Pith. sign in

REVIEW 3 major objections 5 minor 104 references

Global neutron monitors, read by neural networks, can serve as a real-time spectrometer for galactic cosmic-ray protons and helium.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 08:44 UTC pith:AC4CRPWZ

load-bearing objection Solid, honest NN-vs-force-field comparison; the real soft spot is the extrapolation into the 23/24 minimum, not the ML methodology. the 3 major comments →

arxiv 2607.21009 v1 pith:AC4CRPWZ submitted 2026-07-23 astro-ph.IM

GCR Spectra Reconstructed with Neutron Monitor Yield Function and Artificial Neural Networks: Comparison of Two Methods

Stepan Siruk , Vladislav Alekseev , Victor Kuzminov , Rustam Yulbarisov , Andrey Mayorov This is my paper
classification astro-ph.IM
keywords galactic cosmic raysneutron monitorssolar modulationenergy spectraartificial neural networksforce-field approximationproton and helium fluxesspace weather
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that the worldwide network of neutron monitors—ground-based detectors of cosmic-ray secondaries—can act as a single real-time spectrometer for galactic cosmic-ray protons and helium. By training artificial neural networks on seventeen monitor count rates together with sunspot number, geomagnetic Ap index, and heliospheric polarity, the authors reconstruct daily energy spectra that match satellite data from 2011–2019 substantially better than a classic yield-function plus force-field approach, reducing error by a factor of 2–4 and keeping chi-squared per degree of freedom near unity. The same trained networks are then applied to 2006–2011 and 2019–2022, periods lacking daily satellite spectra, and the results agree with independent measurements within known systematic differences. If correct, the global neutron monitor network becomes a continuous cosmic-ray spectrometer covering solar-cycle phases past, present, and future without waiting for spacecraft data releases.

Core claim

The central claim is that a supervised neural network, trained on daily AMS-02 proton and helium fluxes with multi-station neutron-monitor count rates plus heliophysical indices as inputs, learns the solar-modulation mapping well enough to reproduce spectra in 2011–2019 with roughly 10% error at the lowest energies, falling to about 1% mid-range, chi-squared per degree of freedom mostly between 0.1 and 2, and about 90% of test days reconstructed within 3 sigma in every energy bin. Applied outside the training interval, the networks extrapolate to 2006–2011 and 2019–2022, matching available averaged satellite measurements within established inter-experiment differences; the paper acknowledges

What carries the argument

The load-bearing machinery is the trained neural-network mapping from twenty predictor variables—seventeen neutron-monitor count rates spanning geomagnetic cutoff rigidities from 0 to about 17 GV, sunspot number with charge-sign-dependent propagation delays, geomagnetic Ap index, and heliospheric polarity A—to the normalized daily AMS-02 proton and helium energy spectra. Complementary machinery is the modified yield-function force-field pipeline, which parameterizes spectra with two solar-modulation parameters fitted to five low-cutoff monitors and includes refined heavy-nuclei scaling and helium isotopic composition. Feature-attribution analysis shows the networks behave consistently with p

Load-bearing premise

The load-bearing premise is that a neural network trained on 2011–2019 conditions continues to work in periods outside that range—especially the deep 2008–2009 solar minimum, which has no counterpart in the training data, and post-2019 activity states—so the claimed extension to 2006–2011 and 2019–2022 stands or falls on extrapolation reliability.

What would settle it

Retrain the networks on data from 2011–2015 only and validate on 2016–2019 satellite daily spectra; if the error on that contiguous held-out period is substantially worse than on a random 30% test split, the model has not learned a solar-cycle-general mapping and the extrapolated reconstructions are unsupported.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the central claim holds, the global neutron-monitor network can produce daily proton and helium spectra covering 2006–2022 and continue in near-real time, with no dependence on spacecraft data availability.
  • The neural-network method gives a factor-of-2-to-4 lower mean absolute percentage error than the yield-function/force-field method across energies in 2011–2019, with chi-squared per degree of freedom typically 0.1–2.
  • The network-based reconstructions fill gaps in satellite records and reproduce short-term phenomena such as 27-day variations and Forbush decreases, though the paper notes that amplitude accuracy of these events needs further study.
  • The comparison exposes the force-field approximation's systematic shortcomings—overestimated short-term low-energy variation and underestimated high-energy modulation—and reveals a polarity-dependent hysteresis that constrains future solar-modulation models.

Where Pith is reading between the lines

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

  • Beyond the paper: a temporal holdout validation—training on 2011–2015 and testing on 2016–2019 satellite data—would directly measure extrapolation skill across solar-cycle phases, something the random 70/30 split does not test.
  • Beyond the paper: the same twenty-input architecture could be retargeted to sub-daily (hourly) proton and helium spectra, since neutron-monitor data are available at that cadence, turning the network into a storm-time spectrometer.
  • Beyond the paper: the polarity-dependent hysteresis visible in the method comparison suggests the networks have implicitly learned drift-related charge-sign effects; reconstructing electron spectra, where the charge-sign product flips, would test whether the mapping encodes physics rather than memorizing the training epoch.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents two methods for reconstructing time-resolved galactic cosmic-ray proton and helium energy spectra from the global neutron monitor (NM) network. The first method uses calibrated NM yield functions together with a two-parameter force-field approximation. The second uses neural networks (CNN and MLP) trained on daily AMS-02 spectra from 2011–2019, with NM count rates, sunspot number, Ap index, and heliospheric polarity as predictors. The neural-network branch is evaluated on a held-out 30% test set, reports MAPE reduced by factors of ~2–4 relative to the yield-function method and χ²/dof mostly in 0.1–2, and is then applied to 2006–2011 (validated against PAMELA) and 2019–2022 (validated against AMS-02 Bartels averages). The authors conclude that the global NM network can serve as an effective real-time GCR spectrometer.

