Environmental Remote Sensing with Unmanned Aircraft Systems

Small Unmanned Aircraft Systems (sUASs) equipped with optical sensors are capableof remotely sensing landscapes and wildlife at spatial and temporal resolutions that werepreviously inaccessible due to technical and budgetary limitations. Conventional remotesensing and photogrammetric workflows can b...

Full description

Bibliographic Details
Main Author: Burnett, Jonathan D.
Other Authors: Wing, Michael, Parrish, Chris, Morrell, Jeffrey, Hailemariam, Temesgen, Shaw, Dave, Forest Engineering Resources and Management, Oregon State University. Graduate School
Format: Doctoral or Postdoctoral Thesis
Language:English
Published: Oregon State University
Subjects:
Online Access:http://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/qz20sz35q
Description
Summary:Small Unmanned Aircraft Systems (sUASs) equipped with optical sensors are capableof remotely sensing landscapes and wildlife at spatial and temporal resolutions that werepreviously inaccessible due to technical and budgetary limitations. Conventional remotesensing and photogrammetric workflows can be applied to the resulting high resolution imageryto facilitate new types of scientific inquiry. This dissertation explores three novel applicationsusing low-cost consumer grade and commercial grade sensors onboard an sUAS.The first application uses a quadcopter equipped with a consumer grade camera to detectSwiss needle cast disease (SNC) in Douglas-fir stands in western Oregon. Swiss needlecast is a non-fatal foliar disease in Douglas-fir that reduces annual growth and stumpagevalue. Conventional detection methods rely on manned aerial detection surveys that aretedious. However, sUAS technologies offer a potential alternative. The presented methodfuses sUAS technology with Structure from Motion, automatic stem segmentation and binomialclassification with generalized additive models. Four 1.6 ha sites containing morethan 3500 Douglas-fir trees were surveyed with a sUAS. Visibly infected trees were distinguishedfrom not visibly infected trees with much greater than random chance (kappa> 0.4) at the four sites surveyed. Near-infrared (NIR) information was not pertinent tosuccessful SNC detection, vastly simplifying the operational complexity of future surveys.The method described in this chapter facilitates mapping of individual Douglas-fir treesinfected with SNC in the mountains of western Oregon.The second application of sUAS technology expands upon the first by adding a narrowbandmultispectral camera (NMC), additional survey sites, and surveys in differentyears and months. The effectiveness of narrowband multispectral cameras for assessingvegetation condition has been heavily researched, but the application to Swiss needle castdetection in an industrial forest has not been previously described. Eight 1.6 ha sites encompassingmore than 6000 trees were surveyed with a consumer grade camera and a NMCin 2015 and 2016. SNC detection reliability tended to be better with the NMC (kappa difference> 0.10) than the consumer grade camera when surveys were conducted in fullysunny conditions, but the differences were negligible in cloudy conditions. Summer imagingwith the NMC yielded highly variable results in comparison to the more stable springsurveys and suggests that summer surveys are not operationally plausible. Detection surveysof the same sites in two different years revealed higher-than-expected levels of diseasestatus change between years. Employing stricter probability thresholds on the classificationrules reduced detected change from > 200 trees site to < 50 trees site at the cost ofcreating a third class of trees having an uncertain disease status. There was no evidencethat foliage retention related to classified diseased status although additional study is recommendeddue to the limited inferential power afforded by the small sample size (n <28). Many regulatory, technical, and computational hurdles must be overcome before largescale implementation of the method can be attempted.The third application uses the integrated camera on a DJI Phantom 3 sUAS to conductphotogrammetric measurements of baleen whale morphology, which is an indication ofwhale health. UAS photogrammetry has been previously explored and shown to produceaccurate measurements, but methods between surveys vary widely, indicating a need forstandardization. We imaged 89 gray and six blue whales with a Phantom 3 sUAS. Whaleswere measured within the images and scaled to metric units using barometric altitude. Linearmixed models with error terms for flight and date were used to to correct scaling error.Post-correction estimates of 1 m calibration object contained 0.17 m less error and 0.25 mless bias than no correction. Total propagated uncertainty analysis was used to examineerror contributions from scaling and image measurement (digitization) to determine thatdigitization accounted for 97% of total variance. Additionally, we present a new body sizemetric termed Body Area Index (BAI). BAI is scale-invariant and is independent of bodylength (R2 = 0:11), enabling robust comparisons of body size within and among populations,and over time. Along with this study we present a three-program analysis suitethat measures baleen whales and applies scale corrections to produce 11 morphometricattributes from UAS imagery. The photogrammetric method presented and associated softwarefacilitate efficient and standardized analysis of any whales that meet the assumptionof a parabolic shape.Environmental remote sensing with sUAS can produce survey data at very high detail(i.e., tree-level) and provide high measurement precision without the use of high-costsensors. However, regulatory limitations within the United States National Airspace combinedwith the low-endurance of most multirotor sUASs limits efficient use to small areas,or one or two whale sightings. sUAS survey data is of such high resolution that data storageand management because burdensome even when survey areas are small. Furthermore,low-cost sUAS systems suffer from reliability challenges and steep learning curves thatcan heavily limit technology accessibility. In spite of the tradeoffs relative to manned surveys,sUAS remote sensing provides researchers with unprecedented access to data of hightemporal and spatial resolution at low costs without putting human lives into the air.