H53B-01
Spatio-temporal Variability of Precipitation over northern Africa
The statistical properties of precipitation events from various satellite products are investigated for northern Africa during the rainy summer season. Different statistical measures, including correlation function and critical success indices, are employed. The spatial scales of interest range from 4 km to 100 km, and the temporal scales of interest range from instantaneous to monthly. The statistics are conditioned on different rainfall rates. The statistics clearly indicate the considerable geographical variability of precipitation variability. The results of this study will be useful for downscaling coarse precipitation fields, and for evaluating satellite-derived precipitation estimates.
H53B-02
The Decomposition of Heterogeneous Rain into Homogeneous Components: Results From a Statistical Inversion of Count Data
Most variables in meteorology are statistically heterogeneous; that is, in the broadest sense, their statistics depend upon the location (temporal or spatial origin) of the observations. Yet often measurements of variables are gathered at widely disparate locations in space-time and are processed as though the data were fully characterized by just one pdf and one single set of parameters having one mean value. Is there, instead, a better way of treating the observations in a manner that is consistent with the actual statistical heterogeneity of the data? We address this question using a statistical inversion technique based upon Bayesian methodology. Two examples of disdrometer measurements in real rain reveal the presence of multiple mean values of the counts at all the different drop sizes. This immediately exposes the statistical heterogeneity of the two data sets, one 16 hours and the other three minutes long. Furthermore, the analyses reveal that in both cases the heterogeneous rain can be decomposed into five to seven statistically homogeneous components, each characterized by its own steady drop size distribution. Every observation can then be represented as a linear, weighted combination of a properly selected complete set of these components. This is a great reduction in the complexity required to describe and to simulate rain events. Furthermore, rather imprecise concepts such as stratiform and convective rain can also be given more exact, objective meaning in terms of the contributions each component makes to the rain. In these data, for example, large numbers of smaller drops associated with one set of components found in the regions of light stratiform rainfall are sandwiched between those of more intense convective rainfall more closely associated with a different set of components having large drops. These results also again confirm the futility of trying to justify or to interpret physically one power law Z-R relation fit to an entire set of vastly different meteorological conditions. More important, however, this discovery allows for the incorporation of statistical heterogeneity explicitly and analytically into radar theory as illustrated in the other article at this conference.
H53B-03
Climatological Downscaling and Evaluation of AGRMET Precipitation Analyses Over the Continental U.S.
The spatially distributed application of a land surface model (LSM) over a region of interest requires the application of similarly distributed precipitation fields that can be derived from various sources, including surface gauge networks, surface-based radar, and orbital platforms. The spatial variability of precipitation influences the spatial organization of soil temperature and moisture states and, consequently, the spatial variability of land- atmosphere fluxes. The accuracy of spatially-distributed precipitation fields can contribute significantly to the uncertainty of model-based hydrological states and fluxes at the land surface. Collaborations between the Air Force Weather Agency (AFWA), NASA, and Oregon State University have led to improvements in the processing of meteorological forcing inputs for the NASA-GSFC Land Information System (LIS; Kumar et al. 2006), a sophisticated framework for LSM operation and model coupling experiments. Efforts at AFWA toward the production of surface hydrometeorological products are currently in transition from the legacy Agricultural Meteorology modeling system (AGRMET) to use of the LIS framework and procedures. Recent enhancements to meteorological input processing for application to land surface models in LIS include the assimilation of climate-based information for the spatial interpolation and downscaling of precipitation fields. Climatological information included in the LIS-based downscaling procedure for North America is provided by a monthly high-resolution PRISM (Daly et al. 1994, 2002; Daly 2006) dataset based on a 30-year analysis period. The combination of these sources and methods attempts to address the strengths and weaknesses of available legacy products, objective interpolation methods, and the PRISM knowledge-based methodology. All of these efforts are oriented on an operational need for timely estimation of spatial precipitation fields at adequate spatial resolution for customer dissemination and near-real-time simulations in regions of interest. This work focuses on value added to the AGRMET precipitation product by the inclusion of high-quality climatological information on a monthly time scale. The AGRMET method uses microwave-based satellite precipitation estimates from various polar-orbiting platforms (NOAA POES and DMSP), infrared-based estimates from geostationary platforms (GOES, METEOSAT, etc.), related cloud analysis products, and surface gauge observations in a complex and hierarchical blending process. Results from processing of the legacy AGRMET precipitation products over the U.S. using LIS-based methods for downscaling, both with and without climatological factors, are evaluated against high-resolution monthly analyses using the PRISM knowledge- based method (Daly et al. 2002). It is demonstrated that the incorporation of climatological information in a downscaling procedure can significantly enhance the accuracy, and potential utility, of AFWA precipitation products for military and civilian customer applications.
