Applied Weather Associates, LLC (AWA)
AWA - Storm Precipitation Analysis System (SPAS) Brochure
The STORM PRECIPITATION ANALYSIS SYSTEM (SPAS) Table of Contents INTRODUCTION .................................................................................................................................................................. 2 SETUP .................................................................................................................................................................................... 3 Analysis Domain ................................................................................................................................................................. 4 Analysis Time Frame .......................................................................................................................................................... 4 DATA ..................................................................................................................................................................................... 4 Precipitation Gauge Data .................................................................................................................................................... 4 Hourly Precipitation Data ............................................................................................................................................... 5 Daily Precipitation Data .................................................................................................................................................. 6 Supplemental Precipitation Data ..................................................................................................................................... 6 Daily and Supplemental Precipitation to Hourly ............................................................................................................ 6 Basemap .............................................................................................................................................................................. 8 Radar Data .......................................................................................................................................................................... 9 GAUGE QUALITY CONTROL .......................................................................................................................................... 12 Mass Curve Check ............................................................................................................................................................ 12 Gauge Mis-location Check................................................................................................................................................ 12 Co-located Gauge QC ....................................................................................................................................................... 12 SPATIAL INTERPOLATION ............................................................................................................................................. 13 Basic Approach ................................................................................................................................................................. 13 Basemap Approach ........................................................................................................................................................... 13 Radar Approach ................................................................................................................................................................ 14 Z-R Relationship ........................................................................................................................................................... 14 Dual Polarization Radar ................................................................................................................................................ 15 Radar-aided Hourly Precipitation Grids............................................................................................................................ 16 Radar- and Basemap-Aided Hourly Precipitation Grids ................................................................................................... 16 SPAS versus Gauge Precipitation ..................................................................................................................................... 17 OUTPUT ............................................................................................................................................................................... 20 SUMMARY .......................................................................................................................................................................... 20 REFERENCES ..................................................................................................................................................................... 22 INTRODUCTION The Storm Precipitation Analysis System (SPAS) is grounded on years of scientific research with a demonstrated reliability in hundreds of post-storm precipitation analyses. It has evolved into a trusted hydrometeorological tool that provides accurate precipitation data at a high spatial and temporal resolution for use in a variety of sensitive hydrologic applications (Faulkner et al., 2004, Tomlinson et al., 2006 and 2008). In partnership with Applied Weather Associates, LLC, METSTAT, Inc. initially developed SPAS in 2002 for use in producing Depth-Area-Duration values for Probable Maximum Precipitator (PMP) analyses. SPAS utilizes precipitation gauge data, “basemaps” and radar data (when available) to produce gridded precipitation at time intervals as short as 5-minutes, at spatial scales as fine as 1 km2 and in a variety of customizable formats. To date (October 1, 2011) SPAS has been used to analyze over 240 storm centers (see Figure 1) across all types of terrain, among highly varied meteorological settings and some occurring over 100-years ago. a) b) c) Figure 1. SPAS storm locations (a) in/near Arizona, (b) West and (c) East of the continental divide (valid through Jan. 2013). SPAS output has many applications including, but not limited to: hydrologic model calibration/validation, flood event reconstruction, storm water runoff analysis, forensic cases and Probable Maximum Precipitation (PMP) studies. Detailed SPAS-computed precipitation data allow hydrologists to accurately model runoff from basins, particularly when the precipitation is unevenly distributed over the drainage basin or when rain gauge data is limited or not available. The increased spatial and temporal accuracy of precipitation estimates has eliminated the need for commonly made assumptions about precipitation characteristics (such as uniform precipitation over a watershed), thereby greatly improving the precision and reliability of hydrologic