Semi-automated Detection of Seagrass Scars in Tampa Bay from Aerial Imagery: An Application for ArcGIS Pro Deep Learning

Abstract Seagrasses are of critical importance as they provide food and structure for many species from larval to adult stages, stabilize sediments, and sequester carbon and nutrients. Seagrass restoration is expensive and not always possible; therefore, prevention of seagrass loss is key to its management. Seagrass scarring, the mechanical removal of seagrass and sediments by the boat propeller and hull, is a contributor to seagrass loss in shallow coastal areas. Identifying the location and abundance of seagrass scarring can assist management by directing and evaluating interventions, but quantifying scars has largely relied on manual digitization, which is time-consuming and inconsistent. Deep learning is a new tool for object detection, yet it is computationally intensive and can be inaccessible to natural resource managers. We show that ArcGIS Deep Learning tools are effective for object detection from remotely sensed data by developing a model to detect seagrass scars in Tampa Bay, FL, USA. The model detected 23,488 seagrass scars across areas of continuous seagrass in Tampa Bay, and model precision was 0.80, which is within the typical range for related applications. As seagrass abundance, distribution, and scarring change year to year, a deep learning model such as the one developed in the present study could be applied year after year to new imagery to monitor seagrass disturbance and recovery.