Pith. sign in

REVIEW 4 major objections 5 minor 24 references

Forecasting Extreme High Summer Temperatures in Paris and Cairo Using Gradient Boosting and Conformal Prediction Regions

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Paris heat waves can be forecast two weeks ahead from routine weather-station data, with useful skill on the hottest days.

desk verdict Honest setup and useful goal, but the central accuracy claim is unsupported by the analysis as presented; worth a serious referee only if the evaluation is redone. read the letter →

arxiv 2506.02349 v1 pith:ZUZYQTL5 submitted 2025-06-03 stat.AP

Richard A. Berk This is my paper
classification stat.AP
keywords HeatWavesSteadmanIndexForecastingSupervisedMachineLearningGradientBoostingConformalPredictionRegionsParis-MontsourisWeatherStationCairoInternationalAirport
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that unusually high summer temperatures can be usefully forecast two weeks in advance from routinely collected weather-station data, using gradient boosting with a quantile loss and conformal prediction regions to quantify uncertainty. In Paris, the model trained on one year's March through August data and tested on an adjacent year places the June and July 2019 heat waves roughly two weeks ahead of their occurrence, and the ten hottest days in a 2020 evaluation sample fall inside their conformal regions. The same recipe applied to Cairo at the $Q(.99)$ level is less accurate, reflecting the desert climate's sudden temperature swings, but it still flags an early May heat event. If the approach holds, it offers a low-cost, locally trainable early-warning tool for heat-related health and ecosystem risks.

What carries the argument

The load-bearing mechanism is quantile gradient boosting: the gbm implementation of gradient boosting with an asymmetric quantile loss function, trained on seven predictors (wind direction, wind speed, air temperature, atmospheric pressure, visibility, dew point, and relative humidity) each lagged by 14 days, so that each day's forecast uses only information from two weeks earlier. A second component is conformal prediction: after fitting, residuals are passed through an AR(1) filter to weaken temporal dependence, and the whitened residuals serve as nonconformal scores to build adaptive prediction regions around each forecast. The quantile target is $Q(.95)$ for Paris and $Q(.99)$ for Cairo, chosen because the goal is statistically rare, extreme heat rather than average conditions.

What would settle it

Take a held-out Paris summer with known heat waves, train only on the adjacent year, and check whether the 10 hottest observed 2pm temperatures fall inside the two-week-ahead $Q(.95)$ conformal regions at the stipulated 0.70 coverage; if coverage on the hottest days falls far below nominal, or the whitened residuals retain clear autocorrelation, the central forecasting claim would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that $Q(.95)$ values of Paris air temperature and the Steadman Heat Index can be forecast two weeks ahead with promising accuracy by quantile gradient boosting on seven predictors lagged by 14 days; the paper states that the June and July 2019 heat waves can be anticipated by about two weeks. For Cairo, the target quantile is raised to $Q(.99)$ and the results are weaker, with one record-setting May heat wave captured but most days poorly fit. The conformal prediction regions around the Paris forecasts are adaptive, averaging about $\pm 2^\circ$C, and the evaluation on 2020 shows that all ten of the hottest sampled observed temperatures fall within their prediction regions.

Load-bearing premise

The load-bearing premise is that adjacent years are generated by the same underlying weather-generating process, so a model trained on one year's March through August data can be applied to another year; if year-to-year climate shifts change the predictor-response relationship, the entire training and test design collapses.