Significance. If the claims hold, this is a valuable contribution to space-weather and cosmic-ray research: it would enable daily GCR spectra without relying on satellite data, extend coverage to periods where direct measurements are sparse, and provide a quantitative comparison of a physics-guided force-field approach with data-driven reconstruction. The manuscript has notable strengths: the neural-network claims are supported by a genuine holdout test and by out-of-period validation against two independent instruments; the authors provide a public repository for the models; SHAP analysis gives physically sensible station–energy associations; and the inclusion of helium spectra plus the detailed heavy-nuclei and 3He treatment in the yield-function branch goes beyond previous NM-based work. The main weaknesses are the extrapolation regime for the headline deliverable and the lack of a fully independent evaluation of the yield-function branch.

major comments (3)
  1. [§3.2, Figs. 6a and 7a] The headline deliverable—spectra for periods without satellite data—lies precisely in the extrapolation regime. The text concedes that the sunspot numbers and NM count rates during the 2006–2009 minimum fall outside the 2011–2019 training range and that the networks must extrapolate. The reconstructed low-energy proton flux in 2009–2011 is ~10% below PAMELA. The paper attributes this to the known PAMELA–AMS-02 offset (Martucci et al. 2018), but it does not apply or quantitatively correct for that offset before comparing. Since the extrapolation error and the inter-experiment offset have the same sign and comparable magnitude, the conclusion of agreement 'within known systematics' is asserted rather than demonstrated. Please add a quantitative decomposition: apply the quoted offset, show residual distributions with and without it, and give extrapolation-uncertainty estimates. A constructi
  2. [§2.1.2, Eq. (8), and §3.1, Figs. 5a–b] The yield-function branch is calibrated on a randomly selected 30% (K=772 days) of the same 2011–2019 AMS-02 interval on which its MAPE and χ²/dof are then reported in §3.1. The neural-network metrics, by contrast, are computed on a held-out 30% test split. The reported factor-of-2–4 improvement of the neural networks over the YF+FF method may therefore partly reflect calibration/overlap rather than predictive skill. Please report YF+FF metrics on the complement of the calibration subset, or use a cross-validation loop, and state the random-seed dependence of the kappa, h*, and daily force-field fits.
  3. [§3.1 and §4] The short-term-disturbance claim is internally hedged. The text says the networks 'allow to reconstruct short-term disturbances' and the conclusion says they 'seamlessly reproduce' them, while §3.1 states that 'the question of models ability to accurately reproduce amplitudes of periodic variations and FDs needs further exploration.' No event-based validation of Forbush-decrease amplitudes or 27-day wave amplitudes is provided. Because the abstract advertises reproduction of short-term disturbances as a capability, either add quantitative FD/27-day event validation or soften the claim in the abstract and conclusion.
minor comments (5)
  1. [§2.2.2] In the standardization paragraph, the normalized flux j(T_l,t_k) is said to be defined by 'Equation 20', but the displayed normalization is Equation (19); Equation (20) is the η metric. Please correct the cross-reference.
  2. [§2.2.1] The manual jump corrections for TXBY, YKTK, and LMKS are described only as 'corrected manually.' For reproducibility, please provide the dates and correction factors, or at least a machine-readable list in the repository, and state any criteria used to distinguish genuine geophysical jumps from instrumental jumps.
  3. [Figure 2] The architecture schematic is too small to read the layer sizes, kernel sizes, and pooling/stride choices. Include these details in the caption or as a table so the CNN architecture can be reproduced without reverse-engineering the code.
  4. [§3.2] The reconstructed spectral time series are shown without uncertainty bands. For the extrapolation periods, prediction intervals (e.g., from an ensemble or dropout) would make the comparison with PAMELA/AMS-02 much more informative.
  5. [Eq. (20)] The formula for η is typeset awkwardly (the expression 'η = (20) 1/K ...' is broken). Please reformat the display equation.

Circularity Check

1 steps flagged

YF+FF validation includes the AMS-02 calibration subset; the NN branch is a genuine held-out/out-of-sample prediction.

specific steps
  1. fitted input called prediction [Sec. 2.1.2 (Eq. 8; YF calibration) and Sec. 3.1 (Eqs. 22-23; 2011-2019 evaluation, K=2572)]
    "For each NM, we randomly select 30% of the dataset (K= 772 points) and fit the theoretically computed count rates to the experimental values: ... obtaining the optimal values of κ_i and h∗_i ... These values are then fixed ... MAPE(T_l) = ... where ... K= 2572 is the number of days in the time series."

