GloH2O

Menu

MSWEP

Multi-Source Weighted-Ensemble Precipitation

MSWEP provides hourly 0.1° resolution precip­i­tation data from1979 to the present with a 2‑hour delay. 

Superior accuracy 

MSWEP’s sophis­ti­cated machine learning algorithm brings together data from gauges, satel­lites, and models, refining it into the most accurate precip­i­tation estimates available. No matter where the data comes from, the system is designed to deliver excep­tional accuracy, so you get reliable insights every time, wherever you are. 

Gauge-corrected 

With correc­tions based on daily and monthly obser­va­tions from over 70,000 gauges worldwide, MSWEP ensures that your data is not just theoretical — it’s anchored in real-world condi­tions. This minimizes discrep­ancies between satellite and model data and local obser­va­tions, giving you confi­dence in your results. 

Near real-time 

MSWEP delivers the first precip­i­tation estimates with a latency of less than 2 hours, giving you access to the first precip­i­tation estimates almost instantly. As new data sources become available, MSWEP contin­u­ously updates and refines these estimates in near real-time, ensuring you’re always equipped with the best information. 

Last week’s MSWEP precipitation

Global precip­i­tation patterns over the past 7 days, with time expressed in UTC. Each frame repre­sents the hourly precip­i­tation accumu­lation, and the video is updated every 3 hours. 

Proven performance

We conducted the most extensive evaluation of gridded precipitation products to date.

In our compre­hensive peer-reviewed study — freely acces­sible here (in prep) — MSWEP emerged as best overall compared to other global precip­i­tation products.

The bar chart on the right presents median daily Kling-Gupta Efficiency (KGE) scores from over 12,000 independent, quality-controlled rain gauges worldwide. KGE scores provides a useful summary of how well each product reflects real-world condi­tions by capturing key factors like timing, variability, and bias.

MSWEP versus other datasets

MSWEP-NRT V3 77%
MSWEP-Past V3* 69%
ERA5* 61%
JRA-3Q* 57%
SM2RAIN-SMOS-RAINFALL* 52%
IMERG-Late V07 46%
GSMaP-Std v8 38%
CMORPH RAW 33%
PDIR-Now 33%
SM2RAIN-ASCAT* 33%
CHIRP* 31%
CMORPH RT 28%
P‑CCSCDR* 24%
P‑CCS* 18%
Performance comparison between MSWEP and other widely used precipitation datasets, based on independent rain gauge observations from across the globe. Values are rescaled median Kling Gupta Efficiency (KGE) scores, where a KGE of −0.41 (equivalent to using the average observed value) is set to 0, and a perfect KGE of 1 is set to 1. Datasets marked with * are not available in near real‑time.

MSWEP versus ERA5

Our dataset surpasses ERA5 in all evaluated metrics

1 %
Better KGE
score
1 %
Lower peak
bias
1 %
Lower wet-
day bias
1 %
Better event detection

Algorithm

MSWEP leverages machine learning to merge and bias-correct satellite and model data.

The following openly acces­sible, peer-reviewed paper provides a detailed description of the MSWEP V3 algorithm: 
Wang, X., Alharbi, R. S., Baez-Villanueva, O. M., Miralles, D. G., Ma, J., Xu, S., McCabe, M. F., Pappen­berger, F., van Dijk, A. I. J. M., McVicar, T. R., Karthikeyan, L., Fowler, H. J., Pan, M., Gebre­chorkos, S. H., and Beck, H. E. (2026). MSWEP V3: Machine learning-powered global precip­i­tation estimates at 0.1° hourly resolution (1979–present). arXiv preprint arXiv:2602.01436:, doi:10.48550/arXiv.2602.01436.
Along with a new algorithm, MSWEP V3 incor­po­rates new data sources, reduces drizzle overes­ti­mation, and minimizes peak under­es­ti­mation compared to V2. For details on the version history, file naming, data formats, and the near real-time release schedule, see the documentation. Previous version of MSWEP are described in the following papers: 
Beck, H. E., Wood, E. F., Pan, M., Fisher, C. K., Miralles, D. G., Van Dijk, A. I. J. M., McVicar, T. R., and Adler, R. F. (2019). MSWEP v2 Global 3‑hourly 0.1° precip­i­tation: Method­ology and quanti­tative assessment. Bulletin of the American Meteo­ro­logical Society, 100(3), 473–500.
Beck, H. E., van Dijk, A. I. J. M., Levizzani, V., Schellekens, J., Miralles, D. G., Martens, B., and de Roo, A. (2017). MSWEP: 3‑hourly 0.25° global gridded precip­i­tation (1979–2015) by merging gauge, satellite, and reanalysis data. Hydrology and Earth System Sciences, 21, 589–615.

