Multi-Source Weighted-Ensemble Precipitation


MSWEP is a glob¬≠al pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠uct with a 3‚ÄĎhourly 0.1¬į res¬≠o¬≠lu¬≠tion avail¬≠able from 1979 to ~3 hours from real-time. The prod¬≠uct is unique in that it merges gauge, satel¬≠lite, and reanaly¬≠sis data to obtain the high¬≠est qual¬≠i¬≠ty pre¬≠cip¬≠i¬≠ta¬≠tion esti¬≠mates at every location.
MSWEP incor­po­rates dai­ly gauge obser­va­tions and accounts for gauge report­ing times to reduce tem­po­ral mis­match­es between satel­lite-reanaly­sis esti­mates and gauge obser­va­tions. Near real-time esti­mates are avail­able with a laten­cy of ~3 hours. MSWEP is com­pat­i­ble with GloH2O’s oper­a­tional Mul­ti-Source Weath­er (MSWX); MSWX fore­casts can thus be used to extend MSWEP into the future. MSWEP tends to exhib­it bet­ter per­for­mance than oth­er pre­cip­i­ta­tion prod­ucts in both dense­ly gauged and ungauged regions (see the Per­for­mance sec­tion on this page, Beck et al., 2017, and Beck et al., 2019).


Based on state-of-the-art gauge, satel¬≠lite, and reanaly¬≠sis data sources 


Dai¬≠ly cor¬≠rec¬≠tions using obser¬≠va¬≠tions from over 77,000 gauges 

Near real-time 

Near real-time avail¬≠abil¬≠i¬≠ty to sup¬≠port time-crit¬≠i¬≠cal applications 


Meth¬≠ods pub¬≠lished in open-access peer-reviewed sci¬≠en¬≠tif¬≠ic journals 


Mean tem¬≠po¬≠ral cor¬≠re¬≠la¬≠tion (R2) against two ref¬≠er¬≠ences for MSWEP and oth¬≠er wide¬≠ly used pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠ucts, demon¬≠strat¬≠ing the high¬≠er over¬≠all accu¬≠ra¬≠cy of MSWEP in dense¬≠ly gauged and ungauged regions, respectively. 

Reference: Stage-IV gauge-radar data

MSWEP V2.2 74%
IMERG Final V6 58%
GSMaP Std V7 49%
ERA-Inter­im 42%
CPC Uni­fied 37%
TMPA 3B42 V7 34%

Val­ues rep­re­sent the mean dai­ly tem­po­ral R2 over the US
Adapt­ed from Beck et al. (2019)

Reference: 75,540 gauges worldwide 

MSWEP* V2.8 48%
GPM+SM2R V0.1.0 41%
IMERG Late V6 32%
GSMaP Std V7 31%

Val¬≠ues rep¬≠re¬≠sent the mean 3‚ÄĎday tem¬≠po¬≠ral R2
*Ver­sion with­out gauge cor­rec­tions
Pub­li­ca­tion in preparation


The fol¬≠low¬≠ing paper pro¬≠vides a detailed descrip¬≠tion of the MSWEP V2.2 methodology: 
MSWEP V2.8 fea¬≠tures new data sources, improved weight maps, less peaky pre¬≠cip¬≠i¬≠ta¬≠tion esti¬≠mates, a longer record (1979‚Äď2020), near real-time (NRT) esti¬≠mates, and com¬≠pat¬≠i¬≠bil¬≠i¬≠ty with Mul¬≠ti-Source Weath¬≠er (MSWX). See the tech¬≠ni¬≠cal doc¬≠u¬≠men¬≠ta¬≠tion for the ver¬≠sion his¬≠to¬≠ry and more infor¬≠ma¬≠tion on MSWEP-NRT, the file nam¬≠ing con¬≠ven¬≠tion, and the data for¬≠mat. See the sys¬≠tem sta¬≠tus page for near real-time infor¬≠ma¬≠tion on the avail¬≠abil¬≠i¬≠ty and per¬≠for¬≠mance of MSWEP-NRT. 

Data license

MSWEP is released under the Cre¬≠ative Com¬≠mons Attri¬≠bu¬≠tion-Non¬≠Com¬≠mer¬≠cial 4.0 Inter¬≠na¬≠tion¬≠al (CC BY-NC 4.0) license. Please con¬≠tact us if you are affil¬≠i¬≠at¬≠ed with a com¬≠mer¬≠cial enti¬≠ty and want to tri¬≠al MSWEP. If you do not have a com¬≠mer¬≠cial affil¬≠i¬≠a¬≠tion and you intend to use the prod¬≠uct for non-com¬≠mer¬≠cial pur¬≠pos¬≠es, please send us a request using the fol¬≠low¬≠ing form. You will receive a link to the Google Dri¬≠ve con¬≠tain¬≠ing MSWEP once your request has been approved. 

