Land surface precipitation and hydrology in MERRA-2

Land surface precipitation and hydrology in MERRA-2 R. Reichle, R. Koster, C. Draper, Q. Liu, M. Girotto, S. Mahanama, G. De Lannoy, G. Partyka, and many others 5th International Conference on Reanalysis 13 17 November 2017, Rome

MERRA Model precip. Land Surface in MERRA Products MERRA-2 Model precip. (M2AGCM) MERRA-Land Observations Observations Corr. precip. (M2CORR) AGCM LSM + updated land model AGCM + updated AGCM and atmospheric analysis 2

Outline 1. Precipitation Corrections and Evaluation 2. Evaluation of Land Surface Hydrology 3

MERRA-Land Precipitation Corrections Land surface precipitation corrected to CPCU gauge product everywhere. Separately for each day / 0.5 grid cell. Sub-daily variations from MERRA. 4

MERRA-2 Precipitation Corrections daily 0.5 pentad 2.5 Land surface precipitation corrected to observations-based products except at high latitudes. Separately for each day / 0.5 grid cell (CPCU) or pentad / 2.5 grid cell (CMAP). Sub-daily/pentad variations from MERRA (through Feb 2016) and GEOS FP-IT thereafter. 5

Time Series Correlation (vs. GPCPv2.2) R( M2CORR ) R( M2CORR ) minus R( M2AGCM ) avg = 0.82 avg = 0.09 MERRA-2 corrected precipitation: agrees with GPCPv2.2 in well-observed regions & is better than model precipitation. Similar results for RMSE and anomaly correlation. 6

Time Series Correlation (vs. GPCPv2.2) R( M2CORR ) R( M2CORR ) minus R( M2AGCM ) avg = 0.82 avg = 0.09 MERRA-2 corrected precipitation: agrees with GPCPv2.2 in well-observed regions & Error in CPCU gauge product! is better than model precipitation. Similar results for RMSE and anomaly correlation. 7

Time Series Correlation (vs. GPCPv2.2) R( M2CORR ) R( M2CORR ) minus R( M2AGCM ) avg = 0.82 avg = 0.09 R( M2CORR ) minus R( MERRA-Land ) MERRA-2 corrected precipitation also better than MERRA-Land. avg = 0.02 8

Observing System Impacts # Gauges No obvious impact from observing system. Corrected precipitation impacted by change in gauges. Model precipitation impacted by change in atmospheric analysis (+AMSU). 9

TRMM MERRA-2 corrected precipitation MERRA-2 model precipitation Amplitude Diurnal Cycle MERRA-2 corrected precipitation inherits diurnal cycle from MERRA. The diurnal cycle of the MERRA-2 corrected precipitation has better amplitude than MERRA-2 model precipitation. 10

Amplitude Diurnal Cycle Phase MERRA-2 corrected TRMM precipitation inherits diurnal cycle from MERRA. MERRA-2 The diurnal cycle of the corrected precipitation MERRA-2 corrected precipitation has better amplitude and MERRA-2 model precipitation worse phase than MERRA-2 model precipitation. 11

Outline 1. Precipitation Corrections and Evaluation 2. Evaluation of Land Surface Hydrology 12

Terrestrial Water Storage (vs. GRACE) 0 1 MERRA-2 monthly TWS correlates better with GRACE than TWS from MERRA and MERRA-Land. Similar for time series anomalies (not shown). 13-0.5 0.5 MERRA-2 worse MERRA-2 better

Soil Moisture (vs. In Situ) MERRA-2 soil moisture skill is similar to that of MERRA-Land, slightly better than that of ERA-Interim/Land, and significantly better than that of MERRA. 14

Streamflow (vs. Naturalized Gauge Obs.) MERRA-2 streamflow anomaly R is better than that of MERRA and similar to that of land-only products. MERRA-2 runoff still biased low (not shown). 15

Snow (SWE vs. CMC, SCA vs. MODIS) SWE bias vs. CMC MERRA-2 (0.4 mm) MERRA-2 slightly overestimates SWE but under-estimates SCA (because of a snow model parameter change). Snow Cover Area [-] [mm] MERRA-Land (-9.7 mm) MERRA (-3.4 mm) ERA-Int/Land (14.1 mm) Distance from observation 16

