pith. sign in

arxiv: 2510.02050 · v3 · submitted 2025-10-02 · 📊 stat.AP · cs.LG

Multidata Causal Discovery for Statistical Hurricane Intensity Forecasting

Saranya Ganesh S , Frederick Iat-Hin Tam , Milton S. Gomez , Marie McGraw , Mark DeMaria , Kate Musgrave , Jakob Runge , Tom Beucler This is my paper

Pith reviewed 2026-05-18 10:46 UTC · model grok-4.3

classification 📊 stat.AP cs.LG
keywords tropical cyclone intensitycausal discoverystatistical forecastingfeature selectionSHIPS modelERA5 reanalysishurricane prediction
0
0 comments X

The pith

Causal feature selection from multidata discovery outperforms correlation-based methods for tropical cyclone intensity forecasts.

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

This paper tries to establish that identifying predictors through causal discovery rather than correlation or random forest importance produces more accurate linear regression models for tropical cyclone intensity at various lead times. A sympathetic reader would care because better intensity forecasts can improve preparation and reduce harm from hurricanes. The authors apply a multidata causal discovery framework to a replicated SHIPS dataset based on ERA5 reanalysis to select variables causally linked to intensity changes. They compare the resulting models against baselines across 1- to 5-day lead times and show consistent gains with causal selection, especially under three days. Adding the selected features creates an enhanced SHIPS+ set that raises short-term skill, with operational tests confirming gains from three of the six new predictors.

Core claim

The central claim is that causal feature selection using a multidata causal discovery framework on SHIPS and ERA5 data consistently outperforms correlation, random forest feature importance, and no selection on unseen test cases for TC intensity prediction, especially at lead times shorter than 3 days. Top causal features are vertical shear, mid-tropospheric potential vorticity, and surface moisture conditions. An extended SHIPS+ predictor set improves predictive skill at 24, 48, and 72 hours. Operational SHIPS tests confirm that three of the six added causally discovered predictors improve forecast skill, while the framework remains regional and does not require global forecast data.

What carries the argument

Multidata causal discovery framework that recovers predictors causally linked to TC intensity changes from replicated SHIPS and ERA5 data.

If this is right

  • Causal selection yields higher skill on unseen cases especially for lead times under 3 days.
  • SHIPS+ with the added causal predictors increases skill at 24, 48, and 72 hours.
  • Three of the six added predictors improve operational SHIPS forecast skill.
  • Adding nonlinearity via multilayer perceptron extends skill gains to longer lead times.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same causal selection process could be tested on other extreme weather variables where confounding variables are common.
  • Operational centers might periodically rerun causal discovery on updated reanalysis to refresh predictor lists.
  • Combining the causal features with nonlinear models may produce further accuracy gains beyond the linear regressions shown.

Load-bearing premise

The multidata causal discovery framework accurately recovers true causal links between the chosen meteorological variables and TC intensity changes without substantial bias from unobserved confounders or violations of its assumptions.

What would settle it

Independent verification on new tropical cyclone data where correlation-based or random forest selection matches or exceeds causal selection in short-lead-time forecast skill would falsify the claimed advantage.

Figures

Figures reproduced from arXiv: 2510.02050 by Frederick Iat-Hin Tam, Jakob Runge, Kate Musgrave, Marie McGraw, Mark DeMaria, Milton S. Gomez, Saranya Ganesh S, Tom Beucler.

Figure 1
Figure 1. Figure 1: Multidata causal feature selection methodology. Step 1: Preprocessed spatiotemporal fields for all TC cases [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example results for the 24-hour intensity change forecast (DELV24) from Fold 3 using the SHIPS+ERA5 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Summary of results for the 24-hour intensity change forecast (DELV24) using SHIPS+ERA5 predictors [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of test 𝑅 2 values across forecast lead times (24–120 hr) for the original SHIPS predictors (blue/green boxes) and the expanded SHIPS+ predictors (orange/yellow boxes), using no feature selection. Both MLR and MLP runs are shown to illustrate the added value of nonlinear modeling. Dashed brown lines indicate the median, and solid black lines mark the mean. Overall, the MLP consistently outperfor… view at source ↗
Figure 5
Figure 5. Figure 5: Predictor importance and dependencies for models trained on SHIPS+. (a–b) Global feature importance [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hurricane Larry (2021): (a) Best track from IBTrACS showing the time used for testing (b) performance [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The improvement in the SHIPS intensity forecasts with the three new predictors relative to baseline for [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
read the original abstract