Editorial extensions

If this is right

  • A two-week-ahead early-warning system for Paris-style heat waves can be built from free, routinely available station data, without relying on simulation output or satellite retrievals.
  • Conformal regions can serve as policy triggers: a region that contains a locally chosen risk threshold (for example, $35^\circ$C) warrants action, while a region entirely below it can be ignored.
  • The forecasting algorithm must be trained locally; the Paris-to-Cairo contrast shows that one trained model will not transfer across climates.
  • Weather-station forecasts of this kind are a forecasting tool, not an explanatory tool: they provide warning skill but little causal understanding of heat wave mechanisms.
  • In desert settings, shortening the 14-day lag may improve accuracy enough to matter, at the cost of reducing advance warning.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implicit extension is that the same recipe could be scaled to the roughly 14,000 active weather stations in the dataset used, producing standardized local heat-warning products wherever station records are long enough.
  • A direct testable extension would compare these two-week-ahead station-data forecasts against satellite-based remote-sensing forecasts on the same days to see where each data source adds skill.
  • The AR(1) whitening step suggests an automated diagnostic: monitor residual autocorrelation each season and widen conformal intervals when whitening fails, so coverage claims remain honest.
  • For Cairo, a lag profile (1, 3, 7, and 14 days) would map the trade-off between lead time and skill in desert climates, where wind shifts can change temperatures within hours.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper uses quantile gradient boosting with 14-day-lagged weather-station predictors to forecast Q(.95) values of 2pm air temperature and the Steadman heat index for Paris and Cairo, and it constructs conformal prediction regions around the forecasts. The central claim is that two Paris heat waves in 2019 could be anticipated about two weeks in advance and that forecasting accuracy is promising; the Cairo analysis is presented as a harder test in which some progress is claimed. The design trains on one year's March–August data, uses an adjacent year as test data for stopping, and then evaluates a 2019-trained model on 2020 data.

Significance. If the claims were supported, the approach would be practically valuable because it uses routine, publicly available weather-station data and a standard machine-learning method with explicit uncertainty quantification. The paper makes a genuine out-of-sample effort by training on one year and testing on another, and it is candid about the key stationarity assumption and about the exploratory nature of the heat-wave matching. Those are real strengths. However, the empirical evidence as presented does not establish either the conformal coverage claim or the heat-wave anticipation claim: the reported coverage check contradicts the stipulated coverage, and the heat-wave matching is retrospective visual alignment without formal forecast verification.

major comments (4)
  1. [§3.1.4 (coverage evaluation)] The paper stipulates a conformal coverage probability of .70 and then reports that only 12 of 100 sampled 2020 days had observed temperatures inside their Q(.95) prediction regions. This 12% empirical coverage flatly contradicts the stipulated guarantee: under .70 coverage the expected number is 70, and a 12/100 rate is not consistent with that claim. The follow-up statement that all of the 10 highest observed temperatures fall inside their regions conditions on the response, which cannot validate an a priori coverage claim. Moreover, if the conformal regions are constructed around the Q(.95) forecast rather than around the observed response, then observed-temperature coverage is not the quantity to which the conformal guarantee applies; the evaluation conflates the two.
  2. [§3.1.4 and Figures 6–7] The central claim that two Paris heat waves 'can be anticipated by about two weeks' rests on visual alignment between two fitted-value spikes and heat-wave reports that were searched for only after the spikes were observed ('After the two large spike were observed, a search was undertaken'). No hit/miss table, lead-time distribution, false-alarm rate, or comparison with a climatological or persistence baseline is provided. Such retrospective event matching cannot establish forecasting skill; a formal event-based evaluation with a pre-specified heat-wave definition and a baseline comparison is required.
  3. [§3.1 and §6] The entire training–test design assumes that adjacent years are generated by the same underlying physical process, an assumption the paper itself labels 'a major concern' in Section 6. The only support offered is the anecdotal statement that the two-year pairs 'performed about the same' and the assurance that results are not cherry-picked. Without a quantitative assessment of distributional stability across year pairs (for example, comparing error distributions or coverage across all listed pairs), the generality of the forecasts is unsubstantiated. This is load-bearing because the 2020 evaluation is performed with a model trained on 2019.
  4. [§4] The Cairo conclusion of 'some progress forecasting record-setting hot days' is based on a single predicted spike in Figure 9 that was matched post hoc to a reported May heat wave, while the text also states that the overall fit 'was not very good' and that the heat-index version substantially underestimates the extreme values. One retrospectively selected spike, with no formal detection or accuracy measures, cannot support a claim of progress, especially in the harder setting that Cairo is meant to illustrate.
minor comments (5)
  1. [§2] The text first says 'March is not included because high temperature values are expected' and then says 'March is included so that 14 day lagged values for the predictors are available'; this is a direct contradiction that should be resolved.
  2. [§3.1.1 and Figure 2] The stopping iteration is given as 1492 in the text, as 1942 in the figure title, and as 1942 again later in the text; the correct value should be stated consistently.
  3. [§3.1.4] The phrase 'the actual coverage probability will be at least .70' is presented as a guarantee, but the paper later reports an empirical coverage of 12/100; the wording should clearly distinguish the conformal guarantee under exchangeability from an empirical check on a particular year.
  4. [§1 and §7] There are typos such as 'We have now arrived our Paris trip’s 2019 destination' and 'The foresting results'; a careful proofreading pass is needed.
  5. [§4] 'Wind direct' should be 'wind direction' in the summary paragraph of the Cairo analysis.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: forecasts are genuine out-of-sample exercises; the post hoc heat-wave matching is an evaluation weakness, not a derivation that reduces to its inputs.