Frequently asked questions

Numer¬≠ous grid¬≠ded pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠ucts have been devel¬≠oped over the last decades. How¬≠ev¬≠er, MSWEP is the only product: 
  1. to merge gauge, satel­lite, and reanaly­sis pre­cip­i­ta­tion esti­mates to enhance the per­for­mance in dense­ly gauged, con­vec­tion-dom­i­nat­ed, and frontal-dom­i­nat­ed weath­er regimes, respectively;
  2. to incor­po­rate dai­ly gauge obser­va­tions and account for gauge report­ing times (which min­i­mizes tem­po­ral mis­match­es between the satel­lite-reanaly­sis esti­mates and the gauge observations);
  3. with com­pat­i­ble medi­um-range fore­cast ensem­bles (updat­ed dai­ly) and sea­son­al fore­cast ensem­bles (updat­ed month­ly; see MSWX).
In addi¬≠tion, MSWEP has a high 3‚ÄĎhourly 0.1¬į res¬≠o¬≠lu¬≠tion, glob¬≠al cov¬≠er¬≠age, a long record start¬≠ing in 1979, and near real-time esti¬≠mates. MSWEP also tends to exhib¬≠it bet¬≠ter per¬≠for¬≠mance than oth¬≠er pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠ucts in both dense¬≠ly gauged and ungauged regions (see the Per¬≠for¬≠mance sec¬≠tion on this page, Beck et al., 2017, and Beck et al., 2019).

The ‚ÄėPast‚Äô and ‚ÄėPast_nogauge‚Äô vari¬≠ants rep¬≠re¬≠sent the his¬≠tor¬≠i¬≠cal satel¬≠lite-reanaly¬≠sis merge includ¬≠ing and exclud¬≠ing gauge cor¬≠rec¬≠tions, respec¬≠tive¬≠ly. The ‚ÄėNRT‚Äô vari¬≠ant rep¬≠re¬≠sents the near real-time exten¬≠sion of the his¬≠tor¬≠i¬≠cal record to the present (with a laten¬≠cy of ~3 hours). We rec¬≠om¬≠mend using the ‚ÄėPast_nogauge‚Äô vari¬≠ant in pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠uct per¬≠for¬≠mance eval¬≠u¬≠a¬≠tions using gauge obser¬≠va¬≠tions as ref¬≠er¬≠ence, and the ‚ÄėPast‚Äô vari¬≠ant for any oth¬≠er purpose. 

Down¬≠load¬≠ing data from shared Google Dri¬≠ve fold¬≠ers is rel¬≠a¬≠tive¬≠ly easy using cloud store man¬≠ag¬≠er soft¬≠ware such as rclone. The fol¬≠low¬≠ing instruc¬≠tions explain how to set up rclone to down¬≠load MSWEP: 
  1. Down­load and install rclone.
  2. Link rclone to your Google account by fol­low­ing the steps in this video.
  3. Access the MSWEP shared fold¬≠er by vis¬≠it¬≠ing it via your brows¬≠er. Check if the shared fold¬≠er is list¬≠ed under ‚ÄúShared with me‚ÄĚ on your Google Dri¬≠ve page.
  4. Con­firm that rclone can find the shared fold­er:
    $ rclone lsd --drive-shared-with-me GoogleDrive:
              -1 2021-02-03 10:14:35        -1 MSWEP_V280
  5. Down­load dai­ly and month­ly MSWEP data from the shared fold­er to your local dri­ve:
    $ rclone sync -v --exclude 3hourly/ --drive-shared-with-me GoogleDrive:/MSWEP_V280 c:/temp/MSWEP_V280
              2021/02/03 11:09:02 INFO  : Past/Monthly/ Copied (new)
              2021/02/03 11:09:02 INFO  : Past/Monthly/ Copied (new)
              2021/02/03 11:09:02 INFO  : Past/Monthly/ Copied (new)
    If the down­load is inter­rupt­ed, the com­mand can be run again, and files that already exist will be skipped.

Medi¬≠um-range and sea¬≠son¬≠al fore¬≠cast data from GloH2O‚Äôs Mul¬≠ti-Source Weath¬≠er (MSWX) prod¬≠uct is com¬≠pat¬≠i¬≠ble with MSWEP. MSWX can thus be used to extend MSWEP into the future. Note, how¬≠ev¬≠er, that some incon¬≠sis¬≠ten¬≠cies between MSWEP and MSWX may be present due to the inclu¬≠sion of gauge and satel¬≠lite data in MSWEP.

