A degree-day model with gridded air temperature, fitted to 1951–1985 only, predicts 85–91% of the 5.7-day earlier ice-off at 100 Northern Hemisphere lakes in 1986–2020
Zhou et al. (2026, PNAS) found across 724 lakes that ice-off sensitivity to air temperature increases steeply above a mean winter air temperature of about -8.8 °C. They used this space-for-time relation to project accelerated ice loss. We tested the relation over time inside lakes. We used 100 lakes from the NSIDC Global Lake and River Ice Phenology Database (51 USA, 38 Finland, 11 other) with at least 25 ice-off dates in each of 1951–1985 and 1986–2020, and monthly air temperature at each lake from CRU TS 4.09 and GHCN-CAMS. Mean ice-off was 5.7 days earlier in 1986–2020 (95% interval 5.2–6.3; bootstrap of 45 spatial cells). A degree-day model with fixed parameters, fitted to 1951–1985 data only, predicted 85% (77–95%) of this advance with CRU and 91% (82–102%) with CAMS, with the lowest lake-level error. Linear sensitivity to annual temperature multiplied by warming predicted 79–80%; threshold-adjusted sensitivity predicted 86–88%. Within lakes, sensitivity to annual temperature became stronger by 0.57 (CRU) and 0.49 (CAMS) days/°C. The fixed model gave 0.38 and 0.60. The across-lake threshold relation predicted 1.23 and 1.19, but the difference was not resolved. The fixed model also gave the observed within-lake curvature of ice-off against winter temperature (-0.20 vs -0.19 days/°C²). Lakes whose winters crossed -8.8 °C did not advance more than other lakes. Sensitivity to the two months before ice-off increased more than the fixed model gives, but this excess changed with product, window and detrending (10 of 24 variants exclude zero). Thus most of the advance and of the nonlinear sensitivity is an expected result of fixed thermal-time physics under the observed warming. A residual under-prediction of 0.5–0.9 days remains, mostly in Finland. Limitations: gridded monthly temperature and interpolated daily values, records that end by 2020, and lakes concentrated in Finland and the north-eastern and mid-western USA.
lake ice · ice-off · phenology · degree-day model · space-for-time · temperature sensitivity · climate change · hindcast · nonlinear response · thresholds · NSIDC G01377 · CRU TS · GHCN-CAMS
References
11- challengesdoi:10.1073/pnas.2610752123
We test the space-for-time assumption behind its threshold-dependent acceleration of ice-off sensitivity with within-lake data for 1951–2020.
- usesdoi:10.7265/N5W66HP8
Source of all ice-off dates (NSIDC G01377 Global Lake and River Ice Phenology Database).
- usesdoi:10.1038/s41597-020-0453-3
CRU TS monthly air temperature (version 4.09) at each lake.
- usesdoi:10.1029/2007JD008470
GHCN_CAMS monthly air temperature as a second product.
- doi:10.1111/j.1365-2486.2010.02249.x
Earlier evidence that ice timing is most sensitive where air temperature is near 0 °C.
- doi:10.1126/science.289.5485.1743
Long-term trends in Northern Hemisphere lake and river ice.
- doi:10.1038/s41558-018-0393-5
Loss of lake ice and more ice-free winters with warming.
- A fixed thermal-time model predicts lower spring temperature sensitivity at warm US lilac and honeysuckle sites in 1956–2009, but not the observed increase in sensitivity over time sci:sha256:c815add42242e8d2f8f213f08dbc384e8ad794e2578349ae8b7b145e55078c11
Same logic of fixed thermal-time null for changes in apparent temperature sensitivity, applied to lilac phenology.
- Spring phenology temperature sensitivity at Japanese weather stations did not decrease in 1953–2025 beyond a forcing-only model, which also explains lower cherry sensitivity at warmer stations sci:sha256:a30b18e0b9dbed22d7ae3539df648fe0cae472a7014dbd9770521051485120f2
Forcing-only null model for temperature sensitivity of spring phenology.
- Spring plant phenology at 81 Japanese weather stations became more sensitive to temperature from 1953–1986 to 1987–2020, while fixed thermal-time models predict less sensitivity sci:sha256:a372304ce3a1aadd55322628f187e2740d8c2ae143952729f53d47badda6abbb
Observed increase in phenology temperature sensitivity compared with fixed thermal-time models.
- 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
Out-of-sample test of temperature models fitted to early decades of phenology data.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- supportsThermal-time models fitted to 1960–1990 air temperature predict 44–64% of the later lake ice freeze-up in 1991–2026, and 73–84% of the earlier break-up, at lakes in Finland and the LIAG database
Parallel test of a degree-day model fitted to early ice-off data with gridded monthly temperature; our CRU-adjusted LIAG break-up result (84%) and the Finnish residual agree with it. We add freeze-up, Finnish records to 2026 with daily station temperature, and a station-inhomogeneity check.
- supportsAcross Finnish lakes, ice-off sensitivity rises with winter temperature only through the annual-temperature predictor and does not predict within-lake change in Finland or Minnesota, 1949–2026
Same space-for-time question with 100 NSIDC LIAG lakes to 2020. Our partly independent Syke (to 2026) and Minnesota DNR (to 2023) data give the same within-lake result: about 1 day per °C more sensitivity, not predicted by the cross-lake relation or ordered by the -8.8 °C threshold. We add the predictor decomposition of the cross-lake pattern.