Improving statistical forecasts of tropical cyclone (TC) intensity is limited by complex nonlinear interactions and difficulty in identifying relevant predictors. Conventional methods prioritize correlation or fit, often overlooking confounding variables and limiting generalizability to unseen TCs. To address this, we leverage a multidata causal discovery framework with a replicated dataset based on Statistical Hurricane Intensity Prediction Scheme (SHIPS) using ERA5 meteorological reanalysis. We conduct experiments to identify and select predictors causally linked to TC intensity changes. We then train multiple linear regression models to compare causal feature selection with correlation, random forest feature importance, and no feature selection, across five forecast lead times from 1 to 5 days (24 to 120 hours). Causal feature selection consistently outperforms on unseen test cases, especially for lead times shorter than 3 days. Top causal features include vertical shear, mid-tropospheric potential vorticity and surface moisture conditions, which are physically significant yet often underutilized in TC intensity predictions. We build an extended predictor set (SHIPS+) by adding selected features to the standard SHIPS predictors. SHIPS+ yields increased short-term predictive skill at lead times of 24, 48, and 72 hours. Adding nonlinearity using a multilayer perceptron further extends skill to longer lead times, despite our framework being purely regional and not requiring global forecast data. Operational SHIPS tests confirm that three of the six added causally discovered predictors improve forecast skill, with the largest gains at longer lead times. Our results demonstrate that causal discovery improves TC intensity prediction and pave the way toward more empirical forecasts.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces a multidata causal discovery framework applied to a replicated SHIPS dataset from ERA5 reanalysis to select predictors for tropical cyclone (TC) intensity forecasting. It compares causal feature selection against correlation-based selection, random forest importance, and no selection in linear regression models for lead times of 24 to 120 hours. The authors report that causal selection consistently outperforms on unseen test cases, particularly at shorter lead times, and construct an extended SHIPS+ model that shows improved predictive skill at 24, 48, and 72 hours, with some predictors confirmed in operational tests. Nonlinear extensions with multilayer perceptrons are also explored.

Significance. Should the causal discovery reliably identify true causal relationships without substantial bias from unobserved confounders, this approach could meaningfully improve statistical TC intensity prediction by focusing on physically relevant features that generalize better. The use of held-out testing and partial operational validation strengthens the empirical claims. The work highlights underutilized predictors like vertical shear and mid-tropospheric potential vorticity.

major comments (2)
  1. [§3 (Multidata Causal Discovery)] The central claim depends on the multidata causal discovery accurately recovering causal links between meteorological variables and TC intensity changes. However, the manuscript provides no explicit validation or sensitivity analysis addressing potential violations of assumptions such as causal sufficiency or faithfulness in the presence of unobserved confounders common in reanalysis data for TCs. This is load-bearing because the superiority is attributed to the causal nature of the selection.
  2. [§5 (Experimental Results)] The reported consistent outperformance on held-out tests lacks accompanying error bars, confidence intervals, or statistical significance tests on the skill scores. Additionally, there is no discussion of multiple-testing correction given the multiple lead times and comparison methods, which weakens the strength of the performance claims.
minor comments (2)
  1. [Notation and Methods] The description of how the replicated dataset is constructed from SHIPS and ERA5 could include more details on variable selection criteria and preprocessing steps to improve reproducibility.
  2. [Figures] Figure captions for performance plots should explicitly state the metrics (e.g., RMSE or correlation) and the number of test cases used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments, which have helped us identify areas to strengthen the manuscript. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§3 (Multidata Causal Discovery)] The central claim depends on the multidata causal discovery accurately recovering causal links between meteorological variables and TC intensity changes. However, the manuscript provides no explicit validation or sensitivity analysis addressing potential violations of assumptions such as causal sufficiency or faithfulness in the presence of unobserved confounders common in reanalysis data for TCs. This is load-bearing because the superiority is attributed to the causal nature of the selection.

