Pith. sign in

REVIEW 3 major objections 6 minor 210 references

This paper claims that training an AI precipitation model on satellite-derived rain estimates and feeding it raw satellite observations makes forecasts substantially more accurate, beating a conventional operational baseline and matching a

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 ↗

Training on IMERG satellite precipitation and directly ingesting satellite observations improves medium-range AI precipitation forecasts, with the largest gains at lead times under 40 hours.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A careful, well-executed finetuning study with two genuinely useful results—satellite-derived training targets and a sensor-agnostic observation encoder—but the headline generalization claim rests on thin independent validation and an untreated circularity risk. the 3 major comments →

arxiv 2608.03959 v1 pith:FCGS6ZU7 submitted 2026-08-04 physics.ao-ph

Prithvi-Precip: Integrating Satellite Observations into an Atmospheric AI Foundation Model for Precipitation Forecasting

Simon Pfreundschuh , Christian D. Kummerow , Johannes Schmude , Sujit Roy , Rahul Ramachandran , Tsengdar Lee , Valentine Anantharaj , Katherine H. Breen This is my paper
classification physics.ao-ph
keywords AI weather predictionprecipitation forecastingsatellite observationsIMERGMERRA-2foundation model fine-tuningautoregressive rolloutquantile regression
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 reading

Prithvi-Precip is a global AI precipitation forecasting system built by fine-tuning the Prithvi-WxC weather foundation model. The paper's central claim is that two interventions—using satellite-based IMERG precipitation estimates rather than MERRA-2 reanalysis fields as training targets, and feeding raw satellite observations directly into the model—each improve forecast skill, and together they substantially beat the GEOS-FP operational baseline while staying competitive with the stronger AIFS system out to about 36–40 hours. The key evidence is generalization: IMERG-trained forecasts agree better with independent ground-radar and gauge data than MERRA-2-trained forecasts, even though MERRA-2-trained forecasts look better when scored against MERRA-2 itself. The paper concludes that better precipitation training targets and fuller use of the satellite observing system are practical levers for advancing medium-range AI precipitation prediction.

Core claim

On its own terms, the paper establishes that an autoregressively rollout-trained version of Prithvi-WxC, extended with a precipitation head and a sensor-agnostic observation encoder, can forecast precipitation up to 96 hours at six-hour intervals with skill that substantially exceeds GEOS-FP precipitation forecasts and matches AIFS through roughly 36–40 hours. The method's force comes from two complementary claims. First, training targets matter: models trained on IMERG V07 satellite precipitation estimates transfer better to independent MRMS radar and INMET gauge observations than models trained on MERRA-2 precipitation, which merely learn reanalysis-specific artifacts. Second, observations

What carries the argument

The load-bearing component is a sensor-agnostic observation encoder: each satellite channel is treated as an independent observation layer with metadata (wavelength, polarization, time offset), tiles are stored sparsely, and a Perceiver-style cross-attention block with a single learnable query compresses the variable-length layer sequence into a fixed-size latent that is merged into the Prithvi-WxC hidden state by a residual MLP. A second mechanism is the precipitation head: a pixel-wise four-layer MLP that outputs 32 quantiles of the precipitation distribution, trained with quantile loss, so the model learns a predictive distribution rather than a point estimate. Autoregressive rollout with

Load-bearing premise

The global conclusion that IMERG targets and satellite ingestion improve skill rests on independent validation over just two regions, CONUS and Brazil; if those regions are unrepresentative of global precipitation, the global claim may overstate the benefit.

What would settle it

Run the same trained-model comparison against independent high-quality precipitation references over, say, Europe, Australia, and tropical Africa. If IMERG-trained forecasts do not beat MERRA-2-trained forecasts there, or if satellite ingestion gives no short-lead improvement outside CONUS and Brazil, the generalization claim collapses.

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

If this is right

  • Forecast skill against independent radar and gauge references improves across the full 96-hour window when training targets switch from MERRA-2 precipitation to IMERG V07.
  • Autoregressive rollout, even with only precipitation in the loss, beats direct lead-time conditioning for precipitation forecasts.
  • Ingesting passive-microwave and geostationary brightness temperatures alongside reanalysis improves short-lead forecasts, with the largest gains in tropical and subtropical regions.
  • The same model can produce forecasts with full, partial, or no satellite observations because of input dropout, which could simplify operational deployment.
  • Prithvi-Precip substantially outperforms GEOS-FP precipitation forecasts and remains competitive with AIFS up to roughly 36–40 hours, after which AIFS pulls ahead.

