Models with rain and census date did not show clear plant count trends on two Portal control plots in 1990–2025

We tested whether plant counts changed through time on two continuously unmanipulated Portal plots in Arizona. We used release 6.117.0 and the same 32 quadrats, with a total area of 8 square metres. Rain before each census had to have at least 95% hourly coverage. The sample had 29 winter censuses from 1990 to 2025 and 26 summer censuses from 1990 to 2024. The response was the natural log of one plus recorded plants per square metre. Models included time, log rain, and census date. The winter time slope per decade was −0.379, with a nominal 95% interval of −0.845 to +0.086. The summer slope was −0.531, with an interval of −1.281 to +0.219. Both Holm-adjusted p values were 0.212. These results establish neither a persistent linear decline nor stable counts. The winter slope before 2022 was close to zero, so the four recent censuses strongly affect the full-period estimate. A later check that omitted 2011–2016 gave a clearer summer decline of −0.788, with an interval of −1.364 to −0.212. This omission removes years with known gauge problems and high plant counts. The summer conclusion therefore depends on year selection. Limits include one site, missing years, rain-gauge error, count methods, and a simple linear time model. The package preserves source tables, hashes, a census-window data table, all results, figures, code, and instructions. Independent CSV and matrix calculations matched the primary results.

desert plants · rainfall · long-term monitoring · Portal Project · plant abundance · reproducible analysis