    Authors: We agree that explicit discussion of the causal discovery assumptions would strengthen the paper. Although the multidata framework leverages multiple reanalysis sources to improve robustness against some violations of causal sufficiency, unobserved confounders remain a valid concern in TC data. In the revision we will add a new subsection in §3 that explicitly states the assumptions (sufficiency, faithfulness, and no selection bias), discusses their plausibility for ERA5-based TC data, and reports a sensitivity analysis in which we vary the significance threshold, subsample the datasets, and examine the stability of the recovered causal parents. This will clarify that our performance gains are attributed to improved generalization from the selected features rather than a claim of recovering the complete true causal graph. revision: yes

  2. Referee: [§5 (Experimental Results)] The reported consistent outperformance on held-out tests lacks accompanying error bars, confidence intervals, or statistical significance tests on the skill scores. Additionally, there is no discussion of multiple-testing correction given the multiple lead times and comparison methods, which weakens the strength of the performance claims.

    Authors: We thank the referee for this observation. We will revise the experimental results section and associated figures to include error bars (standard deviation across 10 random train/test splits) and 95% confidence intervals for all skill-score metrics. We will also add paired statistical significance tests (Wilcoxon signed-rank) between causal selection and each baseline for every lead time, together with a Bonferroni correction for the family of 20 comparisons (5 lead times × 4 methods). The corrected p-values and a brief discussion of multiple-testing implications will be reported in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results rely on external reanalysis data and standard methods with unseen test evaluation

full rationale

The paper's central claims rest on applying multidata causal discovery to ERA5-based SHIPS data, selecting features, and evaluating linear regression (and MLP) performance on held-out test cases across lead times. Comparisons to correlation and random forest baselines are performed on the same split data. No equations or steps reduce a reported prediction or skill score to a quantity fitted on the identical evaluation set by construction. No load-bearing self-citation chain or uniqueness theorem imported from the authors' prior work is evident in the provided text; the framework is presented as leveraging existing causal discovery algorithms. This is the expected honest outcome for a paper whose performance metrics are externally benchmarked on unseen tropical cyclone cases.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; the approach rests on standard causal discovery assumptions and the representativeness of the ERA5-derived dataset for real TC events.

axioms (1)
  • domain assumption No unobserved confounders affect the relationships between selected meteorological variables and TC intensity changes.
    Implicit in any causal discovery applied to observational reanalysis data for atmospheric processes.

pith-pipeline@v0.9.0 · 5837 in / 1196 out tokens · 31501 ms · 2026-05-18T10:46:57.625244+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages · 1 internal anchor

  1. [1]

    The formation of tropical cyclones.Meteorology and atmospheric physics, 67(1):37–69, 1998

    William M Gray. The formation of tropical cyclones.Meteorology and atmospheric physics, 67(1):37–69, 1998

    work page 1998

  2. [2]

    Recent advancements in dynamical tropical cyclone track predictions.Meteorology and Atmospheric Physics, 56(1):81–99, 1995

    RL Elsberry. Recent advancements in dynamical tropical cyclone track predictions.Meteorology and Atmospheric Physics, 56(1):81–99, 1995

    work page 1995

  3. [3]

    Advances in research and forecasting of tropical cyclones from 1963–2013.Asia-Pacific Journal of Atmospheric Sciences, 50:3–16, 2014

    Russell L Elsberry. Advances in research and forecasting of tropical cyclones from 1963–2013.Asia-Pacific Journal of Atmospheric Sciences, 50:3–16, 2014

    work page 1963

  4. [4]

    Normalized hurricane damages in the united states: 1925–95

    Roger A Pielke Jr and Christopher W Landsea. Normalized hurricane damages in the united states: 1925–95. Weather and forecasting, 13(3):621–631, 1998

    work page 1925

  5. [5]

