M.S. Thesis
Combining Multispectral and Three-Dimensional Data From Drone Imagery to Detect Forest Insect Damage: An Evaluation of a Novel Approach to Identify the Vertical Structure of Damage in Trees in the Northern Rocky Mountains, USA
Project description: My M.S. research project focused on evaluating methods of detecting forest insect damage and mortality using Unmanned Aerial Systems (UAS)-derived data products in conifer forests of the western United States. The project explored the utlity of point cloud data derived from Structure for Motion (SfM) for the classification and characterization of vertical tree damage of a UAS mission area. I used tools such as ArcGIS Pro, Agisoft Metashape, CloudCompare, and R and Python programming to create new algorithms and establish a novel methodology.
Drone data used for this MS thesis was collected for a broader project assessing tree damage funded by the NASA Commercial SmallSat Data Acquisition Program (NASA CSDA, award #80NSSC21K115) (see ‘Collaborative Research’ section below).
A DJI Matrice 210 drone equipped with a MicaSense RedEdge-MX sensor (5-band calibrated sensor) from Washington State University’s Forest Ecosystem Dynamics Lab (research partners) was used for imagery acquisition. UAS image acquisition mission was planned and operated by Dr. Amanda Stahl. Here preview of the UAS-photogrammetry derived point cloud data I am working with (data is displayed on CloudCompare):
Journal article
Evaluating a Novel Approach to Detect the Vertical Structure of Insect Damage in Trees Using Multispectral and Three-Dimensional Data from Drone Imagery in the Northern Rocky Mountains, USA.
Graphical abstract 
Interactive preview of methodology applied on the point cloud of a single tree
Use the left mouse button to tilt the 3D model and the mouse scroll wheel to zoom in and out.
True-color render and random forest classification of point cloud
Left panel: True color representation of the point cloud. Right panel: RF classification; green is healthy, red is red, gray is gray, and black is shadow.
Random forest classification probability and top-kill algorithm
Left panel: The probabilities of classes shown in left panel on the above interactive layout; darker colors represent higher probabilities of classification. Right panel: Top-kill algorithm applied to point cloud, 3D plane represents the height of top-kill detected by the algorithm.
Presentations
I presented a condensed version of my MS research for University of Idaho’s GIS day 2023 as a contributed talk.
To view the recording of the presentation, please follow this link:
To view the pdf of the presentation file, please follow this link: https://objects.lib.uidaho.edu/gisday/shrestha_gisday2023.pdf