H53B-04
Comparison of Eight Different Precipitation Datasets for South America
Long and continuous meteorological data series for large areas are hard to obtain, so several groups have developed climate datasets generated through the combination of models and observed and remote sensing data, including reanalysis products. This study compares eight different precipitation datasets for South America (NCEP/NCAR-2, ERA-40, CMAP, GPCP, CRU, CPTEC, TRMM, Legates and Willmott, Leemans and Cramer). For each dataset, we analyze the four moments of the data distribution (mean, variance, skewness, kurtosis), for latitudinal variation, for the major river basins and for the major vegetation types in the continent, allowing to identify the geographical variations in each dataset. We verified that significant differences exist among the precipitation products.
H53B-05
Rainfall Sampling Uncertainties: A Rain Gauge Perspective
Rain gauge networks provide rainfall measurements with high degree of accuracy at specific locations but, in most of the cases, these networks are too sparse to accurately capture the high spatial and temporal variability of the precipitation systems. Radar and satellite remote sensing of rainfall has become a viable approach for effective addressing of this problem. However, among other sources of uncertainties, the remote-sensing based rainfall products are unavoidably affected by sampling errors that need to be evaluated and characterized. Using a large (seven years) dataset of rainfall measurements by a highly dense rain gauge network (50 gauges in about 140 km2) deployed in the Brue catchment (south-west part of England), this study sheds some light on the temporal and spatial sampling uncertainties: the former are defined as the errors resulting from temporal gaps in rainfall observations, while the latter as the uncertainties due to the approximation of an areal estimate with a point measurement. As far as the temporal sampling uncertainties are concerned, it will be shown that they increase with the sampling interval according to a scaling law and decrease with increasing pixel size with no strong dependence on local orography. On the other hand, the spatial sampling uncertainties tend to decrease for increasing accumulation time, with no apparent dependence on location of the gauge within the pixel or on the gauge elevation. Additionally, results pertaining to their dependence on the rainfall intensity will be presented.
H53B-06
A Satellite-Gauge Merged Analysis of Hourly Precipitation over Southern China
A new technique has been developed to construct analyses of hourly precipitation on a 0.125olat/lon over Guang- Dong province in southern China by merging gauge observations and satellite estimates. Hourly precipitation reports from ~400 stations are available on a real-time basis and used in this study to create the merged precipitation analysis over this province of ~150,000 km2. The high-resolution satellite precipitation estimates used here are those of CPC Morphing Technique (CMORPH, Joyce et al. 2004) which generates 30-min precipitation rates on an 8kmx8km grid over the globe by combining information from satellite-based microwave and infrared observations. The original CMORPH precipitation estimates are regridded into hourly and 0.125olat/lon resolution for use as inputs to our merging procedures. A two-step approach is designed to merge the hourly gauge observations and CMORPH satellite precipitation estimates. In the first step, the CMORPH precipitation estimates are calibrated against the gauge data to remove the inherent biases. To this end, ratio between the mean precipitation for the most recent 30 days from the gauge observations and that from the CMORPH estimates is computed for each gauge location and for each target date. An analyzed field of the gauge-vs-CMORPH ratio is then defined by interpolating the station values through the optimal interpolation (OI) technique of Gandin (1965). Biases in the CMORPH are finally removed by multiplying the ratio to the original satellite estimates. The second step is intended to improve the quantitative accuracy of the precipitation analysis. The bias-corrected CMORPH is combined with the gauge data, again, through an OI-based objective analysis technique, in which the bias-corrected CMORPH is utilized as the first guess while the station gauge data are employed as observations to calculate the increments. The weighting coefficients are calculated through the error structures of the CMORPH and gauge observations, so that over areas with dense gauge network, the final merged analysis is determined primarily by the gauge data while over gauge sparse regions the satellite observations plays more important roles. Cross-validation tests revealed that the merged analysis presents complete spatial coverage with stable and improved performance statistics compared with the individual inputs. A test product of the hourly precipitation analysis has been created for a three-month period from April 1 - June 30, 2005, and applied to examine the diurnal cycle of precipitation over this sub-tropical area during a pre-monsoon season. Detailed results will be reported at the meeting.