Pre­cip­i­ta­tion data for a par­tic­u­lar day (e.g., April 25, 2020) can be down­loaded using rclone as fol­lows:
rclone sync -v --drive-shared-with-me GoogleDrive:/MSWEP_V280/Past/Daily/ ./
The data are read and plot using MATLAB as fol­lows:
global_precip = ncread('','precipitation')';
imagesc(global_precip,[0 30]);
title('Precipitation on April 25, 2020 (mm/day)')
The same data are read and plot using Python as fol­lows:
from netCDF4 import Dataset
import matplotlib.pyplot as plt

dataset = Dataset('','r')
global_precip = dataset.variables['precipitation'][:]

plt.title("Precipitation on April 25, 2020 (mm/day)")

We did not cor¬≠rect for gauge under-catch in the lat¬≠est ver¬≠sion of MSWEP (unlike in pre¬≠vi¬≠ous ver¬≠sions) for two main rea¬≠sons: (i) in our expe¬≠ri¬≠ence, users tend to give pref¬≠er¬≠ence to pre¬≠cip¬≠i¬≠ta¬≠tion esti¬≠mates that match gauge obser¬≠va¬≠tions as close¬≠ly as pos¬≠si¬≠ble; and (ii) under-catch cor¬≠rec¬≠tions are gen¬≠er¬≠al¬≠ly sub¬≠ject to con¬≠sid¬≠er¬≠able uncer¬≠tain¬≠ty. To obtain a bias-cor¬≠rect¬≠ed ver¬≠sion of MSWEP ‚ÄĒ impor¬≠tant for hydro¬≠log¬≠i¬≠cal mod¬≠el¬≠ing ‚ÄĒ we rec¬≠om¬≠mend using GloH2O‚Äôs PBCOR product.

MSWEP V1 lacks cumu­la­tive dis­tri­b­u­tion func­tion (CDF) cor­rec­tions, and there­fore exhibits sig­nif­i­cant driz­zle and less reli­able trends due to changes in the data sources through time. Con­verse­ly, MSWEP V2 match­es the CDF of each dataset com­bi­na­tion to a sin­gle ref­er­ence CDF, result­ing in less driz­zle and a more homo­ge­neous record. See the tech­ni­cal doc­u­men­ta­tion for the full list of changes.

Pre¬≠cip¬≠i¬≠ta¬≠tion prod¬≠uct eval¬≠u¬≠a¬≠tions car¬≠ried out at the dai¬≠ly time scale should account for the report¬≠ing times of gauges (i.e., the end times of the dai¬≠ly accu¬≠mu¬≠la¬≠tions). Dis¬≠re¬≠gard¬≠ing report¬≠ing times in dai¬≠ly eval¬≠u¬≠a¬≠tions can result in sig¬≠nif¬≠i¬≠cant tem¬≠po¬≠ral mis¬≠match¬≠es between the prod¬≠uct and the gauge, which can affect the results of the eval¬≠u¬≠a¬≠tion. Account¬≠ing for report¬≠ing times requires the use of sub-dai¬≠ly data to com¬≠pute new dai¬≠ly accu¬≠mu¬≠la¬≠tions that cor¬≠re¬≠spond to the report¬≠ing times of the gauges. Alter¬≠na¬≠tive¬≠ly, the eval¬≠u¬≠a¬≠tion can be car¬≠ried out at a 3‚ÄĎday (instead of dai¬≠ly) time scale, which reduces the impact of poten¬≠tial mis¬≠match¬≠es. See Beck et al. (2019) for maps of report¬≠ing times for GHCN‚ÄĎD and GSOD gauges.

There is cur­rent­ly no oth­er way to down­load MSWEP except using rclone.


ECMWF, NASA, and NOAA and thanked for pro¬≠duc¬≠ing the ERA5, IMERG, and Grid¬≠Sat datasets, respec¬≠tive¬≠ly. The Water Cen¬≠ter for Arid and Semi-Arid Zones in Latin Amer¬≠i¬≠ca and the Caribbean (CAZALAC) and the Cen¬≠tro de Cien¬≠cia del Cli¬≠ma y la Resilien¬≠cia (CR) 2 (FONDAP 15110009) are thanked for shar¬≠ing Mex¬≠i¬≠can and Chilean gauge data, respec¬≠tive¬≠ly. We also acknowl¬≠edge the gauge data providers in the Latin Amer¬≠i¬≠can Cli¬≠mate Assess¬≠ment and Dataset (LACA&D) project: IDEAM (Colom¬≠bia), INAMEH (Venezuela), INAMHI (Ecuador), SENAMHI (Peru), SENAMHI (Bolivia), and DMC (Chile). We wish to thank Ali Ali¬≠jan¬≠ian, Koen Ver¬≠bist, and Piyush Jain for pro¬≠vid¬≠ing addi¬≠tion¬≠al gauge data.