analyses. In order to instill consistency in SPAS analyses, many of the core methods have remained consistent from beginning, However, SPAS is constantly evolving and improving through new scientific advancements and as new data and improvements are incorporated. This write-up describes the current inter-workings of SPAS, but the reader should realize SPAS can be customized on a case-by-case basis to account for special circumstances; these adaptations are documented and included in the deliverables. The over arching goal of SPAS is to combine the strengths of rain gauge data and radar data (when available) to provide sound, reliable and accurate spatial precipitation data. Hourly precipitation observations are generally limited to a small number of locations, with many basins lacking observational precipitation data entirely. Meanwhile Next Generation Radar (NEXRAD) data provides valuable spatial and temporal information over data sparse basins, it has historically lacked reliability for determining precipitation rates and reliable quantitative precipitation estimates (QPE). The improved reliability in SPAS is made possible by hourly calibration of the NEXRAD radar-precipitation relationship, combined with local hourly bias adjustments to force consistency between the final result and “ground truth” precipitation measurements. If NEXRAD radar data is available (generally for storm events since the mid-1990's), precipitation at temporal scales as frequent as 5-minutes is available, otherwise the precipitation data is available hourly. A summary of the general SPAS processes are shown in flow chart in Figure 2. Figure 2. SPAS flow chart. SETUP Prior to a SPAS analysis careful definition of the storm analysis domain and time frame to be analyzed is established. Several considerations are made to ensure the domain (longitude-latitude box) and time frame are sufficient for the given application. Analysis Domain For PMP applications it is important to establish an analysis domain that completely encompasses a storm center, meanwhile hydrologic modeling applications are more concerned about a specific basin, watershed or catchment. If radar data is available, then it is also important to establish an area large enough to encompass enough stations (minimum of ~30) to adequately derive reliable radar-precipitation intensity relationships (discussed later). The domain is defined by evaluating existing documentation on the storm as well as plotting and evaluating initial precipitation gauge data on a map. The analysis domain is defined to include as many hourly recording gauges as possible given their importance in timing. The domain must include enough of a buffer to accurately model the nested domain of interest. The domain is defined as a longitude-latitude (upper left and lower right corner) rectangular region. Analysis Time Frame Ideally, the analysis time frame, also referred to as the Storm Precipitation Period (SPP), will extend from a dry period through the target wet period then back into another dry period. This is to ensure that total storm precipitation amounts can be confidently associated with the storm in question and not contaminated by adjacent wet periods. If this is not possible, a reasonable time period is selected that is bounded by relatively lighter precipitation. The time frame of the hourly data must be sufficient to capture the full range of daily gauge observational periods in order for the daily observations to be disaggregated into estimated incremental hourly values (discussed later). For example, if a daily gauge takes observations at 8:00 AM, then the hourly data must be available from 8:00 AM the day prior. Given the configuration of SPAS, the minimum SPP is 72 hours and aligns midnight to midnight. The core precipitation period (CPP) is a sub-set of the Storm Precipitation Period (SPP) and represents the time period with the most precipitation and the greatest number of reporting gauges. The CPP represents the time period of interest and where our confidence in the results is highest. DATA Precipitation Gauge Data The foundation of a SPAS analysis is the “ground truth” precipitation measurements. In fact, the level of effort involved in “data mining” and quality control represent over half of the total level of effort needed to conduct a complete storm analysis. SPAS operates with three primary data sets: precipitation gauge data, a “basemap” and, if available, radar data. Table 1 conveys the variety of precipitation gauges usable by SPAS. For each gauge, the following elements are gathered, entered and archived into to SPAS database: ? Station ID ? Station name ? Station type (H=hourly, D=Daily, S=Supplemental, etc.) ? Longitude in decimal degrees ? Latitude in decimal degrees ? Elevation in feet above MSL ? Observed precipitation ? Observation times ? Source ? If unofficial, the measurement equipment and/or method is also noted. Based on the SPP and analysis domain, hourly and daily precipitation gauge data are extracted from our in-house database as well as the Meteorological Assimilation Data Ingest System (MADIS). Our in-house database is contains data dating back to the late 1800s, while the MADIS system (described below) contains archived data back to 2002. Hourly Precipitation Data Our hourly precipitation database is largely comprised of data from NCDC TD-3240, but also precipitation data from other mesnonets and meteorological networks (e.g. ALERT, Flood Control Districts, etc.) that we have collected and archived as part of previous studies. Meanwhile, MADIS provides data from a large number of networks across the U.S., including NOAA’s HADS (Hydrometeorological Automated Data System), numerous mesonets, the Citizen Weather Observers Program (CWOP), departments of transportation, etc. (see http://madis.noaa.gov/mesonet_providers.html for a list of providers). Although our automatic data extraction is fast, cost-effective and efficient, it never captures all of the available precipitation data for a storm event. For this reason, a thorough “data mining” effort is undertaken to acquire all available data from sources such as U.S. Geological Survey (USGS), Remote Automated Weather Stations (RAWS), Community Collaborative Rain, Hail & Snow Network (CoCoRaHS), National Atmospheric Deposition Program (NADP), Clean Air Status and Trends Network (CASTNET), local observer networks, Climate Reference Network (CRN), Global Summary of the Day (GSD) and Soil Climate Analysis Network (SCAN). Unofficial hourly precipitation are gathered to give guidance on either timing or magnitude in areas otherwise void of precipitation data. The WeatherUnderground and MesoWest, two of the largest weather databases on the Internet, contain a good deal of official data, but also unofficial gauges. Table 1. Different precipitation gauge types used by SPAS. Precipitation Gauge Type Abbreviation Description Hourly H Hourly gauges with complete, or nearly complete, incremental hourly precipitation