Consistency of Land Surface Forcing R( 4-day-avg precip, Tair on 4 th day) for JJA. Correcting precipitation within the coupled land-atmosphere system results in higher consistency of land forcing. 17

National Aeronautics and Space Administration Impact of Precipitation Corrections on T2mmax R2anom( Precip, T2mmax ) Ranom(MERRA-2, CRU) Ranom(MERRA, CRU) Model precip Corr. precip Difference (bottom minus top) MERRA-2 improvement in T2mmax vs. CRU. T2mmax variance explained by precip Sensitivity of MERRA-2 T2mmax (for JJA). to precipitation corrections. 18

Summary Land surface precipitation in MERRA-2 is corrected with observations. Precipitation corrections algorithm is an extension of that from MERRA-Land with o o a different observational product in Africa and no corrections at high latitudes. MERRA-2 precipitation, terrestrial water storage, soil moisture, and runoff agree better with measurements or reference data than same from MERRA. Snow model parameter change yields mixed results for MERRA-2 snow estimates. Precipitation corrections within the coupled land-atmosphere system o o facilitate more consistent land surface forcing compared to MERRA-Land, and improve simulated T2m compared to MERRA. Success critically depends on having high-quality global precipitation products with suitable latency. (Thanks to P. Xie et al. at NOAA CPC!) 19

Thank you for your attention! For details, see MERRA-2 Special Collection in J. Climate: Reichle et al. (2017a), Land surface precipitation in MERRA-2 doi:10.1175/jcli-d-16-0570.1 Reichle et al. (2017b), Assessment of MERRA-2 land surface hydrology estimates doi:10.1175/jcli-d-16-0720.1 Draper et al. (2017), Assessment of MERRA-2 Land Surface Energy Flux Estimates doi:10.1175/jcli-d-17-0121.1 20

EXTRA SLIDES 21

22

Precipitation Climatology (Mean) Mean precipitation diff. owing to corrections. MERRA-2 has more precipitation than MERRA-Land in central Africa (good) and in high latitudes (not so good). 23

24

R( M2CORR ) Time Series Correlation (vs. GPCPv2.2) R( M2CORR ) minus R( M2AGCM ) avg = 0.82 avg = 0.09 R( M2CORR ) minus R( MERRA-Land ) MERRA-2 corrected precipitation: agrees with GPCPv2.2 in well-observed regions & generally agrees better with GPCPv2.2 than model or MERRA-Land precipitation. Similar results for RMSE and anomaly correlation. avg = 0.02 25

Time Series Correlation (vs. GPCPv2.2 Precip.) Correlation Δcorrelation w.r.t. model precip. Δcorrelation w.r.t. MERRA-Land R Anom R Good agreement with GPCPv2.2 in well-observed regions. MERRA-2 corrected precipitation generally agrees better with GPCPv2.2 than model or MERRA-Land precipitation. 26

Time Series Correlation (vs. GPCPv2.2 Precip.) Correlation Δcorrelation w.r.t. model precip. Δcorrelation w.r.t. MERRA-Land R Anom R Good agreement with GPCPv2.2 in well-observed regions. MERRA-2 corrected precipitation generally agrees better with GPCPv2.2 than model or MERRA-Land precipitation. 27

Time Series Correlation (vs. GPCPv2.2 Precip.) Correlation Δcorrelation w.r.t. model precip. Δcorrelation w.r.t. MERRA-Land R Anom R Good agreement with GPCPv2.2 in well-observed regions. MERRA-2 corrected precipitation generally agrees better with GPCPv2.2 than model or MERRA-Land precipitation. 28

National Aeronautics and Space Administration 29

30

31

32

33

34

35

36

Soil Moisture (vs. In Situ) In terms of soil moisture skill, MERRA-2 is similar to MERRA-Land, slightly better than ERA-Interim/Land, and significantly better than MERRA. 37

38

Snow Water Equivalent (vs. CMC) Distance from observations. and over-estimates snow water equivalent. 39

Snow Cover (vs. MODIS) MERRA-2 under-estimates snow cover area because of a snow model parameter change 40

41

National Aeronautics and Space Administration MERRA-2 minus MERRA 42

National Aeronautics and Space Administration Snow Water Equivalent (vs. CMC) Distance from observations. 43

44

45

46

47

48