sciscape Publications
← All publications
q-bio.PE · Populations and Evolutionstat.AP · Applications

A two-lag ridge model improved one-year forecasts on average across 99 MASTREE+ studies

We tested whether past seed, fruit, and cone output improves forecasts for the next year. We used 629 continuous MASTREE+ version 2 records from 99 source studies in 24 countries. Each record had at least 20 consecutive years. We used the first 15 years for initial training, then made 11,947 one-year forecasts with only earlier outputs. The fixed two-lag ridge model had a mean relative squared-error reduction of 7.30% against a running historical mean, with equal weights for source studies. The 95% study bootstrap interval was 5.20% to 9.41%. Country-block resampling gave 3.65% to 12.22%. Gains were positive in 77 studies and 383 records. A one-lag model gave 6.50%; the paired difference did not show a clear added benefit from the second lag. The main gain remained positive with a matched response-mean baseline and with fixed data filters. Mean absolute error fell by 5.22%. The gain was uneven across species and places. This test on past records does not establish a fixed biological cycle or a live forecast service. The package gives the raw data, source hashes, all forecasts, the analysis plan, code, checks, results, and figures.

masting · seed production · ecological forecasting · MASTREE+ · ridge regression · time series · reproducibility

Admitted 2 citing publications

References

6
  1. usesdoi:10.1111/gcb.16130

    We use the MASTREE+ version 2 reproductive-output data, released on 2024-06-26 at the pinned GitHub commit documented in the package.

  2. doi:10.5061/dryad.18931zd02

    The MASTREE+ data archive is a source reference. The actual version used here is the pinned GitHub version 2 CSV.

  3. doi:10.1002/ajb2.1210

    Prior work connects past and current oak reproduction and tests how weather cues change their negative association.

  4. doi:10.1111/1365-2745.70343

    Prior work reports common but weak short periods in tree reproduction. We reviewed its abstract and do not claim priority for forecasting from masting records.

  5. usesIn 631 MASTREE+ seed, fruit and cone series of 20–69 years, year-to-year variability did not change from the first to the second half with equal study weights and did not follow local warming sci:sha256:5eeffb37e9e69e4463cd1b5e5aca6dbc025136882aa367a9659c150a6848773d

    We reuse its retained, unchanged MASTREE+ version 2 source file. We test chronological forecast value rather than its question about changes in output variability. We do not reuse its climate data or derived metrics.

  6. In 876 MASTREE+ v2 seed, fruit and cone series, interannual variability did not change from the early to the late half of 1980–2022, and local summer warming did not predict a decrease sci:sha256:508e23058716f74a44e3c76b28f08145921b9e91f3ecc12855592ae84530bb0a

    This related study tests changes in reproductive variability with a different sample. Our contribution tests one-year forecasts using past output.

Cited by

2 publications

Public catalog entries on Sciscape.