data. Hourly estimated HE Hourly gauges with some estimated hourly values, but otherwise reliable. Hourly pseudo HP Hourly gauges with reliable temporal precipitation data, but the magnitude is questionable in relation to co-located daily or supplemental gauge. Hourly estimated pseudo HEP Combination of hourly estimated and hourly pseudo Daily D Daily gauge with complete data and known observation times. Daily estimated DE Daily gauges with some or all estimated data. Supplemental S Gauges with unknown or irregular observation times, but reliable total storm precipitation data. (e.g. public reports, storms reports, “Bucket surveys”, etc.) Supplemental estimated SE Gauges with estimated total storm precipitation values based on other information (e.g. newspaper articles, stream flow discharge, inferences from nearby gauges, pre-existing total storm isohyetal maps, etc.) Daily Precipitation Data Our daily database is largely based on NCDC’s TD-3206 (pre-1948) and TD-3200 (1948 through present) as well as SNOTEL data from NRCS. Since the late 1990s, the Community Collaborative Rain, Hail & Snow Network (CoCoRaHS) network of more than 12,000 observes in the U.S. has become a very important daily precipitation source. Other daily data is gathered from similar, but smaller gauge networks, for instance the High Spatial Density Precipitation Network in Minnesota. As part of the daily data extraction process, the time of observation, as indicted in database (if available), accompanies each measured precipitation value. Accurate observation times are necessary for SPAS to disaggregate the daily precipitation into estimated incremental values (discussed later). Knowing the observation time also allows SPAS to maintain precipitation amounts within given time bounds, thereby retaining known precipitation intensities. Given the importance of observation times, efforts are taken to insure the observation times are accurate. Hardcopy reports of “Climatological Data,” scanned observational forms (available on-line) and/or gauge metadata forms have proven to be valuable and accurate resources for validating observation times. Furthermore, erroneous observation times are identified in the mass-curve quality-control procedure (discussed later) and can be corrected at that point in the process. Supplemental Precipitation Data For gauges with unknown or irregular observation times, the gauge is considered a “supplemental” gauge. A supplemental gauge can either be added to the storm database with a storm total and the associated SPP as the temporal bounds or as a gauge with the known, but irregular observation times and associated precipitation amounts. For instance, if all that is known is 3” fell between 0800-0900, then that information can be entered. Gauges or reports with nothing more than a storm total are often abundant, but in order to use them, it is important the precipitation is only from the storm period in question. Therefore, it is ideal to have the analysis time frame bounded by dry periods. Perhaps the most important source of data, if available, is from “bucket surveys,” which provide comprehensive lists of precipitation measurements collected during a post-storm field exercise. Although some bucket survey amounts are not from conventional precipitation gauges, they provide important information, especially in areas lacking data. Particularly for PMP-storm analysis applications, it is customary to accept extreme, but valid non-measured precipitation values in order to capture the highest precipitation values. Daily and Supplemental Precipitation to Hourly To obtain one hour temporal resolutions and utilize all gauge data, it is necessary to disaggregate the daily and supplemental precipitation observations into estimated hourly amounts. This process has traditionally been accomplished by distributing (temporally) the precipitation at each daily/supplemental gauge in accordance to a single nearby hourly gauge (Thiessen polygon approach). However, this may introduce biases and not correctly represent hourly precipitation at daily/supplemental gauges situated in-between hourly gauges. Instead, SPAS uses a spatial approach by which the estimated hourly precipitation at each daily and supplemental gauge is governed by a distance weighted algorithm of all nearby true hourly gauges. In order to disaggregate (i.e. distribute) daily/supplemental gauge data into estimate hourly values, the true hourly gauge data is first evaluated and quality controlled using synoptic maps, nearby gauges, orographic effects, gauge history and other documentation on the storm. Any problems with the hourly data are resolved, and when possible/necessary accumulated hourly values are distributed. If an hourly value is missing, the analyst can choose to either estimate it or leave it missing for SPAS to estimate later based on nearby hourly gauges. At this point in the process, pseudo (hourly) gauges can be added to represent precipitation timing in topographically complex locations, areas with limited/no hourly data or to capture localized convention. In order to adequately capture the temporal variations of the precipitation a pseudo hourly gauge is sometimes necessary. A pseudo gauge is created by distributing the precipitation at a co-located daily gauge or by creating a completely new pseudo gauge from other information such as inferences from COOP observation forms, METAR visibility data (if hourly precipitation isn’t already available), lightning data, satellite data, or radar data. Often radar data is the best/only choice for creating pseudo hourly gauges, but this is done cautiously given the potential differences (over-shooting of the radar beam equating to erroneous precipitation) between radar data and precipitation. In any case, the pseudo hourly gauge is flagged so SPAS only uses it for timing and not magnitude. Care is taken to ensure hourly pseudo gauges represent justifiably important physical and meteorological characteristics before being incorporated into the SPAS database. Although pseudo gauges provide a very important role, their use is kept to a minimum. The importance of insuring the reliability of every hourly gauge cannot be over emphasized. All of the final hourly gauge data, including pseudos, are included in the hourly SPAS precipitation database. Using the hourly SPAS precipitation database, each hourly precipitation value is converted into a percentage that represents the incremental hourly precipitation divided by the total storm precipitation period (SPP) precipitation. The GIS-ready x-y-z file is constructed for each hour that contains the latitude (x), longitude(y) and percent of precipitation (z) for a particular hour. Using the GRASS GIS, an inverse-distance-weighting squared (IDW) interpolation technique is applied to each of the hourly files. The result is a continuous grid with percentage values for the entire analysis domain, keeping the grid cells on which the hourly gauge resides faithful