The Tijuana river estuary suffers from overwhelming solid-waste contamination such as marine bound plastics, tires, and sediment. Funded by the US EPA's Border 2020 program (SOLTA-C-19-008), this project intends to establish a framework for bi-national monitoring of trans-boundary, marine bound trash (TBMBT) using light-weight unmanned aerial systems (UASs), also known as drones. The developed framework is intended to benefit 1) border authorities through the establishment of a low-cost, minimally invasive, operational standard to monitor TBMBT in coastal ecosystems and optimize upstream trash capture interventions, 2) natural habitat in the Tijuana River Valley (TRV) and TRE through the identification of trash hotspots to guide clean-up operations, and 3) visitors and residents of the TRV and TRE through the long term benefits of reduced trash volumes from optimized interventions. Three surveys were carried out between 2020 and 2021 collecting data on trash locations using drones and ground-based smartphones. Drone imagery was processed to extract locations of trash using image classification algorithms, while ground-based data was used to complement the data collected from drones and to validate the accuracy of the drone generated data.

UAS monitoring occured at two sites, Border Field State Park on the US side of the Los Laureles Canyon watershed (LLCW), and the Milenio 2000 neighborhood on the Mexico side of the LLCW. Across the duration of the project three surveys were conducted: Jan/Feb 2020, Novemebr 2020, and Jan/Feb 2021. Each survey required one day of field work.

UAS/image monitoring was used as tools to detect trash accumulation. To assist with those goals a fixed-wing AeroVironment Quantix UAS was deployed at border field state park due to its extended flight range capabilities and its dual RGB/NIR camera setup. At the sediment basins locations in Border Field State PArk, and in the Milenio 2000 neighborhood, a Phantom 4 quadcopter was deployed able to navigate accurately at low altitude between the canyon walls. Flights occured between the hours of 10:00 am – 2:00 pm to reduce impact from shadows in the collected imagery. Flight dates were conditional on agreeable weather conditions, in consideration of bird nesting sseason, as well as in coordination with land managers (State Parks, California Border Patrol, and Mexico Aviacion Civil). For US flights, survey dates occured on weekends when FAA airspace at the Tijuana river valley is in class G (unrestricted to UAS deployment). UAS flights were executed at low altitude, approximately 150 ft above ground level, capturing images with a ground resolution of approximately 1.2 cm/pixel irrespective of the imaging sensor used. A fine image resolution was targeted to capture macro and mega trash, which have defined diameters of 2-10 cm, and >10 cm respectively.

Ground data collection campaigns of trash locations and conditions were carried out concurrently with the UAS surveys on the same survey days. Ground survey methods of trash debris developed by the Bay Area Stormwater Management Agencies Association (BASMAA, 2017) and the Stormwater Monitoring Coalition (San Francisco RWQCB, 2004) were adapted and implemented for this project. Surveys were carried out using a digital survey sheet developed using Esri’s Survey123 online platform (ESRI, Redlands, CA) that can be accessed at: https://arcg.is/Tuyf5. During each survey, the survey sheet was accessed in the field using GPS enabled tablets or smartphones and allowed collecting visual qualitative data of trash such as photographs, categorization of trash type, and ratios of trash types within larger patches in accordance with the reference methods. Measurements of trash were also performed using tape measures whenever possible (e.g. length and width of trash patch). Field data collection was organized in conjunction with local land managers across the LLCW.

UAS DATA Processing:

Individual UAS images were photogrammetrically processed using Metashape Pro (Agisoft LLC, St. Petersburg, Russia), a popular commercial Structure from Motion (SfM) photogrammetry software package with low entry costs relative to competing commercial software. Using SfM photogrammetry, the individual UAS images were geometrically corrected and stitched together into orthometric images (orthophotos) with uniform scale across the whole scene. Separate orthophotos were generated for each survey and for each survey location. Ground control points were not available at the survey sites and due to budget limitations SfM processing was carried out without the use of ground control. We estimate 3-5 m horizontal accuracy of the derived orthophotos (equivalent to the UAV on-board GPS receiver).

The focus of this project is to detect trash from UAS imagery, and here we tested various image classification algorithms from the field of computer vision. Image classification algorithms have proven to be capable of segmenting aerial imagery into features and landcover classes, and here we tested automatic unsupervised classification as well as semi-automatic supervised classification strategies.

SHARED DATA:

• Orthophotos of each UAS survey (raster tif format).
• Ground based data of identified trash locations (vector shapefile format).
• Classification results from supervised object-oriented classifier (raster tif format).