    The YF calibration parameters κ_i and h*_i are fitted via Eq. (8) to a random 30% slice (K=772 days) of the AMS-02 2011-2019 spectra. The YF+FF method is then evaluated in Section 3.1 over the full 2011-2019 dataset (K=2572 days), which includes those same 772 calibration days. Thus a substantial part of the reported MAPE and chi^2/dof for the YF branch is computed on data used to tune the instrument response, so it is an in-sample fit rather than an independent prediction. No separate metric excluding the calibration subset is provided, so the comparison between the two methods is partly circular.

full rationale

The neural-network branch is standard supervised learning: features are 17 NM count rates plus SSN, Ap, and polarity A; targets are AMS-02 daily spectra. The authors split the data 70/30, select hyperparameters via cross-validation, and report test-set MAPE, eta, and chi^2/dof, then apply the trained networks to 2006-2011 and 2019-2022 data not used in training. This is genuinely predictive and independently supports the central claim. The only concrete circularity found is in the yield-function branch: kappa_i and h*_i are fitted via Eq. (8) to a random 30% subset (K=772) of AMS-02 2011-2019 spectra, and the YF+FF method's reported 2011-2019 MAPE and chi^2/dof (Eqs. 22-23) are computed over K=2572 days, i.e., they include those calibration days. No separate out-of-sample metric for the YF branch is given, so part of its stated reconstruction quality is in-sample rather than predicted. This contaminates the paper's two-method comparison but does not undercut the NN headline claim. The self-citations (Siruk et al. 2023, 2024; Krainev et al. 2018) are contextual and not load-bearing. The Section 3.2 caveat about extrapolation outside the training range is a correctness risk, not a circularity.

Axiom & Free-Parameter Ledger

7 free parameters · 8 axioms · 0 invented entities

The YF branch rests on the Mishev et al. (2020) yield functions, the Bisschoff–Potgieter LIS, the time-invariant heavy-nuclei scaling S(T), and the Shen et al. (2021) two-parameter force-field form — all external domain assumptions, several of which the paper's own output contradicts (χ²/dof up to ~10² near solar maximum; polarity-dependent hysteresis). The NN branch adds AMS-02-as-ground-truth and a transfer assumption: the 2011–2019 mapping extrapolates to 2006–2011 and 2019–2022. Fitted quantities: κ_i, h*_i, daily φ0/φ1, and the network weights; the h*_i experiment fit back to its own input, which the paper honestly reports. No invented entities.

free parameters (7)
  • κ_i NM calibration scaling factors = 1.276, 1.622, 1.776, 1.966, 0.898 (BRBG, THUL, APTY, TXBY, OULU)
    Fit by Eq (8) to 30% of AMS-02 2011–2019 daily data to match Eq (7) to measured count rates; fixed thereafter and used in Eq (10) for all YF-based reconstructions.
  • h*_i effective atmospheric depth = 1013, 1020, 1002, 1025, 1025 g/cm² (≈ h)
    Second calibration parameter proposed in Eq (6) to capture energy-dependent NM non-ideality; fits returned h* ≈ h and the paper concludes it is unnecessary (§2.1.2) — a self-reported null innovation.
  • φ0(t), φ1(t) daily force-field parameters = daily time series, not tabulated
    Two free parameters of Eq (14) fit per day from five NM count rates via Eq (11); they define the reconstructed spectra, so the 2011–2019 spectra are fitted, not predicted.
  • ANN weights (CNN and MLP) = not tabulated
    Optimized by Eq (21) on 70% of AMS-02 daily spectra 2011–2019; all reconstructed spectra are deterministic functions of these fitted parameters.
  • Propagation delays τ±, δt± = τ+ = 1 BR, δt+ = 4 BR; τ− = 3 BR, δt− = 13 BR
    Hand-fixed in Eq (17) from Cholis et al. 2016, Shen et al. 2021, Strauss et al. 2011, etc.; not fitted here and energy dependence deliberately ignored.
  • Polarity transition half-width Δt = 3 years
    Chosen following Shen et al. (2021) for the smooth-step A(t) model in Eq (16); affects the polarity-reversal input years 2013–2016.
  • Heavy-nuclei multipliers for S(T) = 0.24×Si (Ca/Co/Ni); 1.05×N-group flux (P–Mn)
    Assumed composition ratios in Eq (3) taken from Israel et al. 2018 and Boschini et al. 2020; they propagate into every YF-based count rate through S(T).
axioms (8)
  • domain assumption Mishev et al. (2020) NM64 yield functions Y_p(T,h), Y_He(T,h) are valid for all five monitors (Eqs 5–7).
    Invoked in Eqs (5)–(7); yield-function uncertainty is a known dominant error source (Koldobskiy et al. 2019, cited in §1), and κ_i plus the rejected h*_i are the only corrections applied.
  • domain assumption Same yield function for all Z≥2 nuclei, scaled by nucleon number (Z/A≈0.5 ⇒ identical modulation).
    §2.1.1, following Mishev & Velinov 2011 and Engel et al. 2011; this justifies collapsing all heavy nuclei into the S(T) factor of Eqs (3)–(4).
  • domain assumption S(T) (Eq 3) is time-independent: heavy-nuclei-to-helium ratios are constant over the solar cycle.
    Eqs (2)–(3), citing Koldobskiy et al. 2019 and Aguilar et al. 2025b; the paper itself notes a ~0.2% temporal 3He/4He variation (§2.1.1), so strict time-independence is approximate.
  • domain assumption Shen et al. (2021) two-parameter φ(T) form (Eq 14) captures true spectral modulation at NM-relevant energies.
    §2.1.3; the paper's own findings (χ²/dof up to ~10² near solar maximum, exaggerated short-term amplitudes, polarity-dependent hysteresis) show this assumption fails in measurable ways.
  • domain assumption Bisschoff & Potgieter (2016) local interstellar spectra (Eq 13) are correct.
    Input to Eq (12); the LIS choice is a known source of bias in force-field parameter estimates.
  • domain assumption NM responses are time-stationary after pressure/efficiency corrections; cutoff rigidities from Mishev et al. (2020) are constant.
    Eqs (1), (10), (11); contradicted by the TXBY/YKTK/LMKS count-rate jumps requiring manual correction (§2.2.1) and by storm-time R_c changes (Tyasto et al. 2013).
  • domain assumption AMS-02 daily spectra are unbiased training targets; PAMELA–AMS-02 differences are constant ~10% systematics.
    §2.2.1 targets and §3.2 validation; if inter-experiment offsets vary with energy and time, the 2006–2011 'agreement' claim weakens.
  • domain assumption The NN input–output mapping trained on 2011–2019 transfers to 2006–2011 and 2019–2022 conditions.
    Required for the headline extrapolations; §3.2 admits the 23/24-minimum inputs are out-of-training-range and that tree models were discarded for the same reason.