Where Pith is reading between the lines

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

  • Because the observation encoder is sensor-agnostic, the method should scale with the satellite constellation: adding new sensors should require no architectural change, and the short-lead gain is a testable prediction when next-generation sounders or imagers are included.
  • The regional validation suggests the global improvement maps are a hypothesis rather than a measurement; extending the same evaluation to other continents would either confirm or bound the claimed tropical/subtropical benefit.
  • The 2.7-billion-parameter model underperformed under a reduced rollout budget, which implies model scale is not the current bottleneck; retraining the large model with the full six-step rollout and a longer schedule could extend the competitive lead time beyond 40 hours.
  • Observation-only forecasts beat reanalysis-only at 6 hours but decay quickly, implying raw observations carry immediate precipitation information but little dynamical memory; a hybrid initialization (assimilation for the atmospheric state, observations for precipitation) may be the most robust operational configuration.
Share X Bluesky LinkedIn Reddit HN

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 / 6 minor

Summary. The paper develops Prithvi-Precip, a global AI precipitation forecasting system built by fine-tuning the Prithvi-WxC foundation model with an additional precipitation head. The study examines two interventions: replacing MERRA-2 reanalysis precipitation training targets with IMERG V07 satellite-based precipitation estimates, and directly ingesting raw satellite observations through a proposed sensor-agnostic observation encoder. Forecasting is done autoregressively in 6-hour steps out to 96 hours. The authors evaluate design choices (continuous vs. rollout training, small vs. large model, target dataset), then compare the final system against GEOS-FP and AIFS baselines for March–August 2025, using IMERG, MRMS (CONUS) radar, and INMET (Brazil) gauge data as references. The headline claims are that autoregressive rollout training is clearly superior, that IMERG targets generalize better to independent references than MERRA-2 targets, that satellite observations improve short-lead skill (especially in the tropics/subtropics), and that Prithvi-Precip substantially outperforms GEOS-FP while remaining competitive with AIFS out to roughly 36–40 h.

Significance. If the results hold, the paper makes a useful contribution to AI weather prediction by showing that training-target quality and direct use of satellite observations are both practical levers for precipitation forecast skill. The experimental design has genuine strengths: training on 2000–2020, design tuning on 2022, and evaluation on March–August 2025 is a careful temporal separation; the authors explicitly acknowledge that the IMERG evaluation favors their model; and the paper makes code and data-preparation scripts publicly available. The finding that the choice of the precipitation reference data matters as much as architecture is important for the community. However, the central generalization claim is weakened by the fact that the two supposedly independent evaluation references (MRMS, INMET) likely share gauge information with the gauge correction applied to the IMERG training target, and the paper does not quantify this overlap. In addition, all headline skill differences are presented without uncertainty quantification or significance testing. The significance is therefore contingent on addressing those two issues.

