A Gauge-Satellite Merged Analysis A Gauge-Satellite Merged Analysis
of High-Resolution Global of High-Resolution Global
Precipitation Precipitation
Pingping Xie Pingping Xie
NOAA NOAA ’ ’ s Climate Prediction Center s Climate Prediction Center
acknowledgements acknowledgements
S.-H. Yoo, R. Joyce, and Y. Yarosh S.-H. Yoo, R. Joyce, and Y. Yarosh
2011.03.30.
2011.03.30.
Objective:
Objective:
• To develop a high-resolution gauge-satellite merged analysis of global precipitation
• Up to 8kmx8km over the globe (60
oS-60
oN)
• 30-min from Jan.1998, updated real time
• Through combining information from gauge- based analysis and CMORPH satellite
estimates
Global Daily Gauge Analysis Global Daily Gauge Analysis
• Interpolation of gauge reports from ~30K stations
• Optimal Interpolation (OI) with orographic correction (Xie et al. 2007)
• Interpolated on 0.125olat/lon, then averaged on 0.5o/0.25o lat/lon grid over global land / CONUS for release
• Global fields from1979 to present updated daily on a real-time basis
• CONUS analysis from 1948
• Example
for July 1, 2003
CMORPH Satellite Estimates CMORPH Satellite Estimates
CMORPH : CPC Morphing technique (Joyce et al. 2004)
Combined use of satellite PMW and IR data
8kmx8km / 60oS-60oN;
30-min interval / from March1998 / Real-time
CMORPH back-extended to 1998 to cover the entire TRMM Era
CMORPH Bias [1]
CMORPH Bias [1]
Global Distribution Global Distribution
• 2000-2009 10-yr annual mean precip
• CMORPH captures the spatial distribution patterns very well
• BIAS exists
– Over-estimates over tropical / sub-tropical areas
– Under-estimates over
mid- and hi-latitudes
CMORPH Bias [2]
CMORPH Bias [2]
Time Scales of the Bias Time Scales of the Bias
• Bias over CONUS
• Bias presents
substantial variations of
– seasonal (top),
– sub-monthly (middle), and
– year-to-year (bottom)
time scales
CMORPH Bias [3]
CMORPH Bias [3]
Range Dependence Range Dependence
• Bias as a function of CMORPH Rainfall Intensity over CONUS
• Bias exhibits strong range dependence
Bias Correction [1]
Bias Correction [1]
General Strategy General Strategy
• Seasonal / Year-to-year variations in bias
correction coefficients change with time
• Sub-monthly variations in bias
against sub-monthly gauge data
• Range-dependence in bias
PDF matching
Bias Correction [2]
Bias Correction [2]
PDF Bias Correction against daily gauge data PDF Bias Correction against daily gauge data
• PDF matching
• Collect co-located daily gauge and satellite data over a time / space domain
• Construct cumulated PDF tables for the gauge and satellite data
• Match the PDF of satellite data against that of gauge data to remove the bias
Bias Correction [3]
Bias Correction [3]
Global Implementation Strategy Global Implementation Strategy
• Step 1: Correction using Historical Data
– Establish PDF matching tables
– using historical data – for each 0.25olat/lon grid – for each calendar date
– using data over nearby regions and
– over a period of +/- 15 days centering at the target date
– At least 500 pairs of non-zero data pairs
– to ensure the PDF tables are created using data over a small space domain
• Step 2: Correction using Real-Time Data
– Perform PDF matching using data over a 30-day period ending at the target date
– To account for year-to-year variations in the bias
Bias Correction [4]
Bias Correction [4]
Results over Global Land Results over Global Land
2000-2009 annual mean
Large-scale bias corrected
Daily Gauge
KF-CMORPH
Gauge-Adjusted KF-CMORPH Comparison over Africa
Bias Correction [5]
Bias Correction [5]
Strategy over Ocean
Strategy over Ocean
Bias Correction [6]
Bias Correction [6]
Applications : Evaluation of CFSR JJA Precip.
Applications : Evaluation of CFSR JJA Precip.