pith-pipeline@v1.3.0-alltime-deepseek · 22090 in / 27558 out tokens · 285173 ms · 2026-08-01T08:44:42.769612+00:00 · methodology

0 comments
read the original abstract

We present a framework that reconstructs time-resolved galactic cosmic-ray (GCR) proton and helium energy spectra from the global neutron monitor network, providing data about GCR flux without direct satellite observations. Two methods are utilized and compared: a calibrated yield function plus force-field scheme and artificial neural networks trained on multi-station neutron monitor count rates coupled with heliophysical indices. The reconstructed spectral time series reproduce both large-scale solar-cycle modulation and short-term disturbances and extend to periods lacking daily spacecraft data, including 2006-2011 (consistent with PAMELA) and 2019-2022 (consistent with AMS-02 Bartels rotation averages). Artificial neural networks deliver excellent performance across energies, with markedly lower mean absolute percentage error and $\chi^2/\mathrm{dof}$ near unity. A thorough validation confirms robustness and establishes neutron monitors as an effective real-time GCR spectrometer that can be utilized for various purposes.

Figures

Figures reproduced from arXiv: 2607.21009 by Andrey Mayorov, Rustam Yulbarisov, Stepan Siruk, Victor Kuzminov, Vladislav Alekseev.

Figure 1
Figure 1. Figure 1: — Time series of indices of solar (sunspot number) and geomagnetic (Ap) activity, and the configuration of the heliosphere (heliospheric magnetic field polarity) in 2006–2022. median to fill large gaps, so the response variables retain missing entries. In addition to NM data, we account for the large-scale heliospheric magnetic field configuration via the polar￾ity A. When the polarity is positive (A > 0),… view at source ↗
Figure 1
Figure 1. Figure 1: In addition to SSN and A, we include the Ap index, which is not specially preprocessed. This pa￾rameter reflects geomagnetic activity, which alters the geomagnetic cutoff rigidity (Danilova et al. 2025) and may therefore affect NM count rates (see (Kovalev et al. 2022) and Equation 1). These three datasets are used alongside the count rates of the seventeen NMs, yielding twenty predictor variables in total… view at source ↗
Figure 2
Figure 2. Figure 2: — Schematic for CNN and MLP neural architec￾tures [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: — Mean absolute SHAP values for the MLP mod￾els predicting proton (a) and helium (b) fluxes. Regarding solar and geomagnetic activity predictors, the SSN and the heliospheric magnetic field polarity A are informative for GCR flux prediction. By contrast, the Ap index should matter only during strong geomag￾netic disturbances due to their influence on geomagnetic cutoff rigidity in NMs locations and therefo… view at source ↗
Figure 4
Figure 4. Figure 4: — Comparison of AMS-02 daily data (2011–2019) and the results of GCR reconstruction with two methods. Panel (a): low-energy proton flux; panel (b): low-energy helium flux; panels (c) and (d): high-energy proton and helium fluxes, respectively. derestimates solar modulation effects on both long and short timescales. Specifically, the helium flux at T ≈ 20 GeV/nuc is predicted to remain nearly constant, wher… view at source ↗
Figure 5
Figure 5. Figure 5: — Panels (a) and (b): mean absolute percentage error of FFA-based reconstruction of proton and helium fluxes as a function of particle energy, as well as χ 2/dof as a function of time, respectively. Panels (c) and (d): same for the CNN-based proton flux and MLP-based helium flux reconstruction. lation. The time series of particle fluxes at different energies reconstructed with machine-learning-based method… view at source ↗
Figure 6
Figure 6. Figure 6: — Panels (a)–(d): comparison of PAMELA 1–3-month-averaged data (2006–2011) and the results of GCR reconstruction with two methods. Panel (a): low-energy proton flux; panel (b): low-energy helium flux; panels (c) and (d): high-energy proton and helium fluxes, respectively. Panels (e)–(h): same for AMS-02 BR-averaged data (2019–2022). based on the NM YF and FFA. In addition to MAPE and χ 2/dof, we compute th… view at source ↗
Figure 7
Figure 7. Figure 7: — Panels (a) and (b): proton energy spectra reconstructed with neural networks compared to the results of PAMELA (a) and AMS-02 (b) measurements. Panels (c) and (d): same for helium energy spectra. the NM YF+FFA method, along with the experimental data. For 2019–2022, good agreement is observed between AMS-02 data and neural-network predictions. The sole exception is high-energy helium fluxes (Figure 6h), … view at source ↗
Figure 8
Figure 8. Figure 8: shows scatter plots of fixed-energy fluxes re￾constructed for the entire 2006–2022 period. The color gradient encodes time. For protons (Figures 8a and 8b), points from 2006 are bright blue; as solar activity de￾creases toward the 23/24 minimum (2009), the flux reaches its maximum (dark blue). During the subsequent rise, the color shifts to purple; between the cycle-24 max￾imum (2014) and the 24/25 minimum… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