    Crossett, Thomas J

    Kristen M. Crossett, Thomas J. Culliton, Peter C. Wiley, and Timothy R. Goodspeed. Population trends along the coastal united states, 1980–2008. Technical report, United States National Ocean Service, Special Projects, 2004

    work page 1980

  6. [6]

    Tropical cyclone prediction on subseasonal time-scales.Tropical Cyclone Research and Review, 8(3):150–165, 2019

    Suzana J Camargo, Joanne Camp, Russell L Elsberry, Paul A Gregory, Philip J Klotzbach, Carl J Schreck III, Adam H Sobel, Michael J Ventrice, Frédéric Vitart, Zhuo Wang, et al. Tropical cyclone prediction on subseasonal time-scales.Tropical Cyclone Research and Review, 8(3):150–165, 2019

    work page 2019

  7. [7]

    Review of recent progress in tropical cyclone track forecasting and expression of uncertainties.Tropical Cyclone Research and Review, 8(4):181–218, 2019

    Julian T Heming, Fernando Prates, Morris A Bender, Rebecca Bowyer, John Cangialosi, Phillippe Caroff, Thomas Coleman, James D Doyle, Anumeha Dube, Ghislain Faure, et al. Review of recent progress in tropical cyclone track forecasting and expression of uncertainties.Tropical Cyclone Research and Review, 8(4):181–218, 2019

    work page 2019

  8. [8]

    Recent advances in research on tropical cyclogenesis.Tropical Cyclone Research and Review, 9(2):87–105, 2020

    Brian H Tang, Juan Fang, Alicia Bentley, Gerard Kilroy, Masuo Nakano, Myung-Sook Park, VPM Rajasree, Zhuo Wang, Allison A Wing, and Liguang Wu. Recent advances in research on tropical cyclogenesis.Tropical Cyclone Research and Review, 9(2):87–105, 2020

    work page 2020

  9. [9]

    An updated statistical hurricane intensity prediction scheme (ships) for the atlantic and eastern north pacific basins.Weather and Forecasting, 14(3):326–337, 1999

    Mark DeMaria and John Kaplan. An updated statistical hurricane intensity prediction scheme (ships) for the atlantic and eastern north pacific basins.Weather and Forecasting, 14(3):326–337, 1999

    work page 1999

  10. [10]

    Current understanding of tropical cyclone structure and intensity changes–a review

    Yu-qing Wang and C-C Wu. Current understanding of tropical cyclone structure and intensity changes–a review. Meteorology and Atmospheric Physics, 87(4):257–278, 2004

    work page 2004

  11. [11]

    Operational forecasting of tropical cyclone rapid intensification at the national hurricane center.Atmosphere, 12(6):683, 2021

    Mark DeMaria, James L Franklin, Matthew J Onderlinde, and John Kaplan. Operational forecasting of tropical cyclone rapid intensification at the national hurricane center.Atmosphere, 12(6):683, 2021. 16 Causal Discovery to improve SHIPS

    work page 2021

  12. [12]

    On the predictability and error sources of tropical cyclone intensity forecasts

    Kerry Emanuel and Fuqing Zhang. On the predictability and error sources of tropical cyclone intensity forecasts. Journal of the Atmospheric Sciences, 73(9):3739–3747, 2016

    work page 2016

  13. [13]

    The national hurricane center tropical cyclone model guidance suite.Weather and Forecasting, 37(11):2141–2159, 2022

    Mark DeMaria, James L Franklin, Rachel Zelinsky, David A Zelinsky, Matthew J Onderlinde, John A Knaff, Stephanie N Stevenson, John Kaplan, Kate D Musgrave, Galina Chirokova, et al. The national hurricane center tropical cyclone model guidance suite.Weather and Forecasting, 37(11):2141–2159, 2022

    work page 2022

  14. [14]

    Recent research progress on tropical cyclone structure and intensity.Tropical Cyclone Research and Review, 1(2):254–275, 2012

