Assessing joint effects of sampling design and annotation quality on benthic cover estimates through Monte Carlo simulations

Ecological monitoring increasingly relies on image-based data collection coupled with machine learning annotation. Yet, the combined impact of sampling effort and annotation quality on indicator precision remains poorly understood, limiting evidence-based resource allocation in biodiversity monitoring programs. We investigate this question for benthic cover estimation using photo-quadrat surveys, which involve errors from spatial quadrat sampling and image annotation. While prior studies have examined these error sources independently, their joint effects remain unquantified. We developed a photo-quadrat sampling and annotation simulator and applied it within a Monte Carlo framework to quantify how different combinations of design choices affect cover estimates on known reference maps. We apply our approach to a reference map from shallow Mediterranean benthic habitat (Tyrrhenian Sea), comparing four quadrat placement strategies with varying annotation accuracy across 10,000 replicates per parameter combination. Our results challenge conventional benthic monitoring assumptions: (i) random quadrat placement significantly outperformed structured transect-based strategies, (ii) improved annotation performance did not systematically improve cover precision, and (iii) extensive sampling with imperfect annotation consistently outperformed perfect annotation of fewer samples. Our complete simulation framework, analysis code and data are made publicly available, along with a free web application for broader research and educational use.