to the observed/actual percentage. Since the percentages typically have a high degree of spatial autocorrelation, the spatial interpolation has skill in determining the percentages between gauges, especially since the percentages are somewhat independent of the precipitation magnitude. The end result is a GIS grid for each hour that represents the percentage of the SPP precipitation that fell during that hour. After the hourly percentage grids are generated and QC’ed for the entire SPP, a program is executed that converts the daily/supplemental gauge data into incremental hourly data. The timing at each of the daily/supplemental gauges is based on (1) the daily/supplemental gauge observation time, (2) daily/supplemental precipitation amount and (3) the series of interpolated hourly percentages extracted from grids (described above). This procedure is detailed in Figure 3 below. In this example, a supplemental gauge reported 1.40" of precipitation during the storm event and is located equal distance from the three surrounding hourly recording gauges. The procedure steps are: Step 1. For each hour, extract the percent of SPP from the hourly gauge-based percentage at the location of the daily/supplemental gauge. In this example, assume these values are the average of all the hourly gauges. Step 2. Multiply the individual hourly percentages by the total storm precipitation at the daily/supplemental gauge to arrive at estimated hourly precipitation at the daily/supplemental gauge. To make the daily/supplemental accumulated precipitation data faithful to the daily/supplemental observations, it is sometimes necessary to adjust the hourly percentages so they add up to 100% and account for 100% of the daily observed precipitation. Figure 3. Example of disaggregation of daily precipitation into estimated hourly precipitation based on three (3) surrounding hourly recording gauges. In cases where the hourly grids do not indicate any precipitation falling during the daily/supplemental gauge observational period, yet the daily/supplemental gauge reported precipitation, the daily/supplemental total precipitation is evenly distributed throughout the hours that make up the observational period; although this does not happen very often, this solution is consistent with NWS procedures. However, the SPAS analyst is notified of these cases in a comprehensive log file, and in most cases they are resolvable, sometimes with a pseudo hourly gauge. Basemap “Basemaps” are independent grids of spatially distributed weather or climate variables that are used to govern the spatial patterns of the hourly precipitation. The basemap also governs the spatial resolution of the final SPAS grids, unless radar data is available/used to govern the spatial resolution. Note that a base map is not required as the hourly precipitation patterns can be based on a station characteristics and an inverse distance weighting technique (discussed later). Basemaps in complex terrain are often based on the PRISM mean monthly precipitation (Figure 4a) or HDSC precipitation frequency grids (Figure 4b) given they resolve orographic enhancement areas and micro-climates at a spatial resolution of 30-seconds (about 800 m). Basemaps of this nature in flat terrain are not as effective given the weak precipitation gradients, therefore basemaps for SPAS analyses in flat terrain are often developed from pre-existing (hand-drawn) isohyetal patterns (Figure 4c), composite radar imagery or a blend of both. a) b) c) Figure 4. Sample SPAS “basemaps:” (a) A pre-existing (USGS) isohyetal pattern across flat terrain (SPAS #1209), (b) PRISM mean monthly (October) precipitation (SPAS #1192) and (c) A 100-year 24-hour precipitation grid from NOAA Atlas 14 (SPAS #1138). Radar Data For storms occurring since approximately the mid-1990's, weather radar data is available to supplement the SPAS analysis. A fundamental requirement for high quality radar-estimated precipitation is a high quality radar mosaic, which is a seamless collection of concurrent weather radar data from individual radar sites, however in some cases a single radar is sufficient (i.e. for a small area size storm event such as a thunderstorm). Weather radar data has been in use by meteorologists since the 1960’s to estimate precipitation depths, but it was not until the early 1990’s that new, more accurate NEXRAD Doppler radar (WSR88D) was placed into service across the United States. Currently efforts are underway to convert the WSR88D radars to dual polarization (DualPol) radar. Today, NEXRAD radar coverage of the contiguous United States is comprised of 159 operational sites and 30 in Canada. Each U.S. radar covers an approximate 285 mile (460 km) radial extent and while Canadian radars have approximately 256 km (138 nautical miles) radial extent over which the radar can detect precipitation (See Figure 5). The primary vendor of NEXRAD weather radar data for SPAS is Weather Decision Technologies, Inc. (WDT), who accesses, mosaics, archives and quality-controls NEXRAD radar data from NOAA and Environment Canada. SPAS utilizes Level II NEXRAD radar reflectivity data in units of dBZ, available every 5-minutes in the U.S. and 10-minutes in Canada. Figure 5. U.S. radar locations and their radial extents of coverage below 10,000 feet above ground level (AGL). Each U.S. radar covers an approximate 285 mile (460 km) radial extent over which the radar can detect precipitation. The WDT and National Severe Storms Lab (NSSL) Radar Data Quality Control Algorithm (RDQC) removes non-precipitation artifacts from base Level–II radar data and remaps the data from polar coordinates to a Cartesian (latitude/longitude) grid. Non-precipitation artifacts include ground clutter, bright banding, sea clutter, anomalous propagation, sun strobes, clear air returns, chaff, biological targets, electronic interference and hardware test patterns. The RDQC algorithm uses sophisticated data processing and a Quality Control Neural Network (QCNN) to delineate the precipitation echoes caused by radar artifacts (Lakshmanan and Valente, 2004). Beam blockages due to terrain are mitigated by using 30m DEM data to compute and then discard data from a radar beam that clears the ground by less than 50m and incurs more than 50% power blockage. A clear-air echo removal scheme is applied to radars in clear-air mode when there is no precipitation reported from observation gauges within the vicinity of the radar. In areas of radar coverage overlap, a distance weighting scheme is applied to assign reflectivity to each grid cell, for multiple vertical levels. This scheme is applied to data from the nearest radar that is unblocked by terrain. Once the data from individual radars have passed through the RDQC, they are merged to create a seamless mosaic for the United States and southern Canada as shown in Figure 6. A multi-sensor quality control can be applied by post-processing the mosaic to remove any remaining “false echoes”. This technique uses