Prior three-year precipitation raised forecast error in seven California divisions during water years 1949-2025
We tested whether precipitation in the previous three water years improved forecasts of the next water year in seven California climate divisions. We used a fixed NOAA nClimDiv snapshot for 1896-2025. Each regression used earlier targets only. There were 77 test years, 1949-2025. The predictor raised pooled variance-scaled squared error by 1.65% relative to the training mean. The conditional 95% five-year block interval was an increase of 0.20% to 3.11%. All seven division point estimates had higher error. Mean lag-one correlation was -0.010 for annual totals and 0.657 for three-year running totals. The latter was near the center of the shuffled-year range, 0.563 to 0.734. Shared years can thus explain the apparent running-total memory under the tested null. The results do not prove annual independence or exclude useful forecasts from other predictors. They do not measure runoff, groundwater, or drought recovery. The package gives fixed source rows, 910 annual totals, all forecasts, code, uncertainty draws, figures, and an independent numerical audit.
California · precipitation · water year · drought · persistence · overlapping windows · chronological forecasts · nClimDiv · block bootstrap · reproducibility
References
7- usesJuly-September ENSO gave no clear precipitation forecast gain in seven California divisions in winters 1981-2026 sci:sha256:613b0ddfe29853a816c1880f84c061841d758ecaa5365c184a32464943d46fda
We reuse its public shared precipitation snapshot and saved NOAA format document. We test full water-year persistence with new code.
- doi:10.1175/JCLI-D-16-0423.1
Prior California deficit recovery analysis reports weak annual precipitation dependence and sensitivity to the recovery threshold. The current forecast and overlap tests have a different scope.
- doi:10.1175/JAMC-D-13-0248.1
Source methodology for the gridded observations used to form the division estimates.
- useshttps://www.ncei.noaa.gov/monitoring-content/data/us/climdiv/monthly/current/climdiv-pcpndv-v1.0.0-20261006
Exact precipitation source version. Full hash and exact California rows are preserved.
- useshttps://www.ncei.noaa.gov/monitoring-content/data/us/climdiv/monthly/current/divisional-readme.txt
NOAA format, units, source construction, and missing-value codes. Saved bytes were reused from the cited SCI bundle.
- https://psl.noaa.gov/data/usclimdivs/data/map.html
Official California climate division codes and names.
- https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc%3AC00005%3Bview%3Diso
NOAA metadata describe the gridded source and division area averages.
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesRandom year order matched winter precipitation transition rates in seven California divisions in 1901-1960 and 1966-2025
We reuse its public shared NOAA precipitation snapshot and verified source-format document. We extend the annual dependence question to an explicit transition-count decomposition with new code.
- The largest three-year dry footprint across 344 US divisions was within a shuffled-year range in 1896-2025
Prior campaign work distinguishes overlap correlation from annual forecast information. It motivated a direct count of separate dry water years rather than overlapping precipitation totals.
- Arctic summer extent had little link to next March extent in 45 seasons from 1979 to 2023
Related method context separates shared-value correlation from forecast information. No data or code reused.
- A linear October-December predictor raised mean January-September forecast error in seven California divisions in 1926-2025
This prior campaign study tests interannual persistence and overlap in running totals. The present study tests information within one water year.
- usesWith a 24-month cap, recovery from low 12-month precipitation ranks took longer than in reordered records across 344 US divisions, 1901-2025
We reuse its public shared source snapshot. It tests annual forecast skill and rolling correlation; we test monthly recovery.