Advances and cost reductions in unmanned aerial vehicle (UAV) technology led us to conduct initial testing to identify areas of our invasive species field operations that would benefit from this remote sensing platform. For our study, we used a DJI Mavic Pro drone, DJI GO 4 flight software, and DroneDeploy flight planning and processing software. We collected imagery and video primarily during our regularly scheduled fieldwork and visually reviewed the collection for the presence of our target species. A UAV benefited our miconia (Miconia calvescens) and Rapid ʻŌhiʻa Death (Ceratocystis huliohia and C. lukuohia) operations by enabling us to detect miconia and dead ‘ōhi‘a trees in previously unconfirmed locations. Devil weed (Chromolaena odorata), Himalayan blackberry (Rubus discolor), Cape ivy (Delairea odorata), and Cane ti (Tibouchina herbacea) were not detectable in the sense that we could only identify these species at a confirmed location and from very low altitudes. While surrounding vegetation hindered the detection of many of our target species, UAV surveys offered some benefit to our field operations in challenging and inaccessible areas.
OISC field operations commonly occur in remote wilderness locations that require extensive hiking and can involve adverse conditions such as challenging terrain and vegetation. For this reason, portability was a major consideration in UAV selection. Limited funds and time for developing and learning the system was another important consideration. We selected the DJI Mavic Pro, released in October 2016, as it offers the highest performance in image/video quality, flight time, stability, object avoidance capability, and ease-of-use relative to its unprecedented portability. It also has a large community of users and software developers that can provide support and additional functionality.
We selected software that enabled us to progress to field operations quickly. Different workflows and budgets may lead to other software choices. We conducted manual flights using the DJI GO 4 app. We used DroneDeploy to conduct autonomous mapping flights.
Analysis and Image Processing Software
UAV imagery was visually analyzed during flight by the operator and again post-flight in the office on a larger, full resolution display. We processed and orthorectified imagery from autonomous mapping flights using DroneDeploy.
For the purposes of this study, we defined a detectable plant as a target plant that can be identified in UAV imagery and was found in a previously unconfirmed location. This is in contrast to target plants that were previously identified and then specifically located with the UAV for the purpose of developing a search image or collecting sample imagery.
In general, we selected our target species survey locations based on the regular scheduling of our field surveys to minimize the impact of the flights on our operations. Additional criteria for sites included the presence of good vantage points for maintaining visual line of sight and the absence of controlled airspace.
Miconia was successfully detected in a previously unconfirmed location in Maunawili using a UAV. Manual UAV surveys in an area of recently treated plants yielded video in which a recently hand-pulled plant could be seen in post-flight analysis. The opportunistic use of a UAV during ground surveys also led to confident identification of a suspected miconia plant seen from a distance (Fig. 2). Large areas of monotypic uluhe (Dicranopteris linearis) were surveyed with a UAV and determined to be free of miconia. An autonomous mapping flight resulted in an orthorectified image of a small portion of the survey area (Fig. 3). UAV surveys in areas with challenging vegetation, and proximity to residences and hazards resulted in initial surveys of previously un-surveyed areas (Fig. 4).
Fig. 4. A UAV survey area with challenging vegetation, residences and hazards (powerlines) in close proximity.
Rapid ʻŌhiʻa Death (ROD)
Suspect ROD trees were successfully detected using a UAV (Fig. 5, 6). Trees which could not be identified on foot due to steep terrain and dense vegetation were easily seen in UAV imagery due to the aerial perspective.
Fig. 5. Several ROD suspect trees (circled) in UAV imagery.
Fig. 6. Individual ROD suspect tree in UAV imagery.
Fig. 7. UAV imagery of field crew inspecting devil weed.
Devil weed was not successfully detected in a previously unknown location with a UAV in any of the test locations. Devil weed plants were distinguishable in the imagery only when captured from very low altitude (<50 ft. AGL) and focused on known plant locations (Fig. 7).
Fig. 8. Himalayan blackberry (circled) in UAV imagery.
Himalayan blackberry was not successfully detected with a UAV. Plants were distinguishable in the imagery only when captured from very low altitude (<20 ft. AGL) and focused on known plant locations (Fig. 8).
Fig. 9. Cape ivy (circled) in UAV imagery.
Cape ivy was not successfully detected with a UAV. Plants were distinguishable in the imagery only when captured from very low altitude (<20 ft. AGL) and focused on known plant locations (Fig. 9).
Cane ti was not successfully detected with a UAV. Plants were distinguishable in the imagery only when captured from very low altitude (<30 ft. AGL) and focused on known plant locations (Fig. 10). Flowering structures were the most detectable plant part.
One of the most successful uses of the UAV in this study was surveying for miconia in monotypic uluhe areas in place of time-consuming ground surveys. Although we could survey some of these areas by helicopter, UAVs can save money and increase safety by reducing helicopter use. The very low stature of these uluhe areas allowed for a confident visual assessment of the imagery. However, this was the only test area with such prominent and clearable features and miconia was the only test species with a large enough habit to say it was not present. In addition, while only miconia plants visible through the canopy would be detectable by UAVs, we often find miconia under dense canopy especially in its seedling and immature growth stages. UAVs also provided some level of survey for areas that have remained un-surveyed for miconia.
Although we did not successfully detect devil weed in this study, further testing is highly recommended for this species. Due to UAV permission issues, ideal test locations could not be used. In addition, an aerial search image could not be adequately developed for the operator and post-flight footage reviewers. Greater devil weed detectability may be realized by surveying during its flowering season.
Himalayan blackberry, Cape ivy, and cane ti are small, relatively rare, and often found under canopy in O‘ahu’s landscapes. These plants were completely un-detectable with the UAV. They will likely remain very challenging in future remote sensing detection efforts. Higher (sub-centimeter) resolution may improve detectability of these species when they are exposed through the canopy. Greater utilization of any type of sensor data will likely result from the application and improvement of object detection algorithms coupled with ideal sensor data, which is yet to be determined.
Although the majority of our target species were undetectable in our study, the UAV was helpful in certain aspects of our miconia and ROD operations. Miconia operations benefited through the fast and easy survey of traditionally slow and challenging uluhe-covered areas. The UAV also provided a method for surveying areas for miconia that were previously un-surveyable by ground and by helicopter due to a combination of challenging vegetation, and proximity to residences and hazards. ROD ground sampling operations benefited from the UAV's aerial view, which assisted in the location of suspect ROD trees.
Video edited by Elana Shi.
Mahalo to the Hawai‘i Invasive Species Council, the Honolulu Board of Water Supply and the Hawai‘i Tourism Authority for funding OISC's miconia work and to the Pacific Cooperative Studies Unit for logistic support.