Rain from the previous year gave no clear biomass prediction gain in 17 test years at Konza watershed 001d

We tested whether rain in the previous growing season improved predictions of graminoid biomass at one annually burned, ungrazed Konza Prairie watershed. Public records gave 82 annual soil means in 1984–2024. Two linear models used current April–August rain, with or without previous rain. Each test prediction used only earlier years for training. Strict rain coverage gave 17 test years within 2004–2024. Adding previous rain increased mean squared error by 3.09%, with equal weight for two soil positions. The conditional 95% range for error reduction was −11.54% to +3.43% from 9,999 shared three-year block draws. Different rain windows, fixed training, log biomass and Daymet rain also left the gain uncertain. The source audit found that overlapping rain gauges would raise a sum of all April–August 2010 rows from 596.6 to 1,225.2 mm. We selected one instrument using the documented history. The result applies to the tested models and sample. It does not establish that ecological memory is absent. The package preserves source data, provenance, code, all predictions, figures and offline instructions. A separate run reproduced 17 numerical files and two PNG figures.

grassland · ecological memory · precipitation legacy · Konza Prairie · biomass · chronological prediction · data provenance