full rationale

The paper's core pipeline is not circular by construction. Q(.95) gradient boosting fits are computed from seven weather-station predictors lagged by 14 days, and forecasts for an adjacent year are produced by applying a model trained on a different year's March-August data. The response is the observed 2pm temperature or Steadman heat index; no heat-wave label is used as an input, and the forecasted quantity is not defined in terms of the heat-wave events it is later compared with. The 2020 evaluation uses 100 randomly sampled days whose temperatures were not used to fit the 2019 model, so the reported forecasts are not in-sample fits renamed as predictions. The conformal regions are built from AR(1)-filtered residuals, not from the observed outcomes used in the coverage check, so the coverage statement is not a fitted parameter. The paper's self-citations (Berk et al., 2024; Berk and Braverman, 2025) provide background and a brief residual-exchangeability remark, but the central forecasting derivation does not rest on an unverified uniqueness theorem or on a self-citation chain. Section 3.1.4's statement, 'After the two large spike were observed, a search was undertaken for information on heat waves,' shows post hoc event matching, which cannot by itself establish forecasting skill, but this is an evaluation-design weakness rather than a circular reduction of the forecast to its own output. Similarly, the 12/100 empirical coverage versus the stipulated .70, and the conditional statement that all 10 highest observed temperatures fall inside their regions, undermine the coverage claim but do not make the conformal construction equivalent to its inputs. Section 6's 'major concern' about year-to-year stability of the underlying physics is an explicitly stated assumption and limitation, not a circular step. Because the derivation chain is self-contained and externally checkable, the circularity score is low; the minor self-citations do not carry the paper's main claim.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a year-to-year stationarity assumption that is acknowledged by the author as a major concern and is untestable with 7 years of data. The conformal coverage claim additionally relies on exchangeability of AR(1)-filtered residuals, which is checked with a white-noise test but not formally established. Several user-set choices, quantile level, lag, coverage level, and hyperparameters, shape the results. No new physical entities are introduced.

free parameters (6)
  • Gradient boosting stopping iteration = 1942 trees (text also cites 1492)
    The number of trees is selected by minimizing Q(.95) loss on the test year (2018), meaning the model complexity is tuned on test data.
  • Quantile level tau = 0.95 (Paris), 0.99 (Cairo)
    Chosen by hand to target rare extremes; Cairo uses a higher quantile after initial attempts, a post hoc modeling choice.
  • Predictor lag = 14 days
    Chosen by hand to provide two-week advance warning; not searched or optimized.
  • Stipulated conformal coverage = 0.70
    User-set coverage level for illustration; the actual empirical coverage on a random 2020 sample is 12/100, contradicting the stipulation.
  • Shrinkage and interaction depth = 0.001 and 4
    Set by hand with the intent of gradual fitting; not optimized or justified by data.
  • AR(1) coefficient for residual whitening = 0.42
    Estimated from training residuals and used to make exchangeable nonconformity scores; it affects the conformal intervals but is not the forecast itself.
assumptions (4)
  • domain assumption Adjacent years are sufficiently similar: a model trained on one year's March through August data can forecast another year's extremes, based on the same underlying physics and covariance stationarity conditional on month.
    Stated in Sections 3.1 and 5; the paper calls it the key forecasting assumption and notes it may fail with global warming or El Nino.
  • domain assumption The AR(1)-filtered residuals are exchangeable, so conformal prediction regions are valid.
    Section 3.1.4; the paper reports the filtered residuals are indistinguishable from white noise, but exchangeability of the scores is assumed.
  • domain assumption 2pm solar time temperature is a suitable proxy for daily high heat exposure.
    Section 2; the paper acknowledges daily maxima can occur at other times but chooses 2pm for longitudinal comparability.
  • standard math The gbm quantile procedure correctly estimates conditional quantiles for these data.
    Relies on Friedman (2001) and Kriegler's quantile loss; this is a standard method, not proven anew here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Forecasting Extreme High Summer Temperatures in Paris and Cairo Using Gradient Boosting and Conformal Prediction Regions." pith.science (2026). https://pith.science/paper/ZUZYQTL5