    Yuqing Wang. Recent research progress on tropical cyclone structure and intensity.Tropical Cyclone Research and Review, 1(2):254–275, 2012

    work page 2012

  15. [15]

    Validation of ensemble-based probabilistic tropical cyclone intensity change

    Ryan D Torn and Mark DeMaria. Validation of ensemble-based probabilistic tropical cyclone intensity change. Atmosphere, 12(3):373, 2021

    work page 2021

  16. [16]

    A statistical hurricane intensity prediction scheme (ships) for the atlantic basin

    Mark DeMaria and John Kaplan. A statistical hurricane intensity prediction scheme (ships) for the atlantic basin. Weather and Forecasting, 9(2):209–220, 1994

    work page 1994

  17. [17]

    F. D. Marks and M. DeMaria. Development of a tropical cyclone rainfall climatology and persistence (r-cliper) model. Technical report, NOAA/OAR/AOML/Hurricane Research Division, 2003

    work page 2003

  18. [18]

    Recent progress in tropical cyclone intensity forecasting at the national hurricane center.Weather and Forecasting, 35(5):1913–1922, 2020

    John P Cangialosi, Eric Blake, Mark DeMaria, Andrew Penny, Andrew Latto, Edward Rappaport, and Vijay Tallapragada. Recent progress in tropical cyclone intensity forecasting at the national hurricane center.Weather and Forecasting, 35(5):1913–1922, 2020

    work page 1913

  19. [19]

    Further improvements to the statistical hurricane intensity prediction scheme (ships).Weather and Forecasting, 20(4):531–543, 2005

    Mark DeMaria, Michelle Mainelli, Lynn K Shay, John A Knaff, and John Kaplan. Further improvements to the statistical hurricane intensity prediction scheme (ships).Weather and Forecasting, 20(4):531–543, 2005

    work page 2005

  20. [20]

    Causal network reconstruction from time series: From theoretical assumptions to practical estimation

    Jakob Runge. Causal network reconstruction from time series: From theoretical assumptions to practical estimation. Chaos: An Interdisciplinary Journal of Nonlinear Science, 28(7):075310, 2018

    work page 2018

  21. [21]

    Inferring causation from time series in earth system sciences.Nature communications, 10(1):1–13, 2019

    Jakob Runge, Sebastian Bathiany, Erik Bollt, Gustau Camps-Valls, Dim Coumou, Ethan Deyle, Clark Glymour, Marlene Kretschmer, Miguel D Mahecha, Jordi Muñoz-Marí, et al. Inferring causation from time series in earth system sciences.Nature communications, 10(1):1–13, 2019

    work page 2019

  22. [22]

    Detecting and quantifying causal associations in large nonlinear time series datasets.Science advances, 5(11):eaau4996, 2019

    Jakob Runge, Peer Nowack, Marlene Kretschmer, Seth Flaxman, and Dino Sejdinovic. Detecting and quantifying causal associations in large nonlinear time series datasets.Science advances, 5(11):eaau4996, 2019

    work page 2019

  23. [23]

    Causal discovery in ensembles of climate time series

    Andreas Gerhardus and Jakob Runge. Causal discovery in ensembles of climate time series. Technical report, Copernicus Meetings, 2022

    work page 2022

  24. [24]

    Causal inference for time series.Nature Reviews Earth & Environment, 4(7):487–505, 2023

    Jakob Runge, Andreas Gerhardus, Gherardo Varando, Veronika Eyring, and Gustau Camps-Valls. Causal inference for time series.Nature Reviews Earth & Environment, 4(7):487–505, 2023

    work page 2023

  25. [25]

    Causal discovery for climate research using graphical models.Journal of Climate, 25(17):5648–5665, 2012

    Imme Ebert-Uphoff and Yi Deng. Causal discovery for climate research using graphical models.Journal of Climate, 25(17):5648–5665, 2012

    work page 2012

  26. [26]