major comments (3)
  1. [Sec. 2e1, 2e2, 3c2, 3c3] The manuscript calls MRMS and INMET 'independent validation data' (Sec. 2e2) and uses them to support the claim that IMERG training targets generalize better than MERRA-2 targets (Secs. 3a2, 3c2, 3c3, abstract). However, Sec. 2e1 states that the IMERG V07 Final run 'additionally applies a monthly gauge corrections over land surfaces.' The gauge networks feeding that correction (e.g., GPCC) likely overlap with the surface gauge data used in MRMS gauge correction and with INMET stations over Brazil. If so, the evaluation is partly circular: the IMERG target and the reference share a common gauge-based climatological bias, so higher IMERG-trained skill may reflect agreement with that shared bias rather than a genuinely more realistic precipitation field. The authors should disclose this overlap, quantify the fraction of shared gauges/stations, test sensitivity to excluding those stations, o
  2. [Sec. 3c, Figs. 11–13] The headline comparisons against GEOS-FP and AIFS—and the conclusions about 'substantial' improvements and competitiveness out to 36–40 h—are based on single deterministic curves over a six-month evaluation period, with no uncertainty quantification. No confidence intervals, bootstrap uncertainty, or paired significance tests are reported for NRMSE, CRPS, or correlation. Differences between Prithvi-Precip and AIFS are visually small at many lead times, and the period March–August 2025 samples only a limited set of weather regimes. The authors should add, at minimum, bootstrap confidence intervals over forecast initializations and paired significance tests for the differences shown in Figs. 11–13 (and ideally in Figs. 6 and 8 as well). Without this, the reader cannot assess whether the reported improvements are robust or within sampling variability.
  3. [Sec. 3b, Figs. 8 and 10; Sec. 3c2, 3c3] The abstract and Sec. 3b state that direct satellite ingestion provides additional improvements at short lead times, with the largest gains in tropical and subtropical regions. Figs. 8 and 10 appear to be evaluated against IMERG (or at least the reference product is not stated in the caption), and Sec. 3c itself notes that IMERG evaluation favors the model. The independent MRMS evaluation (Sec. 3c2) shows only 'very small improvements' over CONUS, and the INMET evaluation (Sec. 3c3) is limited to Brazil. Thus the global tropical/subtropical claim is not supported by independent reference data. The authors should clearly state the reference used in Figs. 8–10 and qualify the abstract-level claim, or provide independent global evidence such as evaluation against a satellite-only precipitation product or a different gauge network not used in IMERG correction.
minor comments (6)
  1. [Sec. 2e1] Typo: 'Pritvh-WxC' should be 'Prithvi-WxC'.
  2. [Sec. 3c] The text refers to 'GOES-FP' in the first paragraph and later 'availability of the GOES-FP forecasts'; the model is elsewhere correctly called GEOS-FP. Please correct.
  3. [Table 1] Table 1 caption says 'satellite observations used to drive the Prithvi-Prithvi forecasts'; should be 'Prithvi-Precip forecasts'.
  4. [Fig. 10] The first-row axis label says 'NMSE' while the caption and text refer to 'NRMS'/'NRMSE'. Make the notation consistent.
  5. [Sec. 2c] Typo in the introduction: 'operating completely in observations pace' should be 'observation space'.
  6. [Author affiliations] Morgan State 'Univeresity' should be 'University'.

Circularity Check

0 steps flagged

No significant circularity: core claims rest on independent holdout evaluations; in-distribution IMERG evaluation is explicitly disclosed as favoring the model.

full rationale

The paper's central claims are empirical comparisons of trained forecast models rather than analytical derivations that reduce to their inputs. The target-data claim (IMERG vs MERRA-2 training targets) is decided on MRMS and INMET evaluations (Sec. 3c2, 3c3) that were not used in training, and the paper explicitly flags the IMERG evaluation as favoring Prithvi-Precip: 'The evaluation against IMERG V07 will naturally favor the Prithvi-Precip model since it was trained the data' (Sec. 3c). The satellite-ingestion claim is tested by controlled input ablation on the same trained model (Sec. 3b), and its independent-data gains are reported, including the statement that CONUS gains are small. No fitted parameter is renamed as a prediction; no uniqueness theorem or ansatz is imported from the authors' prior work. Self-citations (Schmude et al. 2024 for Prithvi-WxC; Pfreundschuh et al. for quantile loss and GPROF-NN) provide the base architecture and loss, but the finetuning results are independently evaluated. The only conceivable circularity would be if IMERG Final's monthly gauge correction shared gauge sources with MRMS/INMET, making those references partially non-independent; however, the paper does not state that overlap, and asserting it would require external assumptions rather than an exhibited reduction from the text. Accordingly, there is no demonstrated circular step.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

This is an empirical ML study; the model weights are learned from training data, and all training hyperparameters are disclosed design choices that were not tuned to the independent evaluation metrics. The load-bearing assumptions are the reliability of the reference datasets, the representativeness of the evaluation regions and period, and the inter-calibration consistency of the satellite record.

axioms (4)
  • domain assumption The Prithvi-WxC foundation model is a suitable backbone for finetuned precipitation forecasting
    The entire system is built on this pretrained model; no alternative backbone is benchmarked (Section 2a).
  • domain assumption MRMS gauge-corrected radar over CONUS and INMET gauges over Brazil are reliable independent references at the MERRA-2 grid scale
    These are the only independent references supporting the claimed global improvements (Sections 2e2, 3c).
  • domain assumption Level-1C inter-calibrated brightness temperatures are consistent enough across sensors and generations to be treated as interchangeable observation layers
    The sensor-agnostic encoder relies on this consistency without re-calibration in the model (Section 2c).
  • domain assumption The March-August 2025 evaluation window with 0Z/12Z initializations is representative of long-term forecast skill
    The evaluation period is constrained by GEOS-FP availability and the paper notes diurnal-cycle oscillations from discrete initialization times (Section 3c).