104 extracted references · 12 canonical work pages

  1. [1]

    Examples of its use , author=

    Station-ring method in the study of cosmic-ray variations: 2. Examples of its use , author=. Geomagnetism and Aeronomy , volume=. 2020 , publisher=

    2020

  2. [2]

    , keywords =

    Time Dependence of the Proton Flux Measured by PAMELA during the 2006 July-2009 December Solar Minimum. , keywords =. doi:10.1088/0004-637X/765/2/91 , archivePrefix =. 1301.4108 , primaryClass =

    Pith/arXiv arXiv 2006

  3. [3]

    , year = 2019, volume =

    Properties of Cosmic Helium Isotopes Measured by the Alpha Magnetic Spectrometer. , year = 2019, volume =

    2019

  4. [4]

    , year = 2021, volume =

    The Alpha Magnetic Spectrometer (AMS) on the international space station: Part II - Results from the first seven years. , year = 2021, volume =

    2021

  5. [5]

    , year = 2021, volume =

    Periodicities in the Daily Proton Fluxes from 2011 to 2019 Measured by the Alpha Magnetic Spectrometer on the International Space Station from 1 to 100 GV. , year = 2021, volume =

    2011

  6. [6]

    , year = 2021, volume =

    Properties of Iron Primary Cosmic Rays: Results from the Alpha Magnetic Spectrometer. , year = 2021, volume =

    2021

  7. [7]

    , year = 2021, volume =

    Properties of Heavy Secondary Fluorine Cosmic Rays: Results from the Alpha Magnetic Spectrometer. , year = 2021, volume =

    2021

  8. [8]

    , year = 2021, volume =

    Properties of a New Group of Cosmic Nuclei: Results from the Alpha Magnetic Spectrometer on Sodium, Aluminum, and Nitrogen. , year = 2021, volume =

    2021

  9. [9]

    , year = 2022, volume =

    Properties of Daily Helium Fluxes. , year = 2022, volume =

    2022

  10. [10]

    , year = 2023, volume =

    Properties of Cosmic-Ray Sulfur and Determination of the Composition of Primary Cosmic-Ray Carbon, Neon, Magnesium, and Sulfur: Ten-Year Results from the Alpha Magnetic Spectrometer. , year = 2023, volume =

    2023

  11. [11]

    , volume=

    Properties of cosmic deuterons measured by the Alpha Magnetic Spectrometer , author=. , volume=. 2024 , publisher=

    2024

  12. [12]

    Physical Review Letters , volume=

    Antiprotons and elementary particles over a solar cycle: Results from the Alpha Magnetic Spectrometer , author=. Physical Review Letters , volume=. 2025 , publisher=

    2025

  13. [13]

    Physical Review Letters , volume=

    Solar modulation of cosmic nuclei over a solar cycle: Results from the Alpha Magnetic Spectrometer , author=. Physical Review Letters , volume=. 2025 , publisher=

    2025

  14. [14]

    Neurocomputing , volume=

    Backpropagation and stochastic gradient descent method , author=. Neurocomputing , volume=. 1993 , publisher=

    1993

  15. [15]

    Journal of Geophysical Research (Space Physics) , year = 2017, volume =

    Neutron Monitors and Cosmogenic Isotopes as Cosmic Ray Energy-Integration Detectors: Effective Yield Functions, Effective Energy, and Its Dependence on the Local Interstellar Spectrum. Journal of Geophysical Research (Space Physics) , year = 2017, volume =

    2017

  16. [16]

    Advances in Space Research , volume=

    Exploring relations between solar activity, cosmic rays, clouds and earth climate using machine learning techniques , author=. Advances in Space Research , volume=. 2024 , publisher=

    2024

  17. [17]

    Geomagnetism and Aeronomy , year = 2018, month = may, volume =

    Global Survey Method for the World Network of Neutron Monitors. Geomagnetism and Aeronomy , year = 2018, month = may, volume =. doi:10.1134/S0016793218030039 , adsurl =

    work page doi:10.1134/s0016793218030039 2018

  18. [18]

    , year = 2016, volume =

    New local interstellar spectra for protons, helium and carbon derived from PAMELA and Voyager 1 observations. , year = 2016, volume =. doi:10.1007/s10509-015-2633-8 , archivePrefix =. 1512.04836 , primaryClass =

    Pith/arXiv arXiv 2016

  19. [19]

    , year = 2019, volume =

    New Very Local Interstellar Spectra for Electrons, Positrons, Protons, and Light Cosmic Ray Nuclei. , year = 2019, volume =. doi:10.3847/1538-4357/ab1e4a , archivePrefix =. 1902.10438 , primaryClass =

    Pith/arXiv arXiv 2019

  20. [20]