H53B-07
EPPrePMex: An Operational Real-Time Rainfall Estimation System for Mexico Based on GOES-IR Imagery.
EPPrePMex is a tool for real-time rainfall estimation, calibrated for the Mexican territory and based on satellite imagery. We have implemented, calibrated and operationally evaluated the CST technique (by Adler and Negri) for Mexico. The results show that, in general, EPPrePMex gives an overestimation of the rainfall field. We have realized systematic comparisons for daily rainfall estimations over Mexico and USA with comparable results obtained with HydroEstimator(NOAA). However In order to improve our system, several modifications are planned: a) Calibration for different regions, season and kind of storms, b) A joint use and calibration of satellite and radar images, c) Dynamic study of storms, and d) Real time calibration with microwave and other kind of sensor data from low orbit satellites. We have proposed hydrological applications of the operational results of the EPPrePMex system (real time forecast and warning systems)for central and SW Mexico.
H53B-08
Evaluating Intercalibrated Passive Microwave Rain Rates and Hydrological Consistency
Remote Sensing Systems simultaneously retrieves sea surface temperature, surface wind speed, columnar water vapor, columnar cloud water, and surface rain rate from a variety of passive microwave radiometers including SSM/I, TMI, and AMSR. The rain component of the retrieval algorithm has recently undergone major improvements, and the newly reprocessed data are now available. The physical basis for these changes will be briefly described. We have also completed a major intercalibration effort. In the most recent Version 6, the six SSM/Is have been carefully intercalibrated to a precision of about 0.1 K in brightness temperature, and TMI and AMSR-E have been adjusted to match the SSM/I time series. Wind speed retrievals are very sensitive to brightness temperature calibration errors, and the good agreement with buoys is consistent with an intercalibration error of 0.1 to 0.2 K or less. We use the SSM/I wind and rain products to examine the balance of evaporation and precipitation. Trends over the 20 year time period are used to assess the long-term stability of the dataset. Assessing wind speed errors is easier than assessing precipitation errors, and the magnitude of the error bar on the SSM/I wind trend is estimated to be 0.05 m/s/decade at the 95 percent confidence level. Evaporation is used to help evaluate the accuracy of our precipitation retrievals, and we compare our results to GPCP. Special attention is paid to the middle latitude oceans where passive microwave estimates are still in wide disagreement.