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Prithvi-Precip: Integrating Satellite Observations into an Atmospheric AI Foundation Model for Precipitation Forecasting." pith.science (2026). https://pith.science/paper/FCGS6ZU7

@misc{pith2026260803959,
  author       = {Pith},
  title        = {Pith review of: Prithvi-Precip: Integrating Satellite Observations into an Atmospheric AI Foundation Model for Precipitation Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FCGS6ZU7}},
  note         = {Machine review of arXiv:2608.03959}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Accurate precipitation forecasting remains one of the most challenging problems in weather prediction. While recent AI weather prediction (AIWP) systems have achieved substantial improvements in medium-range forecasting skill, precipitation often remains a secondary target and is commonly learned from reanalysis datasets that contain considerable uncertainty. In this work, we investigate two complementary strategies for improving AI-based precipitation forecasts. Building on the Prithvi-WxC foundation model, we develop Prithvi-Precip, a global precipitation forecasting system, and examine (1) the impact of training targets derived from satellite-based precipitation estimates rather than reanalysis fields and (2) the direct assimilation of satellite observations into the forecasting model. We systematically evaluate key design choices for finetuning the Prithvi-WxC AI foundation model for precipitation forecasting. We find that autoregressive rollout training produces substantially more accurate forecasts than direct conditioning on forecast lead time. Using independent radar-based precipitation estimates for evaluation, we show that training on satellite-derived precipitation targets yields improved forecast accuracy relative to training on MERRA-2 precipitation fields. Furthermore, direct ingestion of satellite observations provides additional improvements at short lead times, with the largest gains occurring in tropical and subtropical regions. Together, these advances enable Prithvi-Precip to substantially improve upon directly comparable precipitation forecasts from the Goddard Earth Observing System. Our results highlight the potential of improved precipitation targets and the direct integration of satellite observations as promising pathways for advancing medium-range AI precipitation forecasting.

Figures

Figures reproduced from arXiv: 2608.03959 by Christian D. Kummerow, Johannes Schmude, Katherine H. Breen, Rahul Ramachandran, Simon Pfreundschuh, Sujit Roy, Tsengdar Lee, Valentine Anantharaj.

Figure 5
Figure 5. Figure 5: In terms of biases, the autoregressive forecast remains nearly bias free up to a lead time of around 36 hours after which it exhibits an increasing tendency to underestimate precipitation. Since the model was trained using roll out up to 36 hours, we suspect that the biases are caused by the model producing predictions outside the lead times it has encountered during training. The direct forecast model pro… 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

210 extracted references · 19 canonical work pages

  1. [1]

    Reviews of Geophysics , author =

    An. Reviews of Geophysics , author =. 2020 , note =. doi:10.1029/2019RG000678 , abstract =

    doi:10.1029/2019rg000678 2020

  2. [2]

    Nat Commun , author =

    Separability and geometry of object manifolds in deep neural networks , volume =. Nat Commun , author =. 2020 , note =. doi:10.1038/s41467-020-14578-5 , abstract =

    doi:10.1038/s41467-020-14578-5 2020

  3. [3]

    Accurate,

    Goyal, Priya and Dollár, Piotr and Girshick, Ross and Noordhuis, Pieter and Wesolowski, Lukasz and Kyrola, Aapo and Tulloch, Andrew and Jia, Yangqing and He, Kaiming , month = apr, year =. Accurate,

  4. [4]

    Journal of Applied Meteorology and Climatology , author =

    Development and. Journal of Applied Meteorology and Climatology , author =. 2020 , note =. doi:10.1175/JAMC-D-20-0084.1 , abstract =

    doi:10.1175/jamc-d-20-0084.1 2020

  5. [5]

    Liu, Zhuang and Mao, Hanzi and Wu, Chao-Yuan and Feichtenhofer, Christoph and Darrell, Trevor and Xie, Saining , year =. A

  6. [6]

    2022 , eprint=

    MaxViT: Multi-Axis Vision Transformer , author=. 2022 , eprint=

    2022

  7. [7]

    , month = mar, year =

    Smith, Leslie N. , month = mar, year =. Cyclical. 2017. doi:10.1109/WACV.2017.58 , abstract =

    doi:10.1109/wacv.2017.58 2017

  8. [8]