    , year = 2020, volume =

    Inference of the Local Interstellar Spectra of Cosmic-Ray Nuclei Z 28 with the GALPROP-HELMOD Framework. , year = 2020, volume =. doi:10.3847/1538-4365/aba901 , archivePrefix =. 2006.01337 , primaryClass =

    Pith/arXiv arXiv 2020

  21. [21]

    EURADOS report , volume=

    Comparison of codes assessing radiation exposure of aircraft crew due to galactic cosmic radiation , author=. EURADOS report , volume=

  22. [22]

    Machine learning , volume=

    Random forests , author=. Machine learning , volume=. 2001 , publisher=

    2001

  23. [23]

    , year = 2000, volume =

    Rigidity dependence of cosmic ray proton latitudinal gradients measured by the Ulysses spacecraft: Implications for the diffusion tensor. , year = 2000, volume =

    2000

  24. [24]

    Journal of Geophysical Research (Space Physics) , keywords =

    Limitations of the force field equation to describe cosmic ray modulation. Journal of Geophysical Research (Space Physics) , keywords =. doi:10.1029/2003JA010098 , adsurl =

    work page doi:10.1029/2003ja010098

  25. [25]

    Journal of Geophysical Research: Space Physics , volume=

    Cosmic-ray yield and response functions in the atmosphere , author=. Journal of Geophysical Research: Space Physics , volume=. 2012 , publisher=

    2012

  26. [26]

    Space Weather , year = 2023, month = apr, volume =

    Astronaut Radiation Dose Calculation With a New Galactic Cosmic Ray Model and the AMS-02 Data. Space Weather , year = 2023, month = apr, volume =. doi:10.1029/2022SW003285 , adsurl =

    work page doi:10.1029/2022sw003285 2023

  27. [27]

    , keywords =

    A predictive analytic model for the solar modulation of cosmic rays. , keywords =. doi:10.1103/PhysRevD.93.043016 , archivePrefix =. 1511.01507 , primaryClass =

    Pith/arXiv arXiv

  28. [28]

    Cosmic Rays and Earth: Proceedings of an ISSI Workshop, 21--26 March 1999, Bern, Switzerland , pages=

    Neutron monitor response functions , author=. Cosmic Rays and Earth: Proceedings of an ISSI Workshop, 21--26 March 1999, Bern, Switzerland , pages=. 2000 , organization=

    1999

  29. [29]

    Il Nuovo Cimento C , volume=

    On cosmic-ray cut-off terminology , author=. Il Nuovo Cimento C , volume=. 1991 , publisher=

    1991

  30. [30]

    , year = 2016, volume =

    Solar Modulation of the Local Interstellar Spectrum with Voyager 1, AMS-02, PAMELA, and BESS. , year = 2016, volume =. doi:10.3847/0004-637X/829/1/8 , archivePrefix =. 1511.08790 , primaryClass =

    Pith/arXiv arXiv 2016

  31. [31]

    Solar-Terrestrial Physics , year = 2025, month = sep, volume =

    Geomagnetic cutoff of cosmic rays during the March 23─24, 2023 magnetic storm: Relationship with solar wind parameters and geomagnetic activity taking into account latitudinal effects. Solar-Terrestrial Physics , year = 2025, month = sep, volume =. doi:10.12737/stp-113202505 , adsurl =

    work page doi:10.12737/stp-113202505 2023

  32. [32]

    8th European Cosmic Ray Symposium, Rome , pages=

    Specific yield function S (P) for a neutron monitor at sea level , author=. 8th European Cosmic Ray Symposium, Rome , pages=

  33. [33]

    Water Resources Research , volume=

    Nature's neutron probe: Land surface hydrology at an elusive scale with cosmic rays , author=. Water Resources Research , volume=. 2010 , publisher=

    2010

  34. [34]

    , keywords =

    Deep Learning the Forecast of Galactic Cosmic-Ray Spectra. , keywords =. doi:10.3847/2041-8213/ada427 , archivePrefix =. 2410.21046 , primaryClass =

    Pith/arXiv arXiv 2041

  35. [35]

    Annals of Statistics , pages=

    Least Angle Regression , author=. Annals of Statistics , pages=. 2004 , publisher=

    2004

  36. [36]

    Annual review of nuclear and particle science , volume=

    Extensive air showers and hadronic interactions at high energy , author=. Annual review of nuclear and particle science , volume=. 2011 , publisher=

    2011

  37. [37]

    Physical Review , year = 1937, volume =

    On the Effects in Cosmic-Ray Intensity Observed During the Recent Magnetic Storm. Physical Review , year = 1937, volume =

    1937

  38. [38]

    Terrestrial Magnetism and Atmospheric Electricity (Journal of Geophysical Research) , year = 1938, volume =

    On cosmic-ray effects associated with magnetic storms. Terrestrial Magnetism and Atmospheric Electricity (Journal of Geophysical Research) , year = 1938, volume =

    1938

  39. [39]

    time series

    Neutron monitors and muon detectors for solar modulation studies: 2. time series. Advances in Space Research , keywords =. doi:10.1016/j.asr.2016.06.027 , archivePrefix =. 1607.01976 , primaryClass =

    Pith/arXiv arXiv 2016

  40. [40]

    Journal of Geophysical Research (Space Physics) , year = 2017, volume =

    An Empirical Modification of the Force Field Approach to Describe the Modulation of Galactic Cosmic Rays Close to Earth in a Broad Range of Rigidities. Journal of Geophysical Research (Space Physics) , year = 2017, volume =. doi:10.1002/2017JA024763 , archivePrefix =. 1710.10834 , primaryClass =

