Making the most of available monitoring data: A grid-summarization method to allow for the combined use of monitoring data collected at random and fixed sampling stations

Statistical habitat models, such as spatial generalized linear mixed models (GLMMs) and spatial generalized additive models (GAMs), provide valuable products to habitat assessments and ecosystem-based fisheries management (EBFM) efforts. In particular, derived spatial distribution maps and quantitative relationships of marine organisms to environmental variables (e.g., suitability indices) can be employed to develop or validate ecosystem simulation models. Recent studies fitted spatial GLMMs and spatial GAMs to encounter/non-encounter data collected by different regional monitoring programs that use random sampling designs, so as to enable the production of distribution maps and suitability indices in bulk. However, despite these large efforts, it was not possible to obtain distribution maps for a number of species and life stage combinations, including the juvenile stages of coastal fish species such as croaker (Micropogonias undulatus). In this study, we introduce a grid-summarization method that allows for the combined use of encounter/non-encounter data collected by multiple monitoring programs at random and fixed sampling stations. We demonstrate our grid-summarization method for contrasting species of the western U.S. Gulf of Mexico: red snapper (Lutjanus campechanus), for which data delivered by monitoring programs employing random sampling designs have a satisfactory spatial coverage and the use of monitoring data collected at fixed sampling stations is not necessary; and croaker and brown shrimp (Farfantepenaeus aztecus), for which the combined use of monitoring data collected at random and fixed sampling stations enables or improves the generation of distribution maps. We compare spatial GLMMs and spatial GAMs that rely on the grid-summarization method (new models) to spatial GLMMs and spatial GAMs that do not rely on the grid-summarization method (status-quo models). We found that the grid-summarization method that allows for the combined use of monitoring data collected at random and fixed sampling stations results in reasonable seasonal distribution maps and suitability indices for the species and life stage combinations (e.g., croaker early juveniles, small brown shrimps) that are undersampled by the monitoring programs that employ random sampling schemes. We also found that the grid-summarization method provides reasonable seasonal distribution maps and suitability relationships for species and life stage combinations (e.g., red snapper adults) for which the status-quo method already provided reasonable results and the combined use of monitoring data collected at random and fixed sampling stations is not necessary. For these species and life stage combinations for which the status-quo method worked well, the choice of the grid-summarization method over the status-quo method depends on whether the fisheries analysts wish to produce smoother distribution maps and whether they target higher predictive accuracy at the expense of lower discrimination accuracy when working with spatial GAMs. Our results suggest that additional monitoring datasets that were previously excluded can be employed by statistical habitat models, thereby enabling generation of distribution maps and suitability indices for a wider range of species, life stage and season combinations.