With three capture inputs, a new moon grid kept 108 forecast targets compared with 88 calendar targets on two Portal plots in 2005–2019

We tested how survey time grids affect a rainfall forecast benchmark at Portal, Arizona. Both grids used the same two untreated plots and three recent capture values. Calendar targets were monthly mean captures three months ahead. New moon targets were captures three new moons ahead. In 2005–2019, the new moon grid kept 108 targets in 99 months. The calendar grid kept 88 targets in 88 months. Thus the new moon test had 23% more targets and 13% more target months. This grid comparison was added after the first calendar results and related SCI studies were read. It is exploratory. Both ridge models used recent captures and season. A second model also used two windows of past county rain. Added rain skill was −0.56% on the calendar grid and −3.59% on the new moon grid. Descriptive 95% intervals from 12-month blocks were −4.73% to +2.61% and −9.32% to +1.35%. Neither showed a clear gain. A sparse 2022–2025 check gave positive point estimates, which limits a general null claim. Changing the grid also changes the targets and issue dates. Absolute errors are therefore not a controlled comparison of grid accuracy. We preserve exact source extracts, provenance, code, all forecasts, fitted states, figures, and repeat instructions. Independent count and numerical checks passed. The results concern linear forecasts of a capture index at one site. They do not identify a causal rain effect or true population abundance.

biodiversity · ecological forecasting · desert rodents · Portal Project · rainfall · sampling interval · missing data · reproducibility