Bias Correction [7]
Bias Correction [7]
Applications : Precip. Diurnal Cycle
Applications : Precip. Diurnal Cycle
Bias Correction [8]
Bias Correction [8]
Applications : Precip. Diurnal Cycle over Oceans
Applications : Precip. Diurnal Cycle over Oceans
Bias Correction [9]
Bias Correction [9]
Applications : Precip. Diurnal Cycle over Land
Applications : Precip. Diurnal Cycle over Land
Bias Correction [10]
Bias Correction [10]
Applications : Evaluations of MJO Precip in CFSR
Applications : Evaluations of MJO Precip in CFSR
Combining Gauge with Satellite [1]
Combining Gauge with Satellite [1]
• Combining bias-corrected satellite estimates with daily gauge over the several regions
– This is only possible for several regions due to different daily ending time in the gauge reports
• Africa (06Z)
• CONUS/MEX (12Z)
• S. America (12Z)
• Australia (00Z)
• China (00Z)
– Combining the bias-corrected CMORPH with gauge
observations through the Optimal Interpolation (OI) over selected regions where gauge observations have the same daily ending time
• in which the CMORPH and gauge data are used as the first guess and observations, respectively
Combining Gauge with Satellite [2]
Combining Gauge with Satellite [2]
Quantifying error in the inputs Quantifying error in the inputs
• Key to the construction of an OI- analysis is the definition of input errors
• Quantified error in the gauge and bias-corrected CMORPH through comparison against real data
• Error variance in gauge analysis
• Proportional to precip intensity
• Inversely proportional to local gauge density
• Error variance in bias-crtd CMORPH
• Proportional to precip intensity
• Correlation between CMORPH error at two different grid boxes
• Decreases exponentially with separation distance
Combining Gauge with Satellite [3]
Combining Gauge with Satellite [3]
Example for Pakistan Flooding Example for Pakistan Flooding
• Gauge analysis depict heavy rain but tend to extend the raining area
• Satellite data tend to under-estimate
• Merged analysis present improved
depiction of the heavy rain
Reports from up to 28 stns available in a 0.25olat/lon grid box over Seoul
Arithmetic mean of 28-stn reports taken as the ‘truth’
Gauge analyses using 1000 combinations of sub-set stations are combined with bias-corrected
CMORPH and compared to the ‘truth’ at the grid box over Seoul
Correlation is calculated for the combined analyses and the input
gauge / CMORPH. Results are plotted in different colors for different gauge network densities
Combining Gauge with Satellite [4]
Combining Gauge with Satellite [4]
Independent tests using data over Korea
Independent tests using data over Korea
Operation System Operation System
Gauge Analysis (0.25olat/lon)
CMORPH (0.25olat/lon)
PDF Matching Bias Correction
Bias-Corrected CMORPH
OI Combining
Merged Analysis
Summary Summary
• We are in final stage of constructing gauge-satellite merged analyses of global precipitation;
• Two sets of gauge-satellite precipitation analyses
• Bias-corrected Satellite Estimates
•Global
•8kmx8km; 30-min
•1998 to the present
• Gauge-satellite combined analyses
•Regional
•0.25olat/lon; daily
•1998 to the present
Related R&D Activities [1]
Related R&D Activities [1]
• Developing 2
ndgeneration Kalman filter based CMORPH satellite estimates (Part of PMM project)
• Capable of including information from additional sources (e.g. IR, model)
• Integrating information based on more accurate statistical framework
Simulation tests of the original and KF CMORPH with inputs from 1,2,4,7, and 9 PMW satellites
Comparisons against radar observations over CONUS for different local times
Related R&D Activities [2]
Related R&D Activities [2]
• Gauge-radar-satellite merged analysis of hourly precipitation over CONUS
• Final goal: a portable module to combine global high-resolution satellite estimates (e.g. CMORPH) with regionally available additional information (e.g. gauge, radar, model) to create precip analysis of higher quality and resolution
Related R&D Activities [3]
Related R&D Activities [3]
• Extending the CMORPH to cover the entire globe
(pole-to-pole)
• Cloud advection vectors
• Cold season
precipitation rates
• Taking advantage of
model simulations
Thank You !!
Thank You !!
Backup Slides
Backup Slides
Gauge Error [1]
Gauge Error [1]
Gauge error is defined by comparing the gauge analysis derived from reports of different gauge network
configurations against the ’truth’
Daily data from an extremely dense station network over Korea is used. Over a 0.25
olat/lon grid box over Seoul,
Korea, reports from 28 stations are available and their arithmetic mean is used as the ‘truth’ for that grid box
Gauge analyses are constructed using reports from 1,000 combinations of stations to mimic the configuration of those over China; and they are compared to the ‘truth’
for 2005-2007
Gauge Error [2]
Gauge Error [2]
Results are analyzed according to the precipitation
intensity and gauge network density
Gauge network density is defined using a parameter called Number of Equivalent Gauges (Neg)
Neg = Ng0 + Ng1 + Ng2
Ng0 : # of gauges inside the target grid box Ng1 : # of gauges inside the grid boxes
neighboring to the target grid box Ng2: # of gauges inside the grid boxes one f
urther layer away.
Gauge Error [3]
Gauge Error [3]
Error variance of the gaugeanalysis linearly proportional to the precipitation intensity and inversely proportional to the local gauge
network density measured by the Number of Equivalent Gauges (Neg)
An empirical relation established between the error variance and the precipitation and the Neg:
E2 = a + b ∙ R / ( Neg + 1 ) a = 0.15 (mm/day)2
b = 4.09 (mm/day)2
CMORPH Error [1]
CMORPH Error [1]
Assuming CMORPH error variance is proportional to the estimated precipitation amount
Proportional constant is determined through
comparison with gauge analysis over grid boxes with at least one gauge using data over China for
summer 2007
CMORPH error is defined for each grid box of
0.25olat/lon using the curve in the bottom panel of the figure
CMORPH Error Correlation CMORPH Error Correlation
Correlation between the bias- corrected CMOROH
error at one location and
that at a different location
Assuming the error
correlation is only a function of separation distance
Correlation calculated for all combinations of grid boxes using data over China for
summer 2007
Results fitted to a Gaussian function