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian , month = jun, year =. Deep. 2016. doi:10.1109/CVPR.2016.90 , abstract =

    doi:10.1109/cvpr.2016.90 2016

  9. [9]

    Revisiting

    Bello, Irwan and Fedus, William and Du, Xianzhi and Cubuk, Ekin D and Srinivas, Aravind and Lin, Tsung-Yi and Shlens, Jonathon and Zoph, Barret , keywords =. Revisiting

  10. [10]

    Wightman, Ross and Touvron, Hugo and Jégou, Hervé , month = oct, year =

  11. [11]

    Yu, Fisher and Wang, Dequan and Shelhamer, Evan and Darrell, Trevor , month = jun, year =. Deep. 2018. doi:10.1109/CVPR.2018.00255 , abstract =

    arXiv 2018

  12. [12]

    Aggregated

    Xie, Saining and Girshick, Ross and Dollar, Piotr and Tu, Zhuowen and He, Kaiming , month = jul, year =. Aggregated. 2017. doi:10.1109/CVPR.2017.634 , abstract =

    doi:10.1109/cvpr.2017.634 2017

  13. [13]

    , month = jul, year =

    Huang, Gao and Liu, Zhuang and Van Der Maaten, Laurens and Weinberger, Kilian Q. , month = jul, year =. Densely. 2017. doi:10.1109/CVPR.2017.243 , language =

    doi:10.1109/cvpr.2017.243 2017

  14. [14]

    Tan, Mingxing and Le, Quoc V , year =

  15. [15]

    Neurocomputing , author =

    Deep visual domain adaptation:. Neurocomputing , author =. 2018 , pages =. doi:10.1016/j.neucom.2018.05.083 , language =

    doi:10.1016/j.neucom.2018.05.083 2018

  16. [16]

    Sandler, Mark and Howard, Andrew and Zhu, Menglong and Zhmoginov, Andrey and Chen, Liang-Chieh , month = jun, year =. 2018. doi:10.1109/CVPR.2018.00474 , abstract =

    arXiv 2018

  17. [17]

    Decoupled

    Loshchilov, Ilya and Hutter, Frank , month = jan, year =. Decoupled

  18. [18]

    Wang, Yingheng and Schiff, Yair and Gokaslan, Aaron and Pan, Weishen and Wang, Fei and De Sa, Christopher and Kuleshov, Volodymyr , month = jun, year =

  19. [19]

    Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil , month = jun, year =. An

  20. [20]

    Aytekin, Caglar , month = oct, year =. Neural

  21. [22]

    Kovachki, Nikola and Li, Zongyi and Liu, Burigede and Azizzadenesheli, Kamyar and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , month = apr, year =. Neural

  22. [23]

    Rethinking

    Chen, Liang-Chieh and Papandreou, George and Schroff, Florian and Adam, Hartwig , month = dec, year =. Rethinking

  23. [24]

    Liu, Ze and Lin, Yutong and Cao, Yue and Hu, Han and Wei, Yixuan and Zhang, Zheng and Lin, Stephen and Guo, Baining , month = oct, year =. Swin. 2021. doi:10.1109/ICCV48922.2021.00986 , abstract =

    arXiv 2021

  24. [25]

    Hu, Han and Zhang, Zheng and Xie, Zhenda and Lin, Stephen , month = apr, year =. Local

  25. [26]

    Lam, Remi and Sanchez-Gonzalez, Alvaro and Willson, Matthew and Wirnsberger, Peter and Fortunato, Meire and Alet, Ferran and Ravuri, Suman and Ewalds, Timo and Eaton-Rosen, Zach and Hu, Weihua and Merose, Alexander and Hoyer, Stephan and Holland, George and Vinyals, Oriol and Stott, Jacklynn and Pritzel, Alexander and Mohamed, Shakir and Battaglia, Peter ...

