Academic Hub – Classifying Vegetative Land Cover

Classification of vegetation location, spatial characteristics, and health using advanced remote sensing techniques are featured in this session. Varying techniques and technologies are used representing machine learning, field collection of foliar characteristics and crown structures, and even dashcam videos.

Estimating Nutritive, Non-nutritive and Defense Foliar Traits in Spruce-fir Stands Using Remote Sensing and Site Data

In this study, we linked information regarding concentration of various canopy traits to landscape-level forest health and pest susceptibility using remote sensing technology. Foliar traits that can affect insect herbivory, including nutritive such as nitrogen (N), phosphorous (P), potassium (K), and copper (Cu), non-nutritive  such as iron (Fe) and calcium (Ca), and defensive such as equivalent water thickness (EWT) and leaf mass per area (LMA), were estimated in this study. We used Sentinel-2 and site data to develop trait estimation models in a forest dominated by spruce and fir. Several machine-learning algorithms, namely Random Forest (RF), Extreme Gradient Boosting (XGB), and Support Vector Machine (SVM) were tested. Based on the model performances, where normalized root mean square error (nRMSE) values were considered, XGB algorithm was selected to estimate Ca (nRMSE: 0.16), EWT (nRMSE: 0.12), Fe (nRMSE: 0.19), and K (nRMSE: 0.14). On the other hand, RF was used to model Cu (nRMSE: 0.18), LMA (nRMSE: 0.14), N (nRMSE: 0.16), and P (nRMSE: 0.22). Almost all best-performing models included Sentinel-2 red-edge indices, and depth to water table (DWT) was the most important site variable. We propose a novel framework to establish a connection between the concentration of foliar traits in SBW host foliage and tree susceptibility to the pest. This approach could allow for the assessment of host susceptibility on a landscape level based on the concentrations of foliar traits.

Rajeev Bhattarai, University of Maine

Remotely Sensed Estimation of Live Crown Ratio (LCR) for Eastern White Pine (EWP) Health Assessment

Eastern white pine (Pinus strobus L.; EWP), a ecologically and economically important tree species of New England, USA, has been experiencing recurring needle damage due to a native fungal disease complex over the past decade. White pine needle damage (WPND) mostly causes needle discoloration in the lower part of the canopy in early spring, followed by casting at the end of summer. The repeated WPND leads to lower canopy density and less live crown ratio (LCR) and severely affects tree health. LCR is an imprtant measure of EWP health and vitality but literature on the estimation of LCR through remote sensing is scarce. In this study, we estimated the LCR of EWP-dominated stands using multi-temporal optical (Sentinel-2) and microwave (Sentinel-1) satellite data in four sites in the State of Maine. The random forest (RF) machine learning (ML)-based regression modelling was applied employing structural, chlorophyll and leaf water content indices, microwave backscatters (ratio of Vertical-Vertical [VV] and Vertical-Horizontal [VH] bands), two satellite data-derived products (canopy height and density), and digital elevation model [DEM]) resulted in the highest estimation accuracy (R2=0.61; RMSE=9.45%). Our study highlighted the potential of multi-sensor satellite data for LCR estimation. Periodic LCR estimation will help assess WPND impact on canopy growth and is essential for tree health monitoring, ecosystem structure modelling, and climate change impact studies

Pulakesh Das, University of Maine

Impacts of Spatial Resolution on Vegetation Cover Classification in a Semiarid Landscape  

This study aims to quantify the uncertainty of satellite-derived fractional vegetation covers and investigate the heterogeneity of hillslope vegetation using unoccupied aerial systems (UASs) in the USDA ARS Reynolds Creek Experimental Watershed, Idaho, USA. This study quantified vegetation and ground cover of three sagebrush-dominated plots (Wyoming Big Sagebrush, Low Sagebrush, and Mountain Big Sagebrush) from UAS images using image segmentation and unsupervised classification.

Tao Huang, Boise State University

Automated Detection and Geolocating Hazard Trees along the Electric Powerlines               

Power outages during extreme weather events in the United States are a common occurrence, often caused by tree failure. This requires utility companies to invest heavily in vegetation management and infrastructure restoration. However, monitoring and managing roadside vegetation and infrastructure at a landscape level can be challenging without proper monitoring of their structure and health conditions. Aerial images and LiDAR have limitations in real-time monitoring due to their spatial and temporal resolution. Therefore, ground-level datasets that accurately describe the vertical profile of roadside forests are crucial in comprehensively understanding forest structure and health, enabling optimal management strategies based on local conditions. In a pioneering effort, we introduced the utilization of Dashcam videos as an alternative data source for characterizing roadside forest conditions using deep learning (DL) convolutional neural net (CNN) algorithms. Our study analyzes Dashcam videos captured under diverse weather and seasonal conditions along roadsides. Using YOLO architectures, we trained DLCNN models to classify multilayer vegetation, detect utility infrastructure, and identify tree trunks alongside roads. Our findings indicate that Dashcam videos can feasibly complement traditional methods in characterizing roadside vegetation, offering a cost-effective way to acquire data in real time.

Durga Joshi, University of Connecticut

Assessing White Pine Needle Damage (WPND) Impact on Eastern White Pine (EWP) Health through Modeling Foliar Traits using Remote Sensing Data

Canopy foliar traits are linked to plant health and productivity. There are common remote sensing techniques using field spectroscopy or remote sensing imagery to relate foliar chemistry to the reflectance spectra and map those properties. However, using remote sensing-derived foliar traits data to detect plant disease is largely unexplored. Eastern White Pine (Pinus strobus L., EWP) is an ecologically and economically significant species for the Northeastern USA. A fungal disease complex known as White Pine Needle Damage (WPND), which has been on the rise over the last two decades, poses a serious threat to the health and longevity of EWP. The extent and severity of the WPND at a landscape level and its impact on EWP health are poorly understood. In this study, to understand the impact of WPND on the EWP foliar traits, leaf samples were collected in spring 2022 in Bethel, Maine. Traits such as nitrogen (N), chlorophyll (Cab) and equivalent water thickness (EWT) were measured for 38 healthy and 60 unhealthy trees. Field spectroradiometry was used to collect hyperspectral data from leaf samples. Preliminary results indicate that spectral vegetation indices had a good agreement with N (R2: 0.27), Cab (R2: 0.51), and EWT (R2: 0.20). Lab-based hyperspectral data are being used to better model foliar traits using random forest algorithm to detect WPND and then for landscape level modeling of WPND using satellite remote sensing data.

Sudan Timalsina, University of Maine