H53B-09
An Inter-comparison of Passive Microwave Rainfall Derived From Various Sensors and Algorithms With TRMM 2A12
A passive microwave (PMW) based blended satellite rainfall estimation technique such as CMORPH will inherently suffer problems when combining and propagating rainfall derived from vastly different frequencies, sampling characteristics, and estimation algorithms resulting in pronounced relative biases. PMW precipitation products are currently derived from AMSR-E, TMI, SSMI & SSMIS, and AMSU-B instruments aboard the AQUA, TRMM, DSMP, and POES satellites respectively. Inter-calibration is crucial because precipitation products that are generated from instruments with different spectral characteristics will not provide the same value for the same scene. However, first rainfall detection frequency, rain rate distribution, and total rainfall from each instrument type must be inter-compared and quantified in order to implement calibration procedures. Certain sensor/algorithms have well known problems such as scan angle dependencies and oceanic rainfall detection deficiencies in the current NESDIS algorithm for the AMSU-B cross-tracking sensor. Despite these problems, the skill of the AMSU-B rainfall estimates is better than IR derived precipitation skill. Furthermore, because the AMSU-B instrument is deployed on four polar orbiting satellites with equatorial crossing times of approximately 1900, 1430, 1330, and 2230, the instrument reasonably samples the diurnal cycle with a broad 2300 km swath. Even algorithms for the same sensor will have significant differences such as the GPROF and EDRR rainfall algorithms for the DMSP SSMI. Rainfall from these various PMW sensor/algorithms and as well as an IR algorithm are inter-compared with temporally and spatially coincident TRMM TMI 2A12 rainfall over various regions, latitudes, seasons, and surface types.
H53B-10
Developing Methodologies for Applying TRMM-Estimated Precipitation Data to Hydrological Modeling of a South TX Watershed - Initial Results
Previous experience with hydrological modeling in South Texas, which is located along the Texas-Mexico border, suggests that NWS ground measurements are too widely scattered to provide reliable precipitation input for modeling. In addition, a significant fraction of the study region is located at the edge of the coverage envelopes of the NWS NEXRAD weather radars present in the region limiting the accuracy of these systems to provide reliable precipitation estimates. Therefore, we are exploring whether TRMM estimated-precipitation data (3B42), in some form, can be used to support hydrological modeling in the Middle Rio Grande and Nueces River Basin watersheds. We have begun our modeling efforts by focusing on the middle Nueces watershed (7770 sq km). To model this largely rural watershed we selected the Soil and Water Assessment Tool (SWAT). Three precipitation datasets were selected for our initial model runs that include: (1) nearest NWS cooperative hourly rain gauge data, (2) three hourly TRMM 3B42 estimated precipitation, and (3) combination TRMM 3B42/NWS rain gauge datasets in which ground measurements are used for three hourly periods lacking high quality satellite microwave precipitation estimates as determined from TRMM 3G68 data. Three dataset were aggregated into an average daily estimate of precipitation for each TRMM grid cell. Manual calibration of was completed achieving model results that yield realistic monthly and annual water balances with both gauge and satellite estimate precipitation datasets. In the future, we plan to use the newly developed automatic calibration routine for SWAT, which is based on the Shuffled Complex Evolution algorithm, to optimize modeled discharge results from this study.
H53B-11
Bayesian Approach in Ensemble Forcing Generation for Improved Ensemble Streamflow Prediction
Many studies have demonstrated that precipitation is the primary source of uncertainty affecting streamflow prediction. As part of the Advance Hydrologic Prediction Service (AHPS) of the National Weather Service, improvements have been made to ensemble streamflow prediction (ESP) production. In the current ESP procedure, precipitation ensemble members from historical time series are treated equally. However, there is no strong reason that such a precipitation ensemble properly represents the precipitation uncertainty. Using a Bayesian approach, a weighting method has been developed to give preference to historical precipitation time series which resemble the hydrologic conditions preceding the current watershed state. In an ensemble forecast, this is done with the intent of decreasing the influence of precipitation records occurring during years that are dissimilar to the current state being simulated. In producing probabilistic river discharge forecasts, this would improve both precision and accuracy of probabilistic forecasts. The method uses, but is not restricted to, the NWSRFS soil moisture accounting model and can be applied to both lumped and distributed watershed simulations.