@misc{pith2026250602349,
  author       = {Pith},
  title        = {Pith review of: Forecasting Extreme High Summer Temperatures in Paris and Cairo Using Gradient Boosting and Conformal Prediction Regions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZUZYQTL5}},
  note         = {Machine review of arXiv:2506.02349}
}
read the original abstract

In this paper, gradient boosting is used to forecast the Q(.95) values of air temperature and the Steadman Heat Index. Paris, France during late the spring and summer months is the major focus. Predictors and responses are drawn from the Paris-Montsouris weather station for the years 2018 through 2024. Q(.95) values are used because of interest in summer heat that is statistically rare and extreme. The data are curated as a multiple time series for each year. Predictors include seven routinely collected indicators of weather conditions. They each are lagged by 14 days such that temperature and heat index forecasts are provided two weeks in advance. Forecasting uncertainty is addressed with conformal prediction regions. Forecasting accuracy is promising. Cairo, Egypt is a second location using data from the weather station at the Cairo Internal Airport over the same years and months. Cairo is a more challenging setting for temperature forecasting because its desert climate can create abrupt and erratic temperature changes. Yet, there is some progress forecasting record-setting hot days.

Figures

Figures reproduced from arXiv: 2506.02349 by the authors.

Figure 1
Figure 1. Histogram of the Paris Daily 2 PM Air temperature. s in Celsius for [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Best Test Data Stopping Iteration for the Paris Data Is 1942 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Fitted Q(.95) values from quantile gradient boosting in red are [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Shown is the percentage contribution to the quantile fit for each [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: A partial dependence plot of the relationship between relative [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: The gradient boosting fit in a time series format. The red dots [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 6
Figure 6. Figure 6: For the higher observed temperatures, the two ideally are very [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: The gradient boosting fit for the Steadman heat index in a time [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: 2pm Cairo Air Temperatures March Through August 2019 with a [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Cairo gradient boosting fit in a time series format. The red dots [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: Cairo gradient boosting fit in a time series format. All of the data [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 22 canonical work pages

  1. [1]

    Transformers in Time- Series Analysis: A Tutorial

    Ahmed, S., Nielsen, I.E., Tripathi, A. et al. (2023) “Transformers in Time- Series Analysis: A Tutorial.” Circuits, Systems, and Signal Processing 42: 7433 –

    work page 2023

  2. [7]

    Avoidable Mortality for Each Temperature Unit Decrease

    Liu, J-M., Ai, S-Q., Jin-Lei Qi, J-L., Wang, L-J., Zhou, M-G., Wang, C-J., Yin, P., Hua-Liang Lin, H-L. (2021) Defining Region-Specific Heatwave in China Based on a Novel Concept of “Avoidable Mortality for Each Temperature Unit Decrease.” Advances in Climate Change Research 12(5): Volume 12, Issue 5: 611 – 618, Mann, M.E., Rahmstorf, S., Kornhuber, K., a...

    work page 2021

  3. [12]

    The Influence of Drought on Heat Wave Intensity, Duration, and Exposure

    Leeper, R. D., Harrington, T., Palecki, M. A., DePolt, K., Scott, E., Runkle, J., and Diamond, H.J. (2025) “The Influence of Drought on Heat Wave Intensity, Duration, and Exposure.” Journal of Applied Meteorology and Climatology 64: 425–438, Li, X., Mann, M.E., Wehnerb, M.F., Rahmstorf, S., Petric, S., Chris- tiansena, S. and Carrilloa, J.” (2024) “Role o...

