Predictibility in Nowcasting of Precipitation
Present short term precipitation forecasting is based on two methods: Lagrangian persistence (nowcasting) and numerical weather prediction (NWP). An improvement over these methods is obtained by the combination of the two. The obvious shortcoming of nowcasting is its severe limitation in capturing new development or dissipation of precipitation. NWP has the ability to predict both but very imprecisely. An attempt to correct model errors by post-processing leads to some improvement in the skill of NWP, but the improvement, although significative, is not very impressive. The goal of our effort is to take a step back and to describe, in a quantitative manner, a) the nature of the uncertainties affecting Lagrangian persistence and NWP forecasts, as well as to determine b) the physical causes of the uncertainties. We quantify the uncertainties in short term forecasting due to limitation of nowcasting algorithms and NWP to capture correctly some of the physical phenomena that determine the predictability of precipitation. The first factor considered is the diurnal cycle that appears as the one physically determined factors that limit the precision of short term prediction. We study the cycle in radar mosaics over US and compare this to nowcasts and model outputs. The seasonal and geographical dependence of the diurnal cycle is quantitatively evaluated.
Environmental Influences on Precipitation Intensity in Simulated Convective Storms
Anticipating whether or not convective storms will produce heavy precipitation is a challenging forecast
problem. In this paper, we utilize a set of over 200 unique cloud-resolving numerical simulations to investigate
vertical profiles of temperature, humidity, and wind that favor increased surface precipitation intensity. Strong
correlations are found to exist between surface rain and hail mixing ratios and deep tropospheric wind shear,
with greater shears frequently associated with better organized convection and more precipitation. Raising the
height of the level of free convection (LFC) is also somewhat correlated with increased surface rain and hail,
which is enhanced by greater updraft overturning efficiency compared to storms in low LFC environments.
Storms in cooler environments having reduced amounts of precipitable water (PW) nevertheless display
greater precipitation efficiency and thus produce almost as much precipitation as storms in warmer, moister
environments with high PW. Interestingly, surface rainfall exhibits no preferential increase as convective
available potential energy (CAPE) increases. Possible changes to storm rainfall production in the presence of
environmental change will also be discussed.
An Improved Polarimetric Radar Rainfall Algorithm With Hydrometeor Classification Optimized For Rainfall Estimation
The efficacy of dual polarimetric radar for quantitative precipitation estimation (QPE) is firmly established. Specifically, rainfall retrievals using combinations of reflectivity (ZH), differential reflectivity (ZDR), and specific differential phase (KDP) have advantages over traditional Z-R methods because more information about the drop size distribution and hydrometeor type are available. In addition, dual-polarization radar measurements are generally less susceptible to error and biases due to the presence of ice in the sampling volume. A number of methods have been developed to estimate rainfall from dual-polarization radar measurements. However, the robustness of these techniques in different precipitation regimes is unknown. Because the National Weather Service (NWS) will soon upgrade the WSR 88-D radar network to dual-polarization capability, it is important to test retrieval algorithms in different meteorological environments in order to better understand the limitations of the different methodologies. An important issue in dual-polarimetric rainfall estimation is determining which method to employ for a given set of polarimetric observables. For example, under what circumstances does differential phase information provide superior rain estimates relative to methods using reflectivity and differential reflectivity? At Colorado State University (CSU), a "blended" algorithm has been developed and used for a number of years to estimate rainfall based on ZH, ZDR, and KDP (Cifelli et al. 2002). The rainfall estimators for each sampling volume are chosen on the basis of fixed thresholds, which maximize the measurement capability of each polarimetric variable and combinations of variables. Tests have shown, however, that the retrieval is sensitive to the calculation of ice fraction in the radar volume via the difference reflectivity (ZDP - Golestani et al. 1989) methodology such that an inappropriate estimator can be selected in situations where radar echo is relatively weak (< 40 dBZ). In this study, a new blended rainfall algorithm is developed using hydrometeor identification (HID) to drive the rainfall estimation algorithm. HID discrimination for rainfall application namely, (1) all rain, (2) mixed precipitation, and (3) all ice, is used to guide the selection of the most appropriate rainfall estimator. Data collected from the CSU-CHILL radar and a network of rain gauges are used to test the performance of the new algorithm in a variety of precipitation situations. The results are compared to similar results using the algorithm from the National Severe Storm Laboratory (NSSL), derived from Oklahoma precipitation events ( Ryzhkov et al. 2005 ). The applicability of the method derived from Oklahoma observations to Colorado precipitation events is also explored.
Analysis of spatial similarities between Stage III NEXRAD precipitation and LDAS combo precipitation data products
Precipitation is one of many essential input components required in hydrological modeling. Although the Stage III/Multi-sensor Precipitation Estimator (MPE) data from NEXRAD (Next Generation Doppler Radar) and the LDAS (Land Data Assimilation Systems) combo precipitation data products have been extensively used in various hydrological and climatic studies, there has been no systematic investigation of the spatial similarities and differences between these two data products, such as their spatial coverage patterns, based on long-term time series data over a large spatial region. In this study, eleven years of precipitation time series data from both NEXRAD and LDAS are employed to investigate the spatial similarities of these two widely used data products over a subregion of the Ohio river basin, which is bounded by longitude (-88.000, -84.000) and latitude (37.750, 41.750). In addition to the analyses of histograms, conditional histograms and spatial correlations, three metrics are employed to evaluate the spatial similarities and differences of these two data products. The first is the Forecast Quality Index (FQI) metric which consists of a numerator of a normalized Hausdorff distance and a denominator that relates to the magnitude. The second is a displacement based metric known as Forecast Quality Measure (FQM) which was originally developed for tracking image motion. This metric has the strength of considering impacts of neighbors in evaluating the spatial patterns. The third is the Kappa coefficient, which considers implicit impacts of both magnitude and location of each cell on spatial patterns and is extensively used in the image processing field. Analyses are also conducted for warm and cold seasons to examine the seasonal characteristics of each data product. Our initial results show that significant differences exist between the Stage III NEXRAD and LDAS combo precipitation products, and that a single metric cannot represent adequately the spatial characteristics of these two precipitation data products. A more comprehensive metric including magnitude, distance, shape and neighborhood is needed to evaluate adequately the spatial similarities and differences between two precipitation data products.