  26. [27]

    Neurocomputing , author =

    A review on the attention mechanism of deep learning , volume =. Neurocomputing , author =. 2021 , pages =. doi:10.1016/j.neucom.2021.03.091 , abstract =

    doi:10.1016/j.neucom.2021.03.091 2021

  27. [28]

    Journal of Geophysical Research: Atmospheres , author =

    Tropical. Journal of Geophysical Research: Atmospheres , author =. 2020 , note =. doi:10.1029/2020JD033107 , abstract =

    doi:10.1029/2020jd033107 2020

  28. [29]

    Clim Dyn , author =

    Global precipitation hindcast quality assessment of the. Clim Dyn , author =. 2019 , keywords =. doi:10.1007/s00382-018-4457-z , abstract =

    doi:10.1007/s00382-018-4457-z 2019

  29. [30]

    Journal of Hydrology , author =

    Evaluation of. Journal of Hydrology , author =. 2021 , keywords =. doi:10.1016/j.jhydrol.2021.127058 , abstract =

    arXiv 2021

  30. [31]

    2021 , pages =

    AAAI , author =. 2021 , pages =. doi:10.1609/aaai.v35i17.17749 , abstract =

    doi:10.1609/aaai.v35i17.17749 2021

  31. [32]

    Zhang, Richard , month = jun, year =. Making

  32. [33]

    Tran, Du and Wang, Heng and Torresani, Lorenzo and Ray, Jamie and LeCun, Yann and Paluri, Manohar , month = apr, year =. A

  33. [34]

    Bulletin of the American Meteorological Society , author =

    Outcomes of the. Bulletin of the American Meteorological Society , author =. 2022 , note =. doi:10.1175/BAMS-D-22-0046.1 , abstract =

    doi:10.1175/bams-d-22-0046.1 2022

  34. [35]

    Parameter-

    Houlsby, Neil and Giurgiu, Andrei and Jastrzebski, Stanislaw and Morrone, Bruna and de Laroussilhe, Quentin and Gesmundo, Andrea and Attariyan, Mona and Gelly, Sylvain , month = jun, year =. Parameter-

  35. [36]

    Gao, Peng and Ma, Teli and Li, Hongsheng and Lin, Ziyi and Dai, Jifeng and Qiao, Yu , month = may, year =

  36. [37]

    Ryali, Chaitanya and Hu, Yuan-Ting and Bolya, Daniel and Wei, Chen and Fan, Haoqi and Huang, Po-Yao and Aggarwal, Vaibhav and Chowdhury, Arkabandhu and Poursaeed, Omid and Hoffman, Judy and Malik, Jitendra and Li, Yanghao and Feichtenhofer, Christoph , month = jun, year =. Hiera:

  37. [38]

    Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , editor =. U-. Medical. 2015 , keywords =. doi:10.1007/978-3-319-24574-4_28 , abstract =

    doi:10.1007/978-3-319-24574-4_28 2015

  38. [39]

    Journal of Geophysical Research: Atmospheres , author =

    Cloud ice:. Journal of Geophysical Research: Atmospheres , author =. 2009 , note =. doi:10.1029/2008JD010015 , abstract =

    doi:10.1029/2008jd010015 2009

  39. [40]

    Journal of Geophysical Research: Atmospheres , author =

    A global survey of the instantaneous linkages between cloud vertical structure and large-scale climate , volume =. Journal of Geophysical Research: Atmospheres , author =. 2014 , note =. doi:10.1002/2013JD020669 , abstract =

    doi:10.1002/2013jd020669 2014

  40. [41]

    Nature Geosci , author =

    Clouds, circulation and climate sensitivity , volume =. Nature Geosci , author =. 2015 , note =. doi:10.1038/ngeo2398 , abstract =

    doi:10.1038/ngeo2398 2015

  41. [42]

    Journal of Climate , author =

    Radiative. Journal of Climate , author =. 2000 , note =. doi:10.1175/1520-0442(2000)013<0264:REOCTV>2.0.CO;2 , abstract =

    doi:10.1175/1520-0442(2000)013 2000

  42. [43]

    Atmospheric Chemistry and Physics , author =

    Assessing observed and modelled spatial distributions of ice water path using satellite data , volume =. Atmospheric Chemistry and Physics , author =. 2011 , pages =. doi:10.5194/acp-11-375-2011 , number =

    doi:10.5194/acp-11-375-2011 2011

  43. [44]

    Atmospheric Chemistry and Physics , author =

    An update on global atmospheric ice estimates from satellite observations and reanalyses , volume =. Atmospheric Chemistry and Physics , author =. 2018 , pages =. doi:10.5194/acp-18-11205-2018 , number =

    doi:10.5194/acp-18-11205-2018 2018

  44. [45]

    and Pfreundschuh, S

    Amell, A. and Pfreundschuh, S. and Eriksson, P. , year =. The. doi:10.5194/egusphere-2023-1953 , journal =

    doi:10.5194/egusphere-2023-1953 2023

  45. [46]