    Pith/arXiv arXiv 2017

  41. [41]

    , year = 1968, volume =

    Solar Modulation of Galactic Cosmic Rays. , year = 1968, volume =

    1968

  42. [42]

    Canadian Journal of Physics , year = 1964, month = dec, volume =

    Experimental Investigation of the NM-64 Neutron Monitor. Canadian Journal of Physics , year = 1964, month = dec, volume =. doi:10.1139/p64-222 , adsurl =

    work page doi:10.1139/p64-222 1964

  43. [43]

    Advances in Space Research , year = 1999, volume =

    Corotating interaction regions. Advances in Space Research , year = 1999, volume =

    1999

  44. [44]

    Elemental Composition at the Cosmic-Ray Source Derived from the ACE-CRIS Instrument. I. _ 6 C to _ 28 Ni. , year = 2018, volume =

    2018

  45. [45]

    Journal of Space Weather and Space Climate , keywords =

    New cosmic ray observations at Syowa Station in the Antarctic for space weather study. Journal of Space Weather and Space Climate , keywords =. doi:10.1051/swsc/2021005 , archivePrefix =. 2101.09887 , primaryClass =

    arXiv

  46. [46]

    Proc 33rd ICRC, Rio de Janeiro , year=

    Reception coefficients and energy characteristics of the ground level cosmic ray detectors , author=. Proc 33rd ICRC, Rio de Janeiro , year=

  47. [47]

    Journal of Geophysical Research (Space Physics) , year = 2019, volume =

    Validation of the Neutron Monitor Yield Function Using Data From AMS-02 Experiment, 2011-2017. Journal of Geophysical Research (Space Physics) , year = 2019, volume =. doi:10.1029/2018JA026340 , archivePrefix =. 1904.01929 , primaryClass =

    Pith/arXiv arXiv 2011

  48. [48]

    Solar Physics , volume=

    Time lag between cosmic-ray and solar variability: Sunspot numbers and open solar magnetic flux , author=. Solar Physics , volume=. 2022 , publisher=

    2022

  49. [49]

    PoS (ICRC2023) , year = 2023, series =

    Reconstruction of solar modulation potential from AMS-02 daily data for the period 2011 – 2019 and its comparison with indirect cosmic-ray measurements. PoS (ICRC2023) , year = 2023, series =

    2011

  50. [50]

    Journal of Atmospheric and Solar-Terrestrial Physics , keywords =

    A proposal to extend the spectrographic global survey method. Journal of Atmospheric and Solar-Terrestrial Physics , keywords =. doi:10.1016/j.jastp.2022.105887 , adsurl =

    arXiv 2022

  51. [51]

    Physics of Atomic Nuclei , year = 2018, volume =

    Relationship between the Longitude Distribution of the Heliospheric Characteristics and the GCR Intensity in 2007-2008 and 2014-2015. Physics of Atomic Nuclei , year = 2018, volume =

    2007

  52. [52]

    Solar-Terrestrial Physics , year = 2025, month = jun, volume =

    Geomagnetic cutoff rigidity in neutron monitor locations. Solar-Terrestrial Physics , year = 2025, month = jun, volume =. doi:10.12737/stp-112202512 , adsurl =

    work page doi:10.12737/stp-112202512 2025

  53. [53]

    34th International Cosmic Ray Conference (ICRC2015) , year = 2015, series =

    The mini neutron monitor programme. 34th International Cosmic Ray Conference (ICRC2015) , year = 2015, series =. doi:10.22323/1.236.0223 , adsurl =

    work page doi:10.22323/1.236.0223 2015

  54. [54]

    IEEE transactions on neural networks and learning systems , volume=

    A survey of convolutional neural networks: analysis, applications, and prospects , author=. IEEE transactions on neural networks and learning systems , volume=. 2021 , publisher=

    2021

  55. [55]

    arXiv preprint arXiv:1908.03265 , year=

    On the variance of the adaptive learning rate and beyond , author=. arXiv preprint arXiv:1908.03265 , year=

    Pith/arXiv arXiv 1908

  56. [56]

    Advances in neural information processing systems , volume=

    A unified approach to interpreting model predictions , author=. Advances in neural information processing systems , volume=. 2017 , pages=

    2017

  57. [57]

    Mathematics , volume=

    Application of solar activity time series in machine learning predictive modeling of precipitation-induced floods , author=. Mathematics , volume=. 2023 , publisher=

    2023

  58. [58]

    Journal of Geophysical Research: Space Physics , volume=

    Dependence of the neutron monitor count rate and time delay distribution on the rigidity spectrum of primary cosmic rays , author=. Journal of Geophysical Research: Space Physics , volume=. 2016 , publisher=

    2016

  59. [59]

    The Astrophysical Journal , volume=

    Time dependence of the flux of helium nuclei in cosmic rays measured by the PAMELA experiment between 2006 July and 2009 December , author=. The Astrophysical Journal , volume=. 2020 , publisher=

    2006

  60. [60]

    The Astrophysical Journal Letters , volume=

    Helium fluxes measured by the PAMELA experiment from the minimum to the maximum solar activity for solar cycle 24 , author=. The Astrophysical Journal Letters , volume=. 2022 , publisher=

