Comparing image-based classification and field sampling methods for beach litter quantification in Can Gio, Southern...

Can Gio District in southern Vietnam, including the Can Gio Biosphere Reserve and the urbanized beaches of Can Thanh Town, is a rapidly developing coastal area facing increasing pressures from marine plastic pollution. This study compared beach litter monitoring using 1 m × 1 m quadrats with in situ collection against smartphone imagery captured via the Mergin Maps app and later annotated in DotDotGoose. Mergin Maps was selected because it provides an open‑source, smartphone‑based GIS platform for rapid, low‑cost capture of georeferenced quadrat images, which is more feasible for routine monitoring in resource‑limited coastal settings than drone surveys or fully manual transect methods. Across four sandy beaches, at Bui Lam Beach field surveys recorded 399 items, while image-based analysis detected 410 items; at Bui Lam-Tac Xuat Beach, field surveys recorded 212 items, compared with 236 from image-based analysis, indicating strong tidal influence and clam farming activities at these sites. At Can Thanh Park, 189 items were counted in the field compared with 149 items in images, whereas at Tac Xuat, a tourist beach with regular cleanups, 82 and 71 items were recorded by field and image-based methods, respectively. Overall litter densities exceeded 10 items m⁻2 for both methods, and plastics accounted for more than 96% of all items, with foam and fragmented single-use products dominating. Although a relatively high error percentage was observed between the two monitoring methods (mean 43.15% ± 32.81%, ranging from 0 to 116.67% across quadrats), this was mainly due to under-detection of items that were partially buried, obscured by vegetation, or mixed with organic matter during image-based classification. Despite this limitation, statistical analysis revealed no significant difference between image-based classification and field data regarding litter count (p > 0.05). Mergin Maps provides a rapid assessment tool for monitoring marine litter, with statistically equivalent results to field sampling and approximately 2.5-3 times faster field data collection than conventional quadrat sampling.