1. TOWARDS A CLIMATE DATA RECORD OF PRECIPITATION FROM SATELLITE MICROWAVE IMAGER DATA
Wesley Berg, Matt Sapiano, Dave Randel, and Chris Kummerow, Colorado State University
INTRODUCTION
DEVELOPING A PARAMETRIC PRECIPITATION RETRIEVAL ALGORITHM
GPROF 2008
The constellation of past and present conically-scanning window channel radiometers with channels applicable to the estimation of precipitation is shown above. While the DMSP satellites carrying the SSM/I and SSMIS series sensors comprises the bulk of the long-term data record, a number of additional sensors have been launched in the past decade or so that provide significantly improved sampling for many precipitation applications. The challenge of creating a precipitation CDR, however, is developing intercalibration techniques and retrieval algorithms that can provide physically-consistent precipitation estimates across the globe extending back to the original SSM/I on board DMSP F08.
The Global Precipitation Mission
Developing Consistent Precipitation Products for a Radiometer Constellation
CONSTELLATION SATELLITES
~8 satellites
Radiometer only
Rely on existing radiometers
Multifrequency radiometer
~3 hour average revisit time
CORE SATELLITE
Dual frequency radar
Non-sun synchronous orbit
~70 deg inclination
~400km altitude
~4km horizontal resolution
250m vertical resolution
MISSION: Understand the horizontal and vertical structure of rainfall and its microphysical elements. Provide training for constellation radiometers
MISSION: Provide enough sampling to reduce uncertainty in short-term rainfall accumulations. Extend scientific and societal applications.
GPROF 2008 Zonal Mean Precipitation
GPROF2008 is the first GRPOF version to use the combined radar/radiometer retrieval for it’s a-priori database over oceans. As such, it represents a significant step forward over GPROF2004 which used only cloud resolving models to construct the a priori database. The key to GPROF2008 is to produce these products in a consistent manner across satellites to ensure that trends and changes can be robustly related to changes in the climate system rather than changes in the sensors themselves.
GPROF 2008
GPROF 2004
1.03 mm/hr
0.75 mm/hr
South Africa
2008 : 0.29 mm/hr
PR : 0.40 mm/hr
2004 : 0.03 mm/hr
Inconsistencies in precipitation estimates between satellite sensors due to calibration errors, differences in information content, and factors such as spatial resolution can lead to time-dependent biases as new sensor are launched and old ones die. Assumptions in the retrieval algorithm can also lead to significant biases in precipitation estimates leading to regional and time-dependent errors. Differences between the SSM/I rainfall trends shown above are likely due to a combination of these issues.
In developing a true Climate Data Record of precipitation in is critical to reduce the errors as much as possible, but since it is impossible to remove all the errors, we need to understand and quantify residual errors. This is the only way to determine if a signal in the data is due to real changes in the climate system or is simply an artifact of the data.
As shown above, the Global Precipitation Mission (GPM) combines a core satellite along with a constellation of radiometers from a diverse range of operational and research satellites. One of the goals of the mission, therefore, is to produce consistent precipitation estimates from the available radiometer constellation. As such, the GPM mission is driving the development of both intercalibration and algorithm development for a diverse constellation of satellite radiometers.
Results from the latest GPROF 2008 retrieval algorithm are shown above compared to the GPROF 2004 estimates for two different cases including Hurricane Floyd in September of 1999 and a light rain case off the coast of South Africa. The apriori database used in the GPROF 2008 retrieval combines information from the TRMM Microwave Imager (TMI), the precipitation radar (PR) and Cloud Resolving Model (CRM) simulations. The light rain case on the left has the PR 2A25 retrieved estimates shown along the narrow center swath indicating significantly improved retrievals of light rainfall. The hurricane case also shows more realistic falloff with a less abrupt transition to non-raining pixels are less artifacts in the spatial structure of the precipitation within the rain bands.
A comparison of zonal mean rainfall from SSM/I F13, F14, F15, and TMI show excellent agreement for two different seasons. While some differences remain, a significant portion of this is due to real differences associated with diurnal variability given differences in the local observing times of the various sensors. Although TMI and SSM/I sensors have very different spatial resolution and differences in channel frequencies etc., this shows that consistent climate estimates from these sensors is possible.