    Atmospheric Measurement Techniques , author =

    Ice water path retrievals from. Atmospheric Measurement Techniques , author =. 2022 , pages =. doi:10.5194/amt-15-5701-2022 , number =

    doi:10.5194/amt-15-5701-2022 2022

  46. [47]

    Atmospheric Measurement Techniques , author =

    A neural network approach to estimating a posteriori distributions of. Atmospheric Measurement Techniques , author =. 2018 , pages =. doi:10.5194/amt-11-4627-2018 , number =

    doi:10.5194/amt-11-4627-2018 2018

  47. [48]

    Surv Geophys , author =

    Lessons. Surv Geophys , author =. 2024 , keywords =. doi:10.1007/s10712-024-09824-0 , abstract =

    doi:10.1007/s10712-024-09824-0 2024

  48. [49]

    IEEE Trans. Geosci. Remote Sensing , author =. 2021 , pages =. doi:10.1109/TGRS.2020.3015155 , abstract =

    arXiv 2021

  49. [50]

    Journal of Advances in Modeling Earth Systems , author =

    Toward an object-based assessment of high-resolution forecasts of long-lived convective precipitation in the central. Journal of Advances in Modeling Earth Systems , author =. 2015 , note =. doi:10.1002/2015MS000497 , abstract =

    doi:10.1002/2015ms000497 2015

  50. [51]

    Convolutional

    SHI, Xingjian and Chen, Zhourong and Wang, Hao and Yeung, Dit-Yan and Wong, Wai-kin and WOO, Wang-chun , year =. Convolutional. Advances in

  51. [52]

    Remote Sensing , author =

    Precipitation. Remote Sensing , author =. 2022 , note =. doi:10.3390/rs14132992 , abstract =

    doi:10.3390/rs14132992 2022

  52. [53]

    Google Docs , month = nov, year =

    Improved temporal and spectral merging of satellite observations for precipitation cloud-property retrievals , url =. Google Docs , month = nov, year =

  53. [54]

    https://www.nasa.gov/wp-content/uploads/2023/09/fy-22-strategic-plan-1.pdf?emrc=ff1a1e , url =

    2023

  54. [55]

    Bulletin of the American Meteorological Society , author =

    The. Bulletin of the American Meteorological Society , author =. 2021 , note =. doi:10.1175/BAMS-D-20-0299.1 , abstract =

    doi:10.1175/bams-d-20-0299.1 2021

  55. [56]

    Journal of Geophysical Research: Atmospheres , author =

    A. Journal of Geophysical Research: Atmospheres , author =. 2021 , note =. doi:10.1029/2020JD034202 , abstract =

    doi:10.1029/2020jd034202 2021

  56. [57]

    Journal of Geophysical Research: Atmospheres , author =

    A. Journal of Geophysical Research: Atmospheres , author =. 2019 , note =. doi:10.1029/2019JD030449 , abstract =

    doi:10.1029/2019jd030449 2019

  57. [58]

    Clim Dyn , author =

    Special issue:. Clim Dyn , author =. 2020 , pages =. doi:10.1007/s00382-020-05240-3 , language =

    doi:10.1007/s00382-020-05240-3 2020

  58. [59]

    A review on regional convection‐permitting climate modeling:

  59. [60]

    2024 , doi =

    Read ". 2024 , doi =

    2024

  60. [61]

    and Bolvin, David T

    Huffman, George J. and Bolvin, David T. and Braithwaite, Dan and Hsu, Kuo-Lin and Joyce, Robert J. and Kidd, Christopher and Nelkin, Eric J. and Sorooshian, Soroosh and Stocker, Erich F. and Tan, Jackson and Wolff, David B. and Xie, Pingping , editor =. Integrated. Satellite. 2020 , doi =

    2020

  61. [62]

    Personal communication , author =

  62. [63]

    doi:10.25921/stkw-7w73 , urldate =

    U. doi:10.25921/stkw-7w73 , urldate =

    doi:10.25921/stkw-7w73

  63. [64]

    2021 , note =

    Geophysical Research Letters , author =. 2021 , note =. doi:10.1029/2020GL092032 , abstract =

    doi:10.1029/2020gl092032 2021

  64. [65]