    Jakob Runge, Vladimir Petoukhov, and Jürgen Kurths. Quantifying the strength and delay of climatic interactions: The ambiguities of cross correlation and a novel measure based on graphical models.Journal of climate, 27(2):720–739, 2014

    work page 2014

  27. [27]

    A causal intercomparison framework unravels precipitation drivers in global storm-resolving models.npj Climate and Atmospheric Science, 8(1):245, 2025

    Lucile Ricard, Tom Beucler, Claudia Christine Stephan, and Athanasios Nenes. A causal intercomparison framework unravels precipitation drivers in global storm-resolving models.npj Climate and Atmospheric Science, 8(1):245, 2025

    work page 2025

  28. [28]

    Memory matters: A case for granger causality in climate variability studies.Journal of climate, 31(8):3289–3300, 2018

    Marie C McGraw and Elizabeth A Barnes. Memory matters: A case for granger causality in climate variability studies.Journal of climate, 31(8):3289–3300, 2018

    work page 2018

  29. [29]

    Jakob Runge, Vladimir Petoukhov, and Jürgen Kurths. Quantifying the strength and delay of climatic interactions: The ambiguities of cross correlation and a novel measure based on graphical models.Journal of Climate, 27(2):720 – 739, 2014

    work page 2014

  30. [30]

    The impact of arctic warming on the midlatitude jet-stream: Can it? has it? will it?Wiley Interdisciplinary Reviews: Climate Change, 6(3):277–286, 2015

    Elizabeth A Barnes and James A Screen. The impact of arctic warming on the midlatitude jet-stream: Can it? has it? will it?Wiley Interdisciplinary Reviews: Climate Change, 6(3):277–286, 2015

    work page 2015

  31. [31]

    Variable selection for tropical cyclogenesis predictive modeling.Monthly Weather Review, 144(12):4605–4619, 2016

    Jasper S Wijnands, Guoqi Qian, and Yuriy Kuleshov. Variable selection for tropical cyclogenesis predictive modeling.Monthly Weather Review, 144(12):4605–4619, 2016

    work page 2016

  32. [32]

    Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasets

    Jakob Runge. Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasets. In Jonas Peters and David Sontag, editors,Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI), volume 124 ofProceedings of Machine Learning Research, pages 1388–1397. PMLR, 03–06 Aug 2020

    work page 2020

  33. [33]

    Latos, I.-J

    B. Latos, I.-J. Moon, and D. H. Kim. Advancing seasonal hurricane predictions using causal ai. InEGU General Assembly 2024, Vienna, Austria, Apr 2024. 17 Causal Discovery to improve SHIPS

    work page 2024

  34. [34]

    Era5 hourly data on pressure levels from 1979 to present, copernicus climate change service (c3s) climate data store (cds), 2018

    H Hersbach, B Bell, P Berrisford, G Biavati, A Horányi, J Muñoz Sabater, J Nicolas, C Peubey, R Radu, I Rozum, et al. Era5 hourly data on pressure levels from 1979 to present, copernicus climate change service (c3s) climate data store (cds), 2018

    work page 1979

  35. [35]

    The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, et al. The era5 global reanalysis.Quarterly Journal of the Royal Meteorological Society, 146(730):1999–2049, 2020

    work page 1999

  36. [36]

    The international best track archive for climate stewardship (ibtracs) unifying tropical cyclone data.Bulletin of the American Meteorological Society, 91(3):363–376, 2010

    Kenneth R Knapp, Michael C Kruk, David H Levinson, Howard J Diamond, and Charles J Neumann. The international best track archive for climate stewardship (ibtracs) unifying tropical cyclone data.Bulletin of the American Meteorological Society, 91(3):363–376, 2010

    work page 2010

  37. [37]

    An evaluation of dvorak technique–based tropical cyclone intensity estimates.Weather and Forecasting, 25(5):1362–1379, 2010

    John A Knaff, Daniel P Brown, Joe Courtney, Gregory M Gallina, and John L Beven. An evaluation of dvorak technique–based tropical cyclone intensity estimates.Weather and Forecasting, 25(5):1362–1379, 2010

    work page 2010

  38. [38]