Algorithm Biases
INTERCALIBRATION OF A RADIOMETER CONSTELLATION
The plot on the left shows regional biases between the Version 6 TMI and PR retrieval algorithms. Note that the V6 TMI algorithm is GPROF Based on the results of an analysis of V6 biases by Berg et al. [2006], the apriori database in the GPROF 2008 algorithm is stratified by both SST and total precipitable water (TPW). This is based on the finding that TPW provides an excellent proxy for freezing height and is strongly correlated with the differences between the PR and TMI estimates. The plot on the lower left shows the frequency of occurrence of different cloud types as a function of cloud top temperature (VIRS ch 4 TBs) and PR echo-top height for various TPW regimes. This stratification is based on an approach developed by Masunaga et al and clearly shows dramatic differences in the structure of precipitating clouds between moist tropical regions and dry subtropical regimes.
Initial comparisons with GPROF 2008 (i.e. V7) indicate a significant reduction in the biases between the PR and TMI estimates, although the V7 products are still undergoing testing and have not yet been released. Minimizing regional and time-dependent biases in the rainfall estimates, however, is critical for developing a precipitation CDR as these biases may either hide real climate signals or produce false ones.
NOAA SDS Project Creating an FCDR of SSM/I and SSMIS TBs
Technique 1: Comparisons with TRMM Microwave Imager (TMI)
SSMI and TMI Ground tracks for 2-3am on 20 Nov 2004
Goals
Generate a transparent and documented Fundamental Climate Data Record (FCDR) of SSM/I and SSMIS brightness temperatures from Present.
Develop/implement multiple approaches for intercalibration and then identify and apply the best approach in the stewardship code to produce FCDR.
Intercompare and examine calibration procedures and results to provide not only the best possible product, but also an assessment of this product.
Getting Matchups with TMI
Obtain points where SSMI and TMI groundtracks cross within 30 minutes & 50km (exclude duplicates within 60 seconds of each other)
Calculate 1° average Brightness Temperature (Tb) for each sensor
Include only clear sky: require that SD(85GHz) < 5K
Used TMI V7 beta (ITE 200) which has some extra corrections over V6 (UCF correction)
Approach
Reformat SSMI/SSMIS TDR files into NetCDF “Base files”. These files contain all the original data and nothing is modified except to make orbit granules, add ephemeris and reformatted time, and reformat to NetCDF.
Create a well documented software package (“Stewardship code”) that ingests the Base files, applies corrections (i.e QC, cross-track bias, TA-TB, geolocation, calibration) and outputs the final FCDR in NetCDF for use by the broader community.
Expert users can be given access to the Base files and the “stewardship code”. This gives them access to the beta versions without confusing the general users.
Removal of sensor dependent differences
Basefiles
FCDR files
QC flags
Intercalibration
Geolocation
Ta to Tb
Cross track bias
correction
Stewardship
Code
Use Elsaesser and Kummerow (2008) Optimal Estimation (OE) approach to retrieve clear sky geophysical parameters (Wind, TPW, LWP) from Tbs and SST (Reynolds)
Technique finds optimal geo parameters given Tbs based on inversion of a Radiative Transfer (RT) model
Use OE to get Geophysical parameters from TMI Tbs
Use same RT model to simulate idealized TMI Tbs from geo parameters
Use RT again to simulate SSMI Tbs based on same geophysical parameters
Compare difference between simulated Tbs (SSMIsim-TMIsim) with difference between actual Tbs (SSMI-TMI) to get calibration offsets
2-3 April, 2007
2D Histograms as a Function of Total Precipitable Water
In addition to regional and temporal biases in rainfall products due to assumptions in the retrieval algorithms, another important issue for developing a precipitation CDR is understanding the limitations of the sensor. For example, comparisons between estimates from the TRMM PR and the 94 GHz CloudSat Cloud Profiling Radar (CPR) suggest that PR misses around 10% of the rain volume and 2/3 of the rain frequency due to its high minimum detectable signal of ~17 dBZ. New technology such as CloudSat and the dual frequency GPM radars are important to improve estimates of light rain as well as to quantify regional impacts.
