotos with a resolution of 0.05 m were utilized as input for canopy spectral analysis, as well as texture analysis (TA). For the spectral analysis, nine vegetation indices (VIs) were applied to evaluate the amount of vegetation cover change of canopy surface between the two seasons, spring and summer of 2019. Moreover, 13 TA variables, including Mean, Variance, Entropy, Contrast, Heterogeneity, Homogeneity, Angular Second Moment, Correlation, Gray-level Difference Vector (GLDV) Angular Second Moment, GLDV Entropy, GLDV Mean, GLDV Contrast, and Inverse Difference, were estimated for the extraction of canopy surface texture. Further, we used the support vector machine (SVM) algorithm to conduct a detailed classification of tree species and health status. Our results highlighted the efficiency of the proposed method for tree species classification with an overall accuracy (OA) of 81.18% (Kappa: 0.70) and health status assessment with an OA of 84.71% (Kappa: 0.66). While SVM proved to be a good classifier, the results also showed that a combination of VI and TA layers increased the OA by 4.24%, providing a new dimension of information derived from UAS platforms. These methods could be used to quickly evaluate large areas that have been impacted by biological disturbance agents for mapping and detection, tree inventory, and evaluating habitat conditions at relatively low costs.

Early Monitoring of Health Status of Plantation-Grown Eucalyptus pellita at Large Spatial Scale via Visible Spectrum Imaging of Canopy Foliage Using Unmanned Aerial Vehicles

“…Detection of early plant stress [51]; nutrient status [52]; and plant phenotyping including height, flower, and canopy cover [53] use NDVI for mapping vegetation condition and status [54]. However, NDVI requires a multispectral sensor operating in the red-edge nearinfrared (NIR) wavelength range [16,17]. Near infrared cameras are more expensive than RGB cameras and require time-consuming calibration procedures [54] unlike VARI-green.…”

“…To monitor for potential outbreak of pest and disease in E. pellita plantations, field surveys are important for pest and disease detection, however they are labor-intensive, timeconsuming, and are not practical in areas with particularly steep terrain. Instead, precision agriculture techniques using high resolution remote sensing from unmanned aerial vehicles (UAV) with a variety of different sensors, or even conventional red-green-blue (RGB) cameras, would lead to cheaper and more practical monitoring of forest health [14][15][16][17]. Furthermore, rapid development of hardware and software, and improvement in the miniaturization of the sensors have led to the widespread use of UAV for obtaining higher temporal and spatial resolution [18][19][20].…”

“…Remotely sensed data have been used for plant phenotyping, including indices to assess plant health status [16][17][18]30,[35][36][37][38]. The authors of [30] claimed that the use of UAV RGB imagery is more effective for estimation of disease resistance of potato light blight compared to visual assessments.…”

Fusion neural networks for plant classification: learning to combine RGB, hyperspectral, and lidar data

“…Different types of passive and active imaging sensors provide unique information about ecosystems that may be most useful when combined (Anderson et al, 2008;Tusa et al, 2020). Multispectral cameras are accessible, affordable and typically require minimal post-processing to be ready for analysis (Gini et al, 2018;Abdollahnejad & Panagiotidis, 2020). Hyperspectral data are valuable for their ability to capture spectral signatures beyond the visible wavelengths, which often contain descriptive reflectance characteristics across plant types and conditions (Dalponte et al, 2012;Ballanti et al, 2016).…”

Estimation of the Conifer-Broadleaf Ratio in Mixed Forests Based on Time-Series Data

“…A mixed broadleaf-conifer forest is essentially a heterogeneously mixed pixel in remote sensing images [10,11]. When applied to the inversion of vegetation parameters, models often assume that a mixed forest is homogeneous at the pixel level, resulting in a decrease in the inversion accuracy [12].…”