    work page 2025

  4. [17]

    Importance of Humidity for Characterization and Communi- cation of Dangerous Heatwave Conditions

    Coles, S. (2001) An Introduction to Statistical Modeling of Extreme Values. Springer Cvijanovic, V., Mistry, M.N., Begg, J.D., Gasparrini A., and Rod´ o, X. (2023) “Importance of Humidity for Characterization and Communi- cation of Dangerous Heatwave Conditions.” npj Climate and Atmo- spheric Science 6(33). Friedman, J.H. (2001) “Greedy Function Approxima...

    work page 2001

  5. [32]

    Magnitude of Extreme Heat Waves in Present Climate and Their Projection in a Warming World

    Russo, S., Dosio, A., Graverson, R.G., Sillmann, J., Carrao, H., Dunbar, M.B., Singleton, A., Montanga, P., Barbola, P., and Vogt, J. (2014) “Magnitude of Extreme Heat Waves in Present Climate and Their Projection in a Warming World.” Journal of Geophysical Research Atmospheres 199(22) 12,500 – 12,512. Smith, T.T., Zaitchik, B.F., Gohlke, J.M. (2013) “Hea...

    work page 2014

  6. [92]

    A review on the Scientific Understanding of Heatwaves –Their Measurement, Driving Mechanisms, and Changes at the Global Scale

    Perkins, S.E. (2015) “A review on the Scientific Understanding of Heatwaves –Their Measurement, Driving Mechanisms, and Changes at the Global Scale.” Atmospheric Research165 - 165: 242 –

    work page 2015

  7. [120]

    Probabilistic Weather Forecasting with Machine Learning

    31 Price, I., Sanchez-Gonzalez, A., Alet, F. et al. (2024) “Probabilistic Weather Forecasting with Machine Learning.” Nature 624: 559 –

    work page 2024

  8. [206]

    Loess: A Nonparametric, Graphical Tool For Depict- ing Relationships Between Variables

    Jacoby, W.G. (2000) “Loess: A Nonparametric, Graphical Tool For Depict- ing Relationships Between Variables.” Electoral Studies 19(4): 577 –

    work page 2000

Show all 24 references
  1. [207]

    The Effects of Night-Time Warming on Mortality Burden Under Fu- ture Climate Change Scenarios: A Modelling Study

    He, C., Kim, H., Hashizume, M., Lee, W., Honda, Y., and Kim, S.E. (2022) “The Effects of Night-Time Warming on Mortality Burden Under Fu- ture Climate Change Scenarios: A Modelling Study.” The Lancet Plan- etary Health 6(8), 648 –

    2022

  2. [267]

    On the Measurement of Heat Waves

    Perkins, S.E., and Alexander, L.V., (2013) “On the Measurement of Heat Waves.” Journal of Climate26: 4500 –

    2013

  3. [378]

    Adaptive Conformal Inference Under Distribution Shift

    Gettleman, A., and Rood, R.B. (2016) Demystifying Climate Models: A users Guide to Earth System ModelsSpringer. Gibbs, I., and Cand` es, E. (2021) “Adaptive Conformal Inference Under Distribution Shift.” in M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang and J. Wortman Vaug...

    2016

  4. [563]

    Generalized Boosted Models: A Guide to The gbm Package

    Ridgeway. G., (2024) “Generalized Boosted Models: A Guide to The gbm Package.” https://cran.r-project.org/web/packages/gbm/vignettes/gbm.pdf Romano, Y., Patterson, E., and Cand` es, E.J. (2019) “Conformal Quantile Regression.” In H. Wallach et al., (eds)Advances in Neural Info...

