Data from a second pool gave no clear gain in linear forecasts of weekly shrimp catch at Quebrada Prieta in 2001–2008

We tested whether the latest shrimp catch rate at a second pool improved forecasts at two fixed Quebrada Prieta pools in Puerto Rico. The data have three groups: Atya lanipes, Xiphocaris elongata, and Macrobrachium spp. We used 1993–2008 source records and 2,211 test predictions in 2001–2008. Each linear model used only earlier training years. The local model used season, linear time, and the latest local log catch rate. Adding the latest log catch rate of the same group at the other pool increased mean squared log error by 0.49%. A central 95% range from common two-year test blocks allowed an error reduction from -1.87% to +1.51%. The latest local value reduced error by 44.66% against a season-and-time model. Checks with two local lags, other horizons, and no time term did not establish a gain from the other pool. A three-pool check in 2003–2008 gave a 0.49% reduction, with a range from -3.92% to +4.19%. The source audit found repeated year–week keys and many all-group-zero weeks after 2008. We therefore fixed the earlier main scope before forecast fitting. The results assume that unique source week labels match survey weeks. They concern these models, one stream, and eight test years. They do not prove biological independence or equal skill. The package has unchanged source data, all predictions, figures, code, provenance, and independent calculation checks. A clean offline run reproduced all 33 result and figure files.

ecological forecasting · freshwater shrimp · Luquillo · catch per trap · spatial information · null result · reproducibility