observations of infra-red cloud top temperatures by GOES satellite and surface temperature to create a precipitation/no-precipitation mask. Figure 6 shows the impact of WDT’s quality control measures. Upon completing all QC, WDT converts the radar data from its native polar coordinate projection (1 degree x 1.0 km) into a longitude-latitude Cartesian grid (based on the WGS84 datum), at a spatial resolution of 1 km2(~1/3rd mi2) for processing in SPAS. a) b) Figure 6. (a) Level-II radar mosaic of CONUS radar with no quality control, (b) WDT quality controlled Level-II radar mosaic. SPAS conducts further QC on the radar mosaic by infilling areas contaminated by beam blockages. Beam blocked areas are objectively determined by evaluating a total storm reflectivity grid which naturally amplifies areas of the SPAS analysis domain suffering from beam blockage as shown in Figure 7. a) b) Figure 7. Illustration of SPAS-beam blockage infilling where (a) is raw, blocked radar and (b) is filled for a 42-hour storm event. GAUGE QUALITY CONTROL Exhaustive quality control measures are taken throughout the SPAS analysis. Below are the most significant QC measures taken: Mass Curve Check A mass curve-based QC-methodology is used to ensure the timing of precipitation at all gauges is consistent with nearby gauges. SPAS groups each gauge with the nearest four gauges (regardless of type) into a single file. These files are subsequently used in software for graphing and evaluation. Unusual characteristics in the mass curve are investigated and the gauge data corrected, if possible and warranted. See Figure 8 for an example. Figure 8. Sample mass curve plot depicting a precipitation gauge with an erroneous observation time (blue line). X-axis is the SPAS index hour and the y-axis is inches. The statistics in the upper left denote gauge type, distance from target gauge (in km), and station ID. Gauge Mis-location Check Although the gauge elevation is not explicitly used in SPAS, it is however used as a means of QCing gauge location. Gauge elevations are compared to a high-resolution 15-second DEM to identify gauges with large differences, which may indicate erroneous longitude and/or latitude values. Co-located Gauge QC Care is also taken to establish the most accurate precipitation depths at all co-located gauges. In general, where a co-located gauge pair exists, the highest precipitation is accepted (if accurate). If the hourly gauge reports higher precipitation, then the co-located daily (or supplemental) is removed from the analysis since it would not add anything to the analysis. Often daily (or supplemental) gauges report greater precipitation than a co-located hourly station since hourly tipping bucket gauges tend to suffer from gauge under-catch, particularly during extreme events, due to loss of precipitation during tips and high winds. In these cases the daily/supplemental is retained for the magnitude and the hourly used as a pseudo hourly gauge for timing. Large discrepancies between any co-located gauges are investigated and resolved since SPAS can only utilize a single gauge magnitude at each co-located site. SPATIAL INTERPOLATION At this point the QCed observed hourly and disaggregated daily/supplemental hourly precipitation data are spatially interpolated into hourly precipitation grids. SPAS has three options for conducting the hourly precipitation interpolation, depending on the terrain and availability of radar data, thereby allowing SPAS to be optimized for any particular storm type or location. Figure 9 depicts the results of each spatial interpolation methodology based on the same precipitation gauge data. a) b) c) Figure 9. Depictions of total storm precipitation based on the three SPAS interpolation methodologies for a storm (SPAS #1177, Vanguard, Canada) across flat terrain: (a) no basemap, (b) basemap-aided and (3) radar. Basic Approach The basic approach interpolates the hourly precipitation point values to a grid using an inverse distance weighting squared GIS algorithm. This is sometimes the best choice for convective storms over flat terrain when radar data is not available, yet high gauge density instills reliable precipitation patterns. This approach is rarely used. Basemap Approach Another option includes the use of a “basemap”, also known as a climatologically-aided interpolation (Hunter, 2005). As noted before, the spatial patterns of the basemap govern the interpolation between points of hourly precipitation estimates, while the actual hourly precipitation values govern the magnitude. This approach to interpolating point data across complex terrain is widely used. In fact, it was used extensively by the NWS during their storm analysis era from the 1940s through the 1970s. In application, the hourly precipitation gauge values are first normalized by the corresponding grid cell value of the basemap before being interpolated. The normalization allows information and knowledge from the basemap to be transferred to the spatial distribution of the hourly precipitation. Using an IDW squared algorithm, the normalized hourly precipitation values are interpolated to a grid. The resulting grid is then multiplied by the basemap grid to produce the hourly precipitation grid. This is repeated each hour of the storm. Radar Approach The coupling of SPAS with NEXRAD provides the most accurate method of spatially and temporally distributing precipitation. To increase the accuracy of the results however, quality-controlled precipitation observations are used for calibrating the radar reflectivity to rain rate relationship (Z-R relationship) each hour instead of assuming a default Z-R relationship. Also, spatial variability in the Z-R relationship is accounted for through local bias corrections (described later). The radar approach involves several steps, each briefly described below. The radar approach cannot operate alone – either the basic or basemap approach must be completed before radar data can be incorporated. Z-R Relationship SPAS derives high quality precipitation estimates by relating quality controlled level–II NEXRAD radar reflectivity radar data with quality-controlled precipitation gauge data in order to calibrate the Z-R (radar reflectivity, Z, and rainfall, R) relationship. Optimizing the Z-R relationship is essential for capturing temporal changes in the Z-R. Most current radar-derived precipitation techniques rely on a constant relationship between radar reflectivity and precipitation rate for a given storm type (e.g. tropical, convective), vertical structure of reflectivity and/or reflectivity magnitudes. This non-linear relationship is described by Equation 1 below: Equation 1. Z = A Rb Where Z is the radar reflectivity (measured in units of dBZ), R is the rainfall (precipitation) rate (millimeters per hour), A is the “multiplicative coefficient” and b is the “power coefficient”. Both A and b are directly related to the rain drop size distribution (DSD) and rain drop number distribution (DND) within a cloud (Martner and Dubovskiy, 2005). The variability in the results of Z versus R is a direct result of differing DSD, DND and air mass characteristics (Dickens, 2003). The DSD and DND are determined by complex interactions of microphysical processes that fluctuate regionally, seasonally, daily, hourly, and even within the same cloud. For these reasons, SPAS calculates an optimized Z-R relationship across the analysis domain each hour based on observed precipitation rates and radar reflectivity (see Figure 11). The National Weather Service (NWS) utilizes different default Z-R algorithms, depending on the precipitation-causing event, to estimate precipitation through the use of NEXRAD radar reflectivity data across the United States (See Figure 11) (Baeck and Smith, 1998 and Hunter, 1999). A default Z-R relationship of Z = 300R1.4 is the primary algorithm used throughout the continental U.S. However, it is widely known that this, compared to unadjusted radar-aided estimates of precipitation, suffers from deficiencies that may lead to significant over or under-estimation of precipitation. Figure 10. Example SPAS (denoted as “Exponential”) vs. default Z-R relationship (SPAS #1218, Georgia September 2009). Figure 11. Commonly used Z-R algorithms used by the National Weather Service. Instead of adopting a standard Z-R, SPAS utilizes a least squares fit procedure for optimizing the Z-R relationship each hour of the SPP. The process begins by determining if sufficient (minimum 12) observed hourly precipitation and radar data pairs are available to compute a reliable Z-R. If insufficient (<12) gauge pairs are available, then SPAS adopts the previous hour Z-R relationship, if available, or applies a user-defined default Z-R algorithm from Figure 9. If sufficient data are available, the one hour sum of NEXRAD reflectivity (Z) is related to the 1-hour precipitation at each gauge. A least-squares-fit exponential function using the data points is computed. The resulting best-fit, one hour-based Z-R is subjected to several tests to determine if the Z-R relationship and its resulting precipitation rates are within a certain tolerance based on the R-squared fit measure and difference between the derived and default Z-R precipitation results. Experience has shown the actual Z-R versus the default Z-R can be significantly different as shown in Figure 13. Dual Polarization Radar During 2011-2013 the National Weather Service deployed Dual Polarization (Dual-Pol) at all WSR-88D NEXRAD radar sites across the United States. The standard WSR-88D Doppler radar transmits and receives information horizontally, while the Dual-Pol transmits and receives information both horizontally and vertically which allows the radar to determine approximate sizes and shapes of objects in the atmosphere. The size and shape of objects in the atmosphere are illustrated through three new base products: Differential Reflectivity (ZDR), Correlation Coefficient (CC), and Specific Differential Phase (KDP). The ZDR shows the difference between the horizontal and vertical reflectivity factors (units dBz). The CC indicates the similarity of the horizontal and vertical pulses. The KDP is the range derivative of the differential phase shift between the horizontal and vertical pulse phases. These products allow for better estimates of radar-estimated precipitation amounts. Additionally, the Dual-Pol radars provide higher spatial resolution reflectivity data – 250 m2 versus 1 km2 (See Figure ). The suite of Dual-Pol products are being incorporated into SPAS, with research and development continuing to ensure proper use and quantification of the data in the rainfall analyses. Currently, SPAS is utilizing the higher spatial resolution (250 m2) standard reflectivity data for storms when available. This improves the accuracy of localized precipitation events and decrease the difference between the average radar pixel value and observed point gauge values. 1-km results 250-m results Figure 12. Total storm precipitation comparison of (a) standard 1-km radar and (b) 250-m Dual-Pol radar resolution. Radar-aided Hourly Precipitation Grids Once a mathematically optimized hourly Z-R relationship is determined, it is applied to the total hourly Z grid to compute an initial precipitation rate (inches/hour) at each grid cell. To account for spatial differences in the Z-R relationship, SPAS computes residuals, the difference between the initial precipitation analysis (via the Z-R equation) and the actual “ground truth” precipitation (observed – initial analysis), at each gauge. The point residuals, also referred to as local biases, are normalized and interpolated to a residual grid using an inverse distance squared weighting algorithm. A radar-based hourly precipitation grid is created by adding the residual grid to the initial grid; this allows the precipitation at the grid cells for which gauges are “on” to be true and faithful to the gauge measurement. The pre-final radar-aided precipitation grid is subject to some final, visual QC checks to ensure the precipitation patterns are consistent with the terrain; these checks are particularly important in areas of complex terrain where even QCed radar data can be unreliable. Radar- and Basemap-Aided Hourly Precipitation Grids At this stage of the radar approach, a radar- and basemap-aided hourly precipitation grid exists for each hour. At locations with precipitation gauges, the grids are equal, however elsewhere the grids can vary for a number of reasons. For instance, the basemap-aided hourly precipitation grid may depict heavy precipitation in an area of complex terrain, blocked by the radar, whereas the radar-aided hourly precipitation grid may suggest little, if any, precipitation fell in the same area. Similarly, the radar-aided hourly precipitation grid may depict an area of heavy precipitation in flat terrain that the basemap-approach missed since the area of heavy precipitation occurred in an area without gauges. SPAS uses an algorithm to compute the hourly precipitation at each pixel given the two results. Areas that are completely blocked from a radar signal are accounted for with the basemap-aided results (discussed earlier). The precipitation in areas with orographically effective terrain and reliable radar data are governed by a blend of the basemap- and radar-aided Figure 12. Comparison of the SPAS optimized hourly Z-R relationships (black lines) versus a default Z=75R2.0 Z-R relationship (red line) for a period of 99 hours for a storm over southern California. precipitation. Elsewhere, the radar-aided precipitation is used exclusively. This blended approach has proven effective for resolving precipitation in complex terrain, yet retaining accurate radar-aided precipitation across areas where radar data is reliable. Figure 13 illustrates the evolution of final precipitation from radar reflectivity in an area of complex terrain in southern California. a) b) c) d) Figure 13. A series of maps