Obtaining DDF Curves of Extreme Rainfall Data Using Bivariate Copula and Frequency Analysis
The traditional rainfall intensity-duration-frequency (IDF) curve is a reliable approach for representing the variation of rainfall intensity with duration for a given return period. In reality rainfall variables intensity, depth and duration are dependent and therefore a bivariate analysis using copulas can give a more accurate IDF curve. We study IDF curves using a copula in a bivariate frequency analysis of extreme rainfall. To be able to choose the most suitable copula among candidate copulas (i.e., Gumbel, Clayton, and Frank) we demonstrated IDF curves based on variation of depth with duration for a given return period and name them DDF (depth-duration-frequency) curves. The copula approach does not assume the rainfall variables are independent or jointly normally distributed. Rainfall series are extracted in three ways: (1) by maximum mean intensity; (2) by depth and duration of individual rainfall events; and (3) by storage volume and duration. In each case we used partial duration series (PDS) to extract extreme rainfall variables. The DDF curves derived from each method are presented and compared. This study examines extreme rainfall data from catchment Vedbæ k Renseanlæ g, situated near Copenhagen in Denmark. For rainfall extracted using method 2, the marginal distribution of depth was found to fit the Generalized Pareto distribution while duration was found to fit the Gamma distribution, using the method of L-moments. The volume was fit with a generalized Pareto distribution and the duration was fit with a Pearson type III distribution for rainfall extracted using method 3. The Clayton copula was found to be appropriate for bivariate analysis of rainfall depth and duration for both methods 2 and 3. DDF curves derived using the Clayton copula for depth and duration of individual rainfall events (method 2) are in agreement with empirically derived DDF curves obtained from maximum mean intensity (method 1) for a 10-year return period. For a 100-year return period the estimates differ by 2.5 cm for a 5 hr duration. This difference diminishes to almost zero for a 50 hr duration. If rainfall series are extracted by storage volume and duration (method 3), the difference between DDF curves derived from the Clayton copula and the empirical DDF curves are more appreciable and in general, DDF curves derived from method 3 show a smaller depth for the same duration for any selected return period. The differences between DDF curves illustrates that the method of extracting extreme rainfall as well as the frequency analysis approach have a considerable effect on the resulting DDF curves.
An Apparent Paradox in Verification of Rainfall Estimates.
A problem that is a source of permanent cognitive confusion in comprehensive evaluations of different rainfall estimates is presented. The problem stems from the existence of two conditional biases (CB) inherent to the uncertainties of the estimates. The two CBs, called "CB type 1" and "CB type 2," are recognized by researchers familiar with the distribution-oriented framework for complete verification of hydrological and meteorological products. Although the mathematical definitions of the two CBs are clear, a reality check reveals that their meaningful interpretation is problematic. It can even result in self-contradictory conclusions suggesting both systematic overestimation and underestimation of strong rainfall by the same rainfall estimation products. A solution to this apparent paradox is discussed. This investigation is based on large data samples of different radar rainfall estimates and the corresponding highly accurate ground reference. Understanding the two CBs, their physical consequences and the fundamental inter-relations between them is essential for informed usage of these uncertainty characteristics.
Estimating Snow Volume in the Elbow River Watershed Using Airborne Lidar
In this research snow pack modeling was attempted in the Elbow River watershed, west of Calgary, Alberta using lidar derived elevation data. The City of Calgary and the Department of Sustainable Resources Development (SRD) were interested in determining whether a winter and a summer lidar dataset can be used to estimate the mean snow depth. Lidar is an airborne laser system that calculates the distance to the ground by determining the return time of emitted laser pulses spatially located by a survey grade global positioning system (GPS) and an inertial motion unit (IMU). Subtraction (Digital Elevation Model (DEM) change detection) of the summer dataset from the winter dataset provides a snow depth dataset that is used to determine mean snow depth. Mean snow depth and an average field-measured snow densities were used to calculate snow water equivalent (SWE). An estimate of snow volume was determined using three methods, 1) one mean snow depth, 2) four terrain attributes (slope, aspect, elevation, and canopy fractional cover) individually and 3) a multiple terrain attribute GIS approach. Application of an average snow depth (3.4 x 107m3) rendered a similar approximate value for snow water equivalent for the study site as the results from slope (3.6 x 107m3), aspect (3.5 x 107m3) and canopy fractional cover (3.5 x 107m3) terrain attributes. Elevation (4.2 x 107m3) and the GIS model (4.3 x 107m3) gave higher estimates of snow water equivalent in the Elbow River watershed since elevation plays a strong role in snow accumulation. Future research should include validation of lidar runoff values with stream gauge data; as well re-evaluating the methods proposed in an area of greater snow depth (the average snow depth in the Elbow River watershed was 18cm, which is the accuracy limit of current lidar systems). Preliminary results indicate that the use of lidar to estimate snow depth is viable option for the determination of snow depth in a mountainous region. Application of this research can be used in conjunction with current water resource management strategies to assist in prediction of seasonal runoff volumes. The advancement of water volume prediction for watersheds can aid city planners with regulating water supply as well as prepare for flooding events.