sciscape Publications
← All publications
q-bio.PE · Populations and Evolutionphysics.ao-ph · Atmospheric and Oceanic Physicsstat.AP · Applications

Thermal-time models fitted to 1952–1987 German phenology predict 1988–2026 dates of 32 spring phases 3.3 days too late, and the error does not increase with station warming

Question: do temperature models of spring phenology, fitted to early years, predict later and warmer years without bias, and does any bias scale with local warming? Approach: we used the Deutscher Wetterdienst wild-plant phenology network (32 spring phases of leaf unfolding, shoot growth and flowering; 1,960,361 dates at 1,900 stations) and HYRAS-DE 1 km daily temperature. We fitted a linear preseason model, a thermal-time model and a thermal-time model with chilling on 1952-1987 and predicted 1988-2026. Results: all models predicted too late. The mean error (observed minus predicted) was -3.3 days for thermal time (95% year-resampling interval -3.8 to -2.7; below zero for 28 of 32 phases), -4.3 days for the linear model and -3.7 days with chilling. The chilling term did not improve predictions, and fewer chill days gave earlier, not later, dates. Across station-phase series, the error change did not follow station warming (+0.31 +/- 0.16 days per degC, against about -2.0 expected if the error were proportional to warming). The error grew with time: a step in 1989 plus a later trend (-0.57 +/- 0.18 days per decade after 1993) fitted best. A model fitted to 1952-1969 also predicted 1970-1987 too late (-1.1 days, -2.1 to -0.03) although the preseason did not warm, and a model fitted to 1988-2006 predicted 2007-2026 too late (-1.5 days). GHCN-CAMS and CRU TS give the same station warming as HYRAS within 0.1 degC. These results independently reproduce SCI publication 9c018f33 with more phases, and show that the drift follows elapsed time, not the amount of warming. Limitations: the data cannot separate observer or practice changes, local change that the grid does not show, and drivers that are not temperature. Intervals resample years with fixed parameters and do not include serial correlation.

phenology · spring leaf-out · flowering · thermal time · growing degree days · chilling · out-of-sample validation · temperature sensitivity · DWD · HYRAS · Germany · climate change · model transfer · observer effects

Admitted 2 citing publications

References

8
  1. Temperature models fitted to 1952–1990 German leaf-out data predict 1991–2025 leaf-out 1.7 to 5.1 days too late for seven tree species sci:sha256:5e63bb36c8ace247e7020a6acd9cf5d6e4f40feaffa50a9c8082d6ca2dff5bf3

    [Legacy relation: reproduces] Same network and question for seven tree species (1952-1990 to 1991-2025). We reproduce the late-prediction bias with separate code for 32 phases, add a cross-station warming test and a placebo split, and find a continuing post-step trend where it reported a nearly flat one.

  2. doi:10.1038/nature15402

    Reported declining apparent temperature sensitivity of leaf unfolding and suggested chilling limits.

  3. doi:10.1038/s44383-025-00004-6

    Showed that declining apparent sensitivity is expected under warming without chilling or photoperiod limits.

  4. doi:10.21203/rs.3.rs-9436708/v1

    Preprint arguing that spring leaf-out keeps pace with warming across the Northern Hemisphere.

  5. doi:10.1016/j.agrformet.2015.11.007

    Earlier evaluation of spring phenology models with cross-validation on PEP725 data.

  6. usesdoi:10.1007/s00704-020-03388-w

    HYRAS daily temperature grid used as model input.

  7. usesdoi:10.1038/s41597-020-0453-3

    CRU TS 4.10 used to check station warming.

  8. usesdoi:10.1029/2007JD008470

    GHCN-CAMS used to check station warming.

Cited by

2 publications

Public catalog entries on Sciscape.