depicting 1-hour of precipitation utilizing (a) inverse distance weighting of gauge precipitation, (b) gauge data together with a climatologically-aided interpolation scheme, (c) default Z-R radar-estimated interpolation (no gauge correction) and (d) SPAS precipitation for a January 2005 storm in southern California, USA. SPAS versus Gauge Precipitation Performance measures are computed and evaluated each hour to detect errors and inconsistencies in the analysis. The measures include: hourly Z-R coefficients, observed hourly maximum precipitation, maximum gridded precipitation, hourly bias, hourly mean absolute error (MAE), root mean square error (RMSE), and hourly coefficient of determination (r2). a) b) Figure 14. Z-R plot (a), where the blue line is the SPAS derived Z-R and the black line is the default Z-R, and the (b) associated observed versus SPAS scatter plot at gauge locations. Comparing SPAS-calculated precipitation (Rspas) to observed point precipitation depths at the gauge locations provides an objective measure of the consistency, accuracy and bias. Generally speaking SPAS is usually within 5% of the observed precipitation (see Figure 14). Less-than-perfect correlations between SPAS precipitation depths and observed precipitation at gauged locations could be the result of any number of issues, including: ? Point versus area: A rain gauge observation represents a much smaller area than the area sampled by the radar. The area that the radar is sampling is approximately 1 km2, whereas a rain gauge only samples approximately 8.0x10-9 km2. Furthermore, the radar data represents an average reflectivity (Z) over the grid cell, when in fact the reflectivity can vary across the 1 km2 grid cell. Therefore, comparing a grid cell radar derived precipitation value to a gauge (point) precipitation depth measured may vary. ? Precipitation gauge under-catch: Although we consider gauge data “ground truth,” we recognize gauges themselves suffer from inaccuracies. Precipitation gauges, shielded and unshielded, can inherently underestimate total precipitation due to local airflow, wind under-catch, wetting, and evaporation. The general rule-of-thumb is 1% of the precipitation is lost for every 1 mph. Therefore, a 10 mph wind can cause up to 10% error (under-catch) (Guo et al. 2001, Duchon and Essenberg 2001, Ciach 2003, Tokay et al. 2010). Tipping buckets miss a small amount of precipitation during each tip of the bucket due to the bucket travel and tip time. As precipitation intensities increase, the volumetric loss of precipitation due to tipping tends to increase. Smaller tipping buckets can have higher volumetric losses due to higher tip frequencies, but on the other hand capture higher precision timing. ? Radar Calibration: NEXRAD radars calibrate reflectivity every volume scan, using an internally generated test. The test determines changes in internal variables such as beam power and path loss of the receiver signal processor since the last off-line calibration. If this value becomes large, it is likely that there is a radar calibration error that will translate into less reliable precipitation estimates. The calibration test is supposed to maintain a reflectivity precision of 1 dBZ. A 1 dBZ error can result in an error of up to 17% in Rspas using the default Z-R relationship Z=300R1.4. Higher calibration errors will result in higher Rspas errors. However, by performing correlations each hour, the calibration issue is minimized in SPAS. ? Attenuation: Attenuation is the reduction in power of the radar beams’ energy as it travels from the antenna to the target and back. It is caused by the absorption and the scattering of power from the beam by precipitation. Attenuation can result in errors in Z as large as 1 dBZ especially when the radar beam is sampling a large area of heavy precipitation. In some cases, storm precipitation is so intense (>12 inches/hour) that individual storm cells become “opaque” and the radar beam is totally attenuated. Armed with sufficient gauge data however, SPAS will overcome attenuation issues. ? Range effects: The curvature of the Earth and radar beam refraction result in the radar beam becoming more elevated above the surface with increasing range. With the increased elevation of the radar beam comes a decrease in Z values due to the radar beam not sampling the main precipitation portion of the cloud (i.e. “over topping” the precipitation and/or cloud altogether). Additionally, as the radar beam gets further from the radar, it naturally samples a larger and larger area, therefore amplifying point versus area differences (described above). ? Radar Beam Occultation/Ground Clutter: Radar occultation (beam blockage) results when the radar beam’s energy intersects terrain features as depicted in Figure 15. The result is an increase in radar reflectivity values that can result in higher than normal precipitation estimates. The WDT processing algorithms account for these issues, but SPAS uses GIS spatial interpolation functions to infill areas suffering from poor or no radar coverage. ? Anomalous Propagation (AP) - AP is false reflectivity echoes produced by unusual rates of refraction in the atmosphere. WDT algorithms remove most of the AP and false echoes, however in extreme cases the air near the ground may be so cold and dense that a radar beam that starts out moving upward is bent all the way down to the ground. This produces erroneously strong echoes at large distances from the radar. Again, equipped with sufficient gauge data, the SPAS bias corrections will overcome AP issues. Figure 15. Depiction of radar artifacts. (Source: Wikipedia) SPAS is designed to overcome many of these short-comings by carefully using radar data for defining the spatial patterns and relative magnitudes of precipitation, but allowing measured precipitation values (“ground truth”) at gauges to govern the magnitude. When absolutely necessary, the observed precipitation values at gauges are nudged up (or down) to force the SPAS results to be consistent with observed gauge values. Nudging gauge precipitation values helps to promote better consistency between the gauge value and the gridcell value, even though these two values sometimes should not be the same since they are sampling different area sizes. For reasons discussed in the "SPAS versus Gauge Precipitation" section, the gauge value and gridcell value can vary. Plus, SPAS is designed to toss observed individual hourly values that are grossly inconsistent with the radar data, hence driving a difference between the gauge and gridcell. In general, when the gauge and gridcell value differ by more than 15% and/or 0.50 inches, and the gauge data has been validated, then it is justified to nudge (artificially increase or decrease) the observed gauge value to "force" SPAS to derive a gridcell value equal to the observed value. Sometimes simply shifting the gauge location to an adjacent gridcell resolves the problems. Regardless, a large gauge