June oxygen reduced August prediction error in fixed linear models for 40 northern lake records in 2001–2017
We tested whether June dissolved oxygen improved August predictions in later years. We used open EDI lake profiles and a fixed depth band within each lake record. Models were fitted with 651 paired observations from 1980–2000 and tested with 574 pairs from 2001–2017. The panel had 40 source lake IDs in seven geographic groups. Adding June oxygen reduced equal-lake mean squared error by 21.9% in the primary model, from 2.187 to 1.708 (mg/L)². A joint region and three-year-block bootstrap gave a nominal 95% change interval of −1.347 to −0.023 (mg/L)². Error decreased in 31 records. This model used separate lake time slopes. Even after adding oxygen, it had more error than the historical August lake mean. Models with one shared time slope or no time slope had lower errors; adding June oxygen reduced their errors by 25.1% and 22.9%. Smaller depth and coverage panels had intervals that included zero. These retrospective predictions assume that the August sample date is known. They do not establish a cause or validate a live service. The package has full source data, code, fixed models, predictions, figures, and repeat instructions.
lake oxygen · seasonal prediction · environmental monitoring · fixed depth bands · later-year test · open data
References
9- usesdoi:10.6073/pasta/7686da566d3589b95b01c530c8cd0a22
The full EDI revision 4 interpolated profiles and lake metadata supply the observations.
- supportsSampling-date drift hid part of the deep-water oxygen loss in 183 temperate lakes (1941–2017), and in 120 northern lakes the loss was present by mid-June with no change in summer depletion rate sci:sha256:790bf5319d249f07cc3718b6795d9676c2e772b0001b7555cab4644dc9ddff0b
Its finding that oxygen loss was present by June motivates the later-year prediction test. No derived layer or fitted result is reused.
- usesWith a 5 mg/L oxygen limit, adding an 18°C limit raised the median paired depth slope by 0.80 percentage points per decade in 149 northern lakes, 1980–2017 sci:sha256:f43507510042557a830bb7925dc492b06f048755bd028b24237659d9d9037449
We reuse its losslessly compressed copy of the unchanged EDI table and metadata. Its analysis code and habitat results are not reused.
- supportsdoi:10.1038/s41586-021-03550-y
Documents the lake oxygen losses that motivate early-season monitoring.
- supportsdoi:10.1111/gcb.16525
Links longer stratification to oxygen loss without a consistent change in depletion rates.
- supportsdoi:10.1002/lno.11543
Shows that winter and spring conditions can set initial oxygen conditions for summer.
- doi:10.1002/env.2765
Prior oxygen forecasting at shorter time horizons with vertical profiles.
- doi:10.1002/ecy.3431
Provides forecast evaluation principles, including separate evaluation data and reference models.
- supportsDeep-water oxygen loss is a shared decline across 195 temperate lakes, but lake-specific trends did not persist between 1990–2003 and 2004–2017 or predict 2018–2025 sci:sha256:9b22f2a196d9f58f7c588670c4612ac959bc78cf54d5bc651797d590bb565bb0
Its poor persistence of individual lake oxygen trends supports the use of simple reference models. It uses the same source compilation, so it is not independent data.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesAn update from prior August reduced squared oxygen forecast error by 8% against past means in 79 lake records, 2001–2017
We reuse its retained unchanged source files and build on its within-season forecast test. We write new code for an annual test.
- supportsA fixed June profile model showed no clear error reduction against recent August oxygen means at seven Wisconsin lakes in 2006–2025
Its June-to-August test showed that added oxygen skill depends on model form and reference rules. We extend the issue-date and recent-mean comparisons in NTL records through 2025.