    Tropical cyclone precipitation, infrared, microwave, and environmental dataset (tc primed).Bulletin of the American Meteorological Society, 104(11):E1980–E1998, 2023

    Muhammad Naufal Razin et al. Tropical cyclone precipitation, infrared, microwave, and environmental dataset (tc primed).Bulletin of the American Meteorological Society, 104(11):E1980–E1998, 2023

    work page 2023

  39. [39]

    S Ganesh, Tom Beucler, Frederick Iat-Hin Tam, Milton S Gomez, Jakob Runge, and Andreas Gerhardus

    S. S Ganesh, Tom Beucler, Frederick Iat-Hin Tam, Milton S Gomez, Jakob Runge, and Andreas Gerhardus. Selecting robust features for machine learning applications using multidata causal discovery.arXiv e-prints, pages arXiv–2304, 2023

    work page 2023

  40. [40]

    Using causal effect networks to analyze different arctic drivers of midlatitude winter circulation.Journal of climate, 29(11):4069–4081, 2016

    Marlene Kretschmer, Dim Coumou, Jonathan F Donges, and Jakob Runge. Using causal effect networks to analyze different arctic drivers of midlatitude winter circulation.Journal of climate, 29(11):4069–4081, 2016

    work page 2016

  41. [41]

    The different stratospheric influence on cold-extremes in Eurasia and North America.npj Climate and Atmospheric Science, 1(1), nov 22 2018

    Marlene Kretschmer, Judah Cohen, Vivien Matthias, Jakob Runge, and Dim Coumou. The different stratospheric influence on cold-extremes in Eurasia and North America.npj Climate and Atmospheric Science, 1(1), nov 22 2018

    work page 2018

  42. [42]

    Samarasinghe, C

    S.M. Samarasinghe, C. Connolly, E.A. Barnes, I. Ebert-Uphoff, and L. Sun. Strengthened causal connections between the MJO and the North Atlantic with climate warming.Geophys. Res. Lett., 48:e2020GL091168, 2021

    work page 2021

  43. [43]

    Causally-informed deep learning to improve climate models and projections.Journal of Geophysical Research: Atmospheres, 129(4):e2023JD039202, 2024

    Fernando Iglesias-Suarez, Pierre Gentine, Breixo Solino-Fernandez, Tom Beucler, Michael Pritchard, Jakob Runge, and Veronika Eyring. Causally-informed deep learning to improve climate models and projections.Journal of Geophysical Research: Atmospheres, 129(4):e2023JD039202, 2024

    work page 2024

  44. [44]

    Optimal model-free prediction from multivariate time series

    Jakob Runge, Reik V Donner, and Jürgen Kurths. Optimal model-free prediction from multivariate time series. Physical Review E, 91(5):052909, 2015

    work page 2015

  45. [45]

    Data generating process to evaluate causal discovery techniques for time series data.arXiv preprint arXiv:2104.08043, 2021

    Andrew R Lawrence, Marcus Kaiser, Rui Sampaio, and Maksim Sipos. Data generating process to evaluate causal discovery techniques for time series data.arXiv preprint arXiv:2104.08043, 2021

    work page arXiv 2021

  46. [46]

    Adam: A Method for Stochastic Optimization

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  47. [47]

    Cross-validation strategy impacts the performance and interpretation of machine learning models.Artificial Intelligence for the Earth Systems, 2(4):e230026, 2023

    Lily-belle Sweet, Christoph Müller, Mohit Anand, and Jakob Zscheischler. Cross-validation strategy impacts the performance and interpretation of machine learning models.Artificial Intelligence for the Earth Systems, 2(4):e230026, 2023

    work page 2023

  48. [48]

    A statistical analysis of the effects of vertical wind shear on tropical cyclone intensity change over the western north pacific.Monthly Weather Review, 143(9):3434–3453, 2015