    2024

  5. [613]

    Deep Learning-Based Extreme Heatwave Forecast

    Jaque-Dumas, V., Ragine, F., Borgnat, P., Arbry, P., and Bouchet, F. (2022) “Deep Learning-Based Extreme Heatwave Forecast” Frontiers in Climate 4 -

    2022

  6. [657]

    Unstable Climates: Exploring the Statistical and Social Constructions of ‘nor- mal’ Climate

    Hyndman, R.J. and Athanasopoulos, G. (2021) Forecasting: Principles and Practice, third edition at http/OTexts.c0m/fpp3/. 29 Hopke, J. E. (2019). Connecting Extreme Heat Events to Climate Change: Media Coverage of Heat Waves and Wildfires. Environmental Commu- nication 14(4), ...

    2021

  7. [667]

    (2004) Desert MeteorologyCambridge University Press

    Warner, T.T. (2004) Desert MeteorologyCambridge University Press. 32

    2004

  8. [800]

    How Unexpected Was the 2021 Pacific Northwest Heatwave?

    McKinnon, K. and Simpson, I.R. (2022) “How Unexpected Was the 2021 Pacific Northwest Heatwave?” Geophysical Research Letters 49(18) (e2022GL100380). Oliveira, R.I., Ornstein, P., Ramos, T., and Romano, J.V. (2024) “Split Conformal Prediction and Non-Exchangeable Data.” Journal...

    2022

  9. [825]

    A Universal Scale of Apparent Temperature

    Steadman, R.G., (1984( “A Universal Scale of Apparent Temperature.” Journal of Applied Meteorology and Climatology23(12): 1674 –

    1984

  10. [1222]

    Stochastic Gradient Boosting

    Friedman J.H. (2002) “Stochastic Gradient Boosting.”Computational Statis- tics & Data Analysis38 (4): 367 –

    2002

  11. [1672]

    101 Degrees in May? Even for Texas, This Is Hot,

    Graff, A., (2025) “101 Degrees in May? Even for Texas, This Is Hot,” New York timesMay 4th, https://www.nytimes.com/2025/05/14/weather/ texas-heat-record-may.html. Granger, C.W.J. (1969) “Prediction with a Generalized Cost of Error Func- tion. Operational Research Quarterly20: 199 –

    2025

  12. [1687]

    Gradient Boost- ing for Extreme Quantile Regression

    Stull, R., (2017) Practical Meteorology: An Algebra-based Survey of Atmo- spheric Science - version 1.02b. University of British Columbia. Velthoen, J., Dombry, C., Cai, JJ., and Engelke, S. (2023) “Gradient Boost- ing for Extreme Quantile Regression.” Extremes 26: 639 –

    2017

  13. [2022]

    Heatwaves: Does Global Research Reflect the Growing Threat in the Light of ClimateChange?

    Klingh¨ ofer, D. Braun, M., Br¨ uggmann, D., and Groneberg, D.A. (2023) “Heatwaves: Does Global Research Reflect the Growing Threat in the Light of ClimateChange?” Globalization and Health(19)56: 1 –

    2023

  14. [4517]

    Quasiresonant Amplification of Planetary waves and Recent Northern Hemisphere Weather Extremes

    Petoukhov, V., Rahmstorf, S., Petri, S., and Schellnhuber, H.J. (2013) “Quasiresonant Amplification of Planetary waves and Recent Northern Hemisphere Weather Extremes.” Proceedings of the National Academy of Sciences 110: 5336 –

    2013

  15. [5341]

    A Review of Recent Studies on Heat wave Definitions, Mechanisms, Changes, and Impact on Mortality

    Pitcar, A., Cheval, S., and Frighenciu, M. (2019) “A Review of Recent Studies on Heat wave Definitions, Mechanisms, Changes, and Impact on Mortality.” Forum GeographicXVII (2): 103 –

    2019

  16. [7466]

    Reconciling Modern Machine- Learning Practice and the Classical Bias–Variance Trade-Off

    Belkin, M., Hsu, D.H., and Mandal, S. (2019) “Reconciling Modern Machine- Learning Practice and the Classical Bias–Variance Trade-Off.” Pro- ceedings of the National Academy of Sciences116(32): 15849 – 15854. Angelopoulos, A.N., Barber, R.F., and Bates, S. (2025) “Theoretical ...

    2019 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.