A stratified baseline-anchored simulator for precision and reliability metric for longline fisheries observer programs coverage design: A case study of Kenya

Document first presented at the 16th session of the IOTC Working Party on Methods: IOTC-2025-WPM16-23_rev1

Effective fisheries management requires accurate, spatially-resolved data on target, bycatch and protected, endangered and threatened species. This study integrates longline observer datasets (2018-2025) from the Kenyan coastline to standardize Catch Per Unit Effort (CPUE), quantify its uncertainty, and evaluate sampling biases. We employed a comprehensive, reproducible framework using modern statistical and computational tools within the R ecosystem. Data were harmonized and mapped onto a 1° × 1° grid of Kenya's Exclusive Economic Zone. Nominal CPUE (kg/1000 hooks) was standardized using generalized linear and delta-lognormal models, adjusting for spatiotemporal and operational covariates.

Uncertainty in CPUE and species composition was rigorously quantified using analytical and bootstrap-derived Coefficients of Variation (CVs). A nested resampling approach simultaneously captured variation in catch proportions and sampling effort, ensuring realistic propagation of uncertainty. The workflow included diagnostic modules to flag statistical deviations, outliers, and spatial coverage gaps, safeguarding data integrity. Spatial analyses produced gridded summaries of species distribution, relative abundance, and biodiversity, highlighting species-specific hotspots and bycatch concentration zones. A critical focus was addressing non-random sampling bias inherent in voluntary observer programs. The framework assesses the representativeness of observed vessels using historical observer data from Kenya, a vital step for bias correction. Our analysis indicates that while 20% observer coverage yields acceptable precision for common species, rare species require coverage exceeding 60% to reduce uncertainty to manageable levels, with requirements varying spatially and seasonally. This CV-based approach provides a quantitative benchmark for designing efficient, stratified sampling strategies that balance logistical constraints with statistical rigor.