    Yuqing Wang, Yunjie Rao, Zhe-Min Tan, and Daria Schönemann. A statistical analysis of the effects of vertical wind shear on tropical cyclone intensity change over the western north pacific.Monthly Weather Review, 143(9):3434–3453, 2015

    work page 2015

  49. [49]

    Hao Fu, Yuqing Wang, Michael Riemer, and Qingqing Li. Effect of unidirectional vertical wind shear on tropical cyclone intensity change—lower-layer shear versus upper-layer shear.Journal of Geophysical Research: Atmospheres, 124(12):6265–6282, 2019

    work page 2019

  50. [50]

    Relationship of environmental relative humidity with north atlantic tropical cyclone intensity and intensification rate.Geophysical research letters, 39(20), 2012

    Longtao Wu, Hui Su, Robert G Fovell, Bin Wang, Janice T Shen, Brian H Kahn, Svetla M Hristova-Veleva, Bjorn H Lambrigtsen, Eric J Fetzer, and Jonathan H Jiang. Relationship of environmental relative humidity with north atlantic tropical cyclone intensity and intensification rate.Geophysical research letters, 39(20), 2012

    work page 2012

  51. [51]

    Michael S Fischer, Paul D Reasor, Brian H Tang, Kristen L Corbosiero, Ryan D Torn, and Xiaomin Chen. A tale of two vortex evolutions: Using a high-resolution ensemble to assess the impacts of ventilation on a tropical cyclone rapid intensification event.Monthly weather review, 151(1):297–320, 2023

    work page 2023

  52. [52]

    Joshua B Wadler, David S Nolan, Jun A Zhang, and Lynn K Shay. Thermodynamic characteristics of downdrafts in tropical cyclones as seen in idealized simulations of different intensities.Journal of the Atmospheric Sciences, 78(11):3503–3524, 2021

    work page 2021

  53. [53]

    Boundary layer recovery and precipitation symmetrization preceding rapid intensification of tropical cyclones under shear

    Xiaomin Chen, Jian-Feng Gu, Jun A Zhang, Frank D Marks, Robert F Rogers, and Joseph J Cione. Boundary layer recovery and precipitation symmetrization preceding rapid intensification of tropical cyclones under shear. Journal of the Atmospheric Sciences, 78(5):1523–1544, 2021. 18 Causal Discovery to improve SHIPS

    work page 2021

  54. [54]

    The genesis of hurricane guillermo: Texmex analyses and a modeling study

    Marja Bister and Kerry A Emanuel. The genesis of hurricane guillermo: Texmex analyses and a modeling study. Monthly weather review, 125(10):2662–2682, 1997

    work page 1997

  55. [55]

    Interactive radiation accelerates the intensification of the midlevel vortex for tropical cyclogenesis.Journal of the Atmospheric Sciences, 77(12):4051–4065, 2020

    Bolei Yang and Zhe-Min Tan. Interactive radiation accelerates the intensification of the midlevel vortex for tropical cyclogenesis.Journal of the Atmospheric Sciences, 77(12):4051–4065, 2020

    work page 2020

  56. [56]

    A unified approach to interpreting model predictions

    Scott M Lundberg and Su-In Lee. A unified approach to interpreting model predictions. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017

    work page 2017

  57. [57]

    Giorgetta, and Veronika Eyring

    Arthur Grundner, Tom Beucler, Pierre Gentine, Fernando Iglesias-Suarez, Marco A. Giorgetta, and Veronika Eyring. Deep learning based cloud cover parameterization for icon.Journal of Advances in Modeling Earth Systems, 14(12):e2021MS002959, 2022. e2021MS002959 2021MS002959

    work page 2022

  58. [58]

    Cangialosi, Brad J

    John P. Cangialosi, Brad J. Reinhart, and Jonathan Martinez. National hurricane center forecast verification report: 2023 hurricane season. Technical Report Verification Report 2023, National Hurricane Center, NOAA, Miami, FL, June 2024. Published June 3, 2024; accessed August 30, 2025. 19 SUPPLEMENTARYMATERIAL: MULTIDATACAUSALDISCOVERY FORSTATISTICALHURR...

    work page 2023