    2022

  61. [61]

    The Astrophysical Journal Letters , volume=

    Proton fluxes measured by the PAMELA experiment from the minimum to the maximum solar activity for solar cycle 24 , author=. The Astrophysical Journal Letters , volume=. 2018 , publisher=

    2018

  62. [62]

    Journal of Geophysical Research: Space Physics , volume=

    Temporal and spatial evolution of the solar energetic particle event on 20 January 2005 and resulting radiation doses in aviation , author=. Journal of Geophysical Research: Space Physics , volume=. 2009 , publisher=

    2005

  63. [63]

    Space weather , volume=

    The geomagnetic Kp index and derived indices of geomagnetic activity , author=. Space weather , volume=. 2021 , publisher=

    2021

  64. [64]

    Advances in space research , volume=

    Neutron monitors and muon detectors for solar modulation studies: Interstellar flux, yield function, and assessment of critical parameters in count rate calculations , author=. Advances in space research , volume=. 2015 , publisher=

    2015

  65. [65]

    , keywords =

    Energetic Particle Influence on the Earth's Atmosphere. , keywords =. doi:10.1007/s11214-015-0185-4 , adsurl =

    doi:10.1007/s11214-015-0185-4

  66. [66]

    Advances in space research , volume=

    Normalized ionization yield function for various nuclei obtained with full Monte Carlo simulations , author=. Advances in space research , volume=. 2011 , publisher=

    2011

  67. [67]

    Journal of Geophysical Research (Space Physics) , year = 2020, volume =

    Updated Neutron-Monitor Yield Function: Bridging Between In Situ and Ground-Based Cosmic Ray Measurements. Journal of Geophysical Research (Space Physics) , year = 2020, volume =

    2020

  68. [68]

    , year = 2020, volume =

    Study of the 27 Day Variations in GCR Fluxes during 2007-2008 Based on PAMELA and ARINA Observations. , year = 2020, volume =. doi:10.3847/1538-4357/abbdac , archivePrefix =. 2104.11193 , primaryClass =

    Pith/arXiv arXiv 2007

  69. [69]

    Il Nuovo Cimento C , volume=

    Response and yield functions of neutron monitor, galactic cosmic-ray spectrum and its solar modulation, derived from all the available world-wide surveys , author=. Il Nuovo Cimento C , volume=. 1989 , publisher=

    1989

  70. [70]

    Frontiers in neurorobotics , volume=

    Gradient boosting machines, a tutorial , author=. Frontiers in neurorobotics , volume=. 2013 , publisher=

    2013

  71. [71]

    Journal of Geophysical Research Machine Learning and Computation , year = 2025, month = dec, volume =

    Approximation of Daily AMS-02 Spectra With Machine Learning Methods. Journal of Geophysical Research Machine Learning and Computation , year = 2025, month = dec, volume =. doi:10.1029/2025JH000676 , adsurl =

    work page doi:10.1029/2025jh000676 2025

  72. [72]

    , journal =

    Parker, E.N. , journal =. The passage of energetic charged particles through interplanetary space , year =

  73. [73]

    Astroparticle Physics , year = 2007, volume =

    PAMELA A payload for antimatter matter exploration and light-nuclei astrophysics. Astroparticle Physics , year = 2007, volume =. doi:10.1016/j.astropartphys.2006.12.002 , archivePrefix =. astro-ph/0608697 , primaryClass =

    Pith/arXiv arXiv 2007

  74. [74]

    WSEAS Transactions on Circuits and Systems , volume=

    Multilayer perceptron and neural networks , author=. WSEAS Transactions on Circuits and Systems , volume=

  75. [75]

    , year = 2017, volume =

    GLE and Sub-GLE Redefinition in the Light of High-Altitude Polar Neutron Monitors. , year = 2017, volume =. doi:10.1007/s11207-017-1202-4 , archivePrefix =. 1711.06161 , primaryClass =

    Pith/arXiv arXiv 2017

  76. [76]

    Advances in Space Research , keywords =

    Cosmic-ray atmospheric cutoff energies of polar neutron monitors. Advances in Space Research , keywords =. doi:10.1016/j.asr.2022.03.037 , adsurl =

    work page doi:10.1016/j.asr.2022.03.037 2022

  77. [77]

    Living Reviews in Solar Physics , year = 2013, volume =

    Solar Modulation of Cosmic Rays. Living Reviews in Solar Physics , year = 2013, volume =. doi:10.12942/lrsp-2013-3 , archivePrefix =. 1306.4421 , primaryClass =

    Pith/arXiv arXiv 2013

  78. [78]

    Advances in Space Research , year = 2016, volume =

    The effects of magnetic field modifications on the solar modulation of cosmic rays with a SDE-based model. Advances in Space Research , year = 2016, volume =. doi:10.1016/j.asr.2016.01.017 , archivePrefix =. 1506.07305 , primaryClass =

    Pith/arXiv arXiv 2016

  79. [79]

    Living Reviews in Solar Physics , year = 2018, volume =

    Solar wind stream interaction regions throughout the heliosphere. Living Reviews in Solar Physics , year = 2018, volume =

    2018

  80. [80]

    Nuclear Energy and Technology , volume=

    Reconstruction of the helium to proton flux ratio in galactic cosmic rays using neutron monitor data , author=. Nuclear Energy and Technology , volume=. 2026 , publisher=

    2026

Showing first 80 references.