H53B-12
Transferability Studies - Evaluating and Improving Simulated Precipitation During CEOP
Transferability studies are a useful methodology for evaluating the capability of regional climate models to simulate different climates. We are participating in the Coordinated Enhanced Observing Period (CEOP) Inter- Continental Transferability Study (ICTS) with the Experimental Climate Prediction Center's (ECPC's) Regional Spectral Model (RSM). We are evaluating the capability of the RSM to simulate energy and water budget components on seven different regional domains. Numerical simulations were conducted for regional climates in tropical, subtropical, mid-latitude and polar-regions. In particular, regional simulations were carried out for small- scale convective systems and various large-scale circulation regimes including: monsoons, the ITCZ, and mid- latitude storms. Sensitivity tests with four different convection schemes showed that either the Kain Fritsch (KF) convection scheme or the Simplified Arakawa Schubert Scheme (SAS) provided the best precipitation simulations for most domains. In those regions where the SAS convection scheme provided the best results, the KF scheme had poor results. On the other hand, in those regions where the KF scheme had the best result, the precipitation simulation using the SAS scheme was only slightly worse. On the basis of these findings we decided to rerun our original long term ICTS runs (July 1999 to December 2004) with the SAS convection scheme. These new long- term runs are now being compared to the previous long-term runs, which used the Relaxed Arakawa Schubert (RAS) convection scheme. Major improvements have been identified over the LBA domain. Also, the annual cycles over the LBA and the AMMA domains were improved in the long-term runs. Further comparisons with other water and energy budget components will be presented at the conference.
H53B-13
The scale dependence of rain: from raindrop stereophotography to global TRMM orbits
Rain is a highly turbulent process over enormous ranges of scale. The stereophotography of rain drops directly demonstrates that rain is strongly coupled with the wind field down to a "relaxation" scale of the order 30-50cm below which drop inertia makes them free. Using the results of this "HYDROP" experiment, we show that while the liquid water density (ro) follows a multifractal generalization of the classical Corrsin-Obukov passive scalar law; deltaro=l**1/3, on the contrary, the drop number density (n) follows the new law deltan=l**1/2. We show how both laws can be derived from dimensional analysis using appropriate turbulent fluxes. The HYDROP experiment only determines the statistical properties of precipitation over the range of scales 3cm to 2m; for larger but intermediate scales (3m- 5km), we use lidar and radar data which also show scaling of the backscatter cross-section and radar reflectivity factors respectively. To extend this up to planetary scales (5 - 20,000km), we use 3 months (1166 orbits) of the TRMM (Tropical Rainfall Monitoring Mission) satellite radar data at heights 250m above the surface. This global data set (over the region ±38 degrees latitude) is remarkable for it's relatively complete and uniform coverage over a range of =4,000 in scale; it is also nearly free of the range dependent biases which plague ground based radar data. Ignoring for the moment the statistics from the weak reflectivities (which are biased due to the rather higher minimum detectable signal), we find that over the observed range 5-20000km, the moments Mean(Z(lambda)**q) for q less than 2 follow the theoretically predicted form Mean(Z(lambda)**q)= lambda**K(q) to within a maximum deviation of 6 percent where lambda=Leff/Lres and Lres, is the resolution of the reflectivities and Leff is the effective outer scale of the cascade. We find Leff= 40000km indicating that at planetary scales (20000km), there is residual variability from other interactions. Analyzing the exponent K(q) we show that it is well reproduced by a theoretically predicted two parameter ("universal") form, with codimension of the mean field C1=0.63 and Levy index α (characterizing the degree of multifractality) =1.5. Finally over the range 20,000 - 5km using multifractal simulations we show that if the minimum detectable signal is =0.5 times the mean, then the slight deviations in the scaling of the low order moments (q less than 0.5) are explained to within 7 percent. These findings allow us to make stochastic precipitation models which reproduce these isotropic statistics over huge ranges of scale. On the one hand, we can make compound Poisson / multifractal/ models of the distribution of individual drops which respect the basic turbulence laws, on the other hand, the same model can be used at much larger scales to accurately model the TRMM reflectivities.