From Masunaga et al. 2005
Green crosses show what SSMI-TMI should be if IA was the only difference. Blue circles show differences due to IA plus other factors (which we want to remove)
Intercalibration Approach
Intercalibration strategy is to compare several different approaches
Goal is to understand differences and use sensor information to select best solution
One of the four techniques is based on coincident overpasses with common instrument: TRMM TMI
Cannot be used for whole period, but can provide verification of calibration from other techniques
TMI and SSMI are similar, but not the same
View angle (Earth Incidence Angle); Channels slightly different (21.3 vs GHz); Footprint sizes
Preliminary Results
Table shows intercal values for F13-F14 from the TMI matchups method, the vicarious cold calibration method and the simultaneous nadir overpass method.
Plot on the left shows resulting calibration offsets as a function of brightness temperature for most current sensors using the TMI comparison method
The Incidence Angle (IA) is currently not known with sufficient accuracy which seems to have a large effect on the results from the TMI matchups.
IA is reliant on changes in spacecraft attitude which are not known. Previous estimates were based on a coastline analysis based on inaccurate geolocation. We are currently working to redo this analysis with more accurate pixel geolocation
We have thus far been unable to obtain intercal numbers from the comparison against in situ observations. However, this approach is an important reality check and underscores the importance of working with TCDR developers
SUMMARY
CONTACT INFO
Wesley Berg
Dept. of Atmospheric Science
Fort Collins, CO
(970)
Satellite Intercalibration:
As part of a funded NOAA project, we will be delivering an intercalibrated dataset of brightness temperatures from SSM/I and SSMIS extending from July 1987 through the present to NCDC in the next two years. This data will be publicly distributed by NCDC.
The existing Remote Sensing Systems (RSS) dataset of calibrated SSM/I TBs will be made publicly available through NCDC shortly. Comparisons between these two dataset should provide valuable insights into the residual errors/issues and the accuracy of the calibrations.
An XCAL working group of the NASA Precipitation Measurement Missions (PMM) is working to develop intercalibration approaches and initial corrections for the existing radiometer constellation prior to the launch of GPM in 2013.
Precipitation Retrieval Algorithms:
GPROF 2008 represents a significant step forward in the development of an ocean precipitation retrieval algorithm that can provide physically consistent estimates from a diverse constellation of radiometers.
Significant work remains in the development of similar retrieval algorithms over land and high latitude regions. The launch of GPM in 2013 with its dual frequency radars and the addition of the 183 GHz water vapor sounding channels will provide valuable information at higher latitudes and for light rain that is currently unavailable.
Errors in Climate Products:
Absolute Calibration of radiometer brightness temperatures is a noble goal but not currently achievable to better than perhaps 1-2K. Relative calibration is more critical, however, with goals of ~0.2 K considered possible.
Estimating biases in temporally and spatially-averaged precipitation will likely continue to be a significant challenge for some time. New technology such as the GPM radars, CloudSat, and polarimetric ground radars, however, will provide valuable new data towards this goal.
Inter-calibration Technqiues
Comparisons with TRMM Microwave Imager (TMI)
Get overpasses of TMI and SSMI; use radiative transfer calculations to remove differences caused by slight differences in channels and incidence angle; only applicable over TRMM period (1998-present)
Vicarious Cold Calibration (and warm calibration)
Each frequency has a theoretical minimum TB which occurs at a given SST; this minimum is stable and can be used to do intercal of two sensors; TB minimum is withi observed range of SSTs for 19, 22 and 37GHz, but not for 85GHz.
Simultaneous Nadir Overpasses
Find crossovers of SSMI instruments and use nadir point to match them. Channels and IA are assumed to be the same; matchups occur at poles only.
Comparison of geophysical parameter retrievals with in situ observations (buoy wind speed)
Compare values from SSMI using several existing wind speed algorithms with those from buoys; Use wind speed bias as a measure of fidelity of intercal
Can wind speed bias be used to estimate intercal?