    2024 , note =

    Atmospheric Measurement Techniques , author =. 2024 , note =. doi:10.5194/amt-17-515-2024 , abstract =

    doi:10.5194/amt-17-515-2024 2024

  65. [66]

    and Becker, A

    Schneider, U. and Becker, A. and Finger, P. and Meyer-Christoffer, A. and Rudolf, B. and Ziese, M. , year =

  66. [67]

    Schmude, Johannes and Roy, Sujit and Trojak, Will and Jakubik, Johannes and Civitarese, Daniel Salles and Singh, Shraddha and Kuehnert, Julian and Ankur, Kumar and Gupta, Aman and Phillips, Christopher E. and Kienzler, Romeo and Szwarcman, Daniela and Gaur, Vishal and Shinde, Rajat and Lal, Rohit and Silva, Arlindo Da and Diaz, Jorge Luis Guevara and Jone...

    Pith/arXiv arXiv doi:10.48550/arxiv.2409.13598

  67. [68]

    Multiscale

    Guilloteau, Clément and Foufoula-Georgiou, Efi , editor =. Multiscale. Satellite. 2020 , doi =

    2020

  68. [69]

    Atmospheric Measurement Techniques , author =

    An improved near-real-time precipitation retrieval for. Atmospheric Measurement Techniques , author =. 2022 , note =. doi:10.5194/amt-15-6907-2022 , abstract =

    doi:10.5194/amt-15-6907-2022 2022

  69. [70]

    Tran, Du and Wang, Heng and Torresani, Lorenzo and Ray, Jamie and LeCun, Yann and Paluri, Manohar , month = apr, year =. A. doi:10.48550/arXiv.1711.11248 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1711.11248

  70. [71]

    Bulletin of the American Meteorological Society , author =

    The. Bulletin of the American Meteorological Society , author =. 2014 , note =. doi:10.1175/BAMS-D-13-00164.1 , abstract =

    doi:10.1175/bams-d-13-00164.1 2014

  71. [72]

    Journal of Applied Meteorology and Climatology , author =

    A. Journal of Applied Meteorology and Climatology , author =. 1988 , note =. doi:10.1175/1520-0450(1988)027<0030:ASITTE>2.0.CO;2 , abstract =

    doi:10.1175/1520-0450(1988)027 1988

  72. [73]

    Monthly Weather Review , author =

    Rain. Monthly Weather Review , author =. 1978 , note =. doi:10.1175/1520-0493(1978)106<1153:REFGSI>2.0.CO;2 , abstract =

    doi:10.1175/1520-0493(1978)106 1978

  73. [74]

    Journal of Hydrometeorology , author =

    Integrating. Journal of Hydrometeorology , author =. 2023 , note =. doi:10.1175/JHM-D-23-0006.1 , abstract =

    doi:10.1175/jhm-d-23-0006.1 2023

  74. [75]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , author =

    The. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , author =. 2015 , keywords =. doi:10.1109/JSTARS.2015.2403303 , abstract =

    arXiv 2015

  75. [76]

    and Weng, Fuzhong , editor =

    Goldberg, Mitchell D. and Weng, Fuzhong , editor =. Advanced. Earth. 2006 , doi =

    2006

  76. [77]

    Bulletin of the American Meteorological Society , author =

    Multi-. Bulletin of the American Meteorological Society , author =. 2016 , note =. doi:10.1175/BAMS-D-14-00173.1 , abstract =

    doi:10.1175/bams-d-14-00173.1 2016

  77. [78]

    Journal of the Atmospheric Sciences , author =

    The. Journal of the Atmospheric Sciences , author =. 2007 , note =. doi:10.1175/2006JAS2375.1 , abstract =

    doi:10.1175/2006jas2375.1 2007

  78. [79]

    Bulletin of the American Meteorological Society , author =

    So,. Bulletin of the American Meteorological Society , author =. 2017 , note =. doi:10.1175/BAMS-D-14-00283.1 , abstract =

    doi:10.1175/bams-d-14-00283.1 2017

  79. [80]

    Hastings, D. A. (David A. ) and Dunbar, Paula K. , month = aug, year =. Global

  80. [81]

    Journal of Hydrology , author =

    Inter-comparison of radar rainfall rate using. Journal of Hydrology , author =. 2015 , keywords =. doi:10.1016/j.jhydrol.2015.08.063 , abstract =

    doi:10.1016/j.jhydrol.2015.08.063 2015

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.