Data access

MSWEP is released under the CC BY-NC 4.0 license and can be used for academic research, nonprofit scien­tific studies, personal projects, and certain government or NGO appli­ca­tions. To request access, simply submit a request, and we’ll email you download instruc­tions.

If you’ve requested before and want the latest version, submit again using a unique email alias by adding + and a new tag before the @ each time (e.g., name+abc123@domain.com).

For questions, read the MSWEP documentation or contact us.

Who’s Using MSWEP?

Leading organi­za­tions relying on MSWEP in their analyses. 
anu_logo-1
European_Commission.svg-1
exeter_logo-1
ghent_logo-1
iiasa_logo-1
KIT-1
KU_Leuven_logo.svg-1
National-Geographic-Logo-1
princeton-logo_1-1
TH_Koeln_Logo-1
tudelft_logo-1
UC_Davis_Logo-1
unesco_logo-1
university-of-melbourne-logo-1
unsw_logo_-1
wageningen_logo-1
wsl-1
nasa_logo-2

Frequently asked questions

Over the years, many gridded precip­i­tation products have been developed, but MSWEP stands out as the only one that merges gauge, satellite, and model precip­i­tation estimates, which enhances its perfor­mance in densely gauged, convection-dominated, and frontal-dominated regions, respec­tively. Furthermore, MSWEP incor­po­rates daily (in addition to monthly) gauge obser­va­tions and accounts for gauge reporting times, reducing temporal mismatches between satellite-model estimates and gauge data. With its high hourly 0.1° resolution, global coverage, historical data starting from 1979, and near real-time updates, MSWEP consis­tently outper­forms other precip­i­tation products (see the “Proven Perfor­mance” section on this page).

The ‘Past’ and ‘Past_nogauge’ variants represent the historical satellite-model merged data, with the ‘Past’ variant including gauge correc­tions and the ‘Past_nogauge’ variant excluding them. The ‘NRT’ variant extends the historical record to near real-time, with a latency of approx­i­mately 3 hours. For perfor­mance evalu­a­tions using gauge obser­va­tions, we recommend the ‘Past_nogauge’ variant, while the ‘Past’ variant is more suitable for general use. For more details, see the documentation.

  • Download the appro­priate Rclone binary for your operating system from the Rclone website.
  • Extract the files into a folder on your hard drive.
  • Open a terminal or command prompt and navigate to the folder using the cd command.
  • Configure Rclone by running: rclone config
  • Press n to create a new remote.
  • Enter the name GoogleDrive.
  • For the storage type, enter drive (for Google Drive).
  • Press Enter to accept the default choices (client_id, client_secret, scope, and service_account_file).
  • Press n for advanced config.
  • Press y to use web browser.
  • Authen­ticate Rclone using your web browser.
  • Press n for Shared Drive (Team Drive).
  • Press y to keep the remote.
  • Press q to exit the config­u­ration menu.
  • Access the MSWEP shared Google Drive folder by clicking the link in the email.
  • To confirm that Rclone can locate the shared folder, run: rclone lsd --drive-shared-with-me GoogleDrive:
  • To download daily and monthly MSWEP data from the shared folder to your local drive, use: rclone sync -v --exclude 3hourly/ --exclude Hourly/ --drive-shared-with-me GoogleDrive:/MSWEP_V315_test c:/temp/MSWEP_V315_test

We have not corrected gauge under-catch in the latest version of MSWEP (V3) for two main reasons: (i) users typically prefer precip­i­tation estimates that align closely with rain gauge obser­va­tions, and (ii) under-catch correc­tions often come with consid­erable uncer­tainty. For a bias-corrected version of MSWEP, partic­u­larly useful for hydro­logical modeling, we recommend using GloH2O’s PBCOR product.

MSWEP V3 intro­duces a cutting-edge machine learning algorithm for merging and bias-correcting satellite and model data. Along with this new algorithm, V3 includes additional data sources, reduces drizzle overes­ti­mation, and minimizes peak under­es­ti­mation compared to V2. For a complete overview of version history, file naming conven­tions, and data formats, please refer to the documentation.

Daily precip­i­tation evalu­a­tions should consider gauge reporting times (i.e., the end time of daily accumu­la­tions) to avoid temporal mismatches between the product and the gauge data. Ignoring reporting times can skew evalu­ation results. To mitigate this, sub-daily data can be used to compute new daily accumu­la­tions that align with the gauges’ reporting times. Alter­na­tively, evalu­a­tions conducted over a 3‑day period can help reduce the impact of these mismatches. For more details, see the global reporting time map in Wang et al. (in prep),

Currently, noncom­mercial users can download MSWEP exclu­sively via Rclone, as outlined in the email received upon requesting access. Alter­native download methods, such as FTP and API access, are available only to commercial users.

Easily access MSWEP using the GloH2O API

The GloH2O API offers seamless access to a wide range of meteo­ro­logical datasets, including MSWEP, for use in commercial applications. 
GloH2O API