versus gridcell difference is a "red flag" and sometimes the result of an erroneous gauge value or a mis-located gauge, but in some cases the difference can only be resolved by nudging the precipitation value. Before final results are declared, a precipitation intensity check is conducted to ensure the spatial patterns and magnitudes of the maximum storm intensities at 1-, 6-, 12-, etc. hours are consistent with surrounding gauges and published reports. Any erroneous data are corrected and SPAS re-run. Considering all of the QA/QC checks in SPAS, it typically requires 5-15 basemap SPAS runs and, if radar data is available, another 5-15 radar-aided runs, to arrive at the final output. OUTPUT Armed with accurate, high-resolution precipitation grids, a variety of customized output can be created (see Figure 16). Among the most useful outputs are sub-hourly precipitation grids for input into hydrologic models. Sub-hourly (i.e. 5-minute) precipitation grids are created by applying the appropriate optimized hourly Z-R (scaled down to be applicable for instantaneous Z) to each of the individual 5-minute radar scans; 5-minutes is often the native scan rate of the radar in the US. Once the scaled Z-R is applied to each radar scan, the resulting precipitation is summed up. The proportion of each 5-minute precipitation to the total 1-hour radar-aided precipitation is calculated. Each 5-minute proportion (%) is then applied to the quality controlled, bias corrected 1-hour total precipitation (created above) to arrive at the final 5-minute precipitation for each scan. This technique ensures the sum of 5-minute precipitation equals that of the quality controlled, bias corrected 1-hour total precipitation derived initially. Depth-area-duration (DAD) tables/plots, shown in Figure 16d, are computed using a highly-computational extension to SPAS. DADs provide an objective three dimensional (magnitude, area size, and duration) perspective of a storms’ precipitation. SPAS DADs are computed using the procedures outlined by the National Weather Service Technical Paper 1 (1946). SUMMARY Grounded on years of scientific research with a demonstrated reliability in post-storm analyses, SPAS is a hydro-meteorological tool that provides accurate precipitation analyses for a variety of applications. SPAS has the ability to compute precise and accurate results by using sophisticated timing algorithms, “basemaps”, a variety of precipitation data and most importantly NEXRAD weather radar data (if available). The approach taken by SPAS relies on hourly, daily and supplemental precipitation gauge observations to provide quantification of the precipitation amounts while relying on basemaps and NEXRAD data (if available) to provide the spatial distribution of precipitation between precipitation gauge sites. By determining the most appropriate coefficients for the Z-R equation on an hourly basis, the approach anchors the precipitation amounts to accepted precipitation gauge data while using the NEXRAD data to distribute precipitation between precipitation gauges for each hour of the storm. Hourly Z-R coefficient computations address changes in the cloud microphysics and storm characteristics as the storm evolves. Areas suffering from limited or no radar coverage, are estimated using the spatial patterns and magnitudes of the independently created basemap precipitation grids. Although largely automated, SPAS is flexible enough to allow hydro-meteorologists to make important adjustments and adapt to any storm situation. a) b) c) d) stid, stname, stntype, lon_dd, lat_dd, elev_ft, Robs_in, Rspas_in, Pct, delta_in,hrs_miss, TotalZ,BBMask 4500,Cesar Chavez Park AZ , H,-112.1422, 33.3693, 1060, 0.94, 0.94, +0.0%, 0.00, 0, 1354.30, N 4575,Laveen Basin AZ , H,-112.1508, 33.3904, 1015, 0.79, 0.79, +0.0%, 0.00, 0, 1380.70, N 4665,East Fork Cave Cr.Ave. AZ , H,-112.0807, 33.6279, 1340, 0.55, 0.55, +0.0%, 0.00, 0, 2022.20, N 4700,Durango Complex AZ , H,-112.1186, 33.4266, 1050, 0.63, 0.64, +1.6%, 0.01, 0, 1588.55, N 4710,Jefferson St. @ 4th Ave. AZ , H,-112.0795, 33.4467, 1080, 0.44, 0.45, +2.3%, 0.01, 0, 1874.05, N 4755,Salt River @ 67th Ave. AZ , H,-112.2040, 33.3998, 980, 0.52, 0.53, +1.9%, 0.01, 0, 1564.60, N 4760,Maryvale Municipal Golf Cou AZ, H,-112.1878, 33.4986, 1110, 0.75, 0.75, +0.0%, 0.00, 0, 1758.40, N 4765,Buckeye Rd. @ 75th Ave. AZ , H,-112.2219, 33.4412, 1025, 0.51, 0.51, +0.0%, 0.00, 0, 1570.45, N 4770,City of Glendale AZ , H,-112.1936, 33.5436, 1150, 1.06, 1.04, -1.9%, -0.02, 0, 1800.65, N ? n = Index number ? stid = Station ID ? stname = Station name ? stntype = Station type (H=hourly, D=Daily, etc.) ? lon_dd = Longitude in decimal degrees ? lat_dd = Latitude in decimal degrees ? elev_ft = Elevation above MSL in feet ? Robs_in = Observed rainfall total ? Rspas_in = SPAS rainfall total at pixel ? Pct = Percent difference between Robs_in and Rspas_in ? delta_in = Difference between Robs_in and Rspas_in ? hrs_missing = Number of hours with missing data ? TotalZ = Storm total NEXRAD radar reflectivity ? BBMask = Radar beam blockage mask flag (Y/N) e) Figure 16. Various examples of SPAS output, including (a) total storm map and its associated (b) basin average precipitation time series, (c) total storm precipitation map, (d) depth-area-duration (DAD) table and plot, and (e) precipitation gauge catalog with total storm statistics. REFERENCES Baeck ML, Smith J.A., 1998. “Rainfall Estimation by the WSR-88D for Heavy Rainfall Events”, Weather and Forecasting: Vol. 13, No. 2, pp. 416–436. Ciach, G.J., 2003: Local Random Errors in Tipping-Bucket Rain Gauge Measurements. J. Atmos. Oceanic Technol., 20, 752–759. Clarke, Beth, Chad Kudym and Bruce Rindahl, 2009. “Gauge-Adjusted Radar Rainfall Estimation and Basin Averaged Rainfall for Use in Local Flash Flood Prediction and Runoff Modeling”, In. Proc. of the 23rd Conference on Hydrology, Phoenix, AZ, January 10-16, 2009. Dickens, J., 2003. “On the Retrieval of Drop Size Distribution by Vertically Pointing Radar”, American Meteorological Society 32nd Radar Meteorology Conference, Albuquerque, NM, October 2005. Duchon, C.E., and G.R. Essenberg, 2001: Comparative Rainfall Observations from Pit and Above Ground Rain Gauges with and without Wind Shields, Water Resources Research, Vol. 37, N. 12, 3253-3263. Faulkner, Ellen, Terry Hampton, Richard M. Rudolph, and Edward M. Tomlinson, 2004. “Technological Updates for PMP and PMF – Can They Provide Value for Dam Safety Improvements?,” Association of State Dam Safety Officials Annual Conference, Phoenix, Arizona, September 26-30, 2004 Guo, J. C. Y., Urbonas, Ben, and Stewart, Kevin, 2001. Rain Catch under Wind and Vegetal Effects. ASCE, Journal of Hydrologic Engineering, Vol. 6, No. 1. Hartzell, C.L., and A.B. Super, 2000: Development of a WSR-88D based Snow Accumulation Algorithm for Quantitative Precipitation Estimates over southwestern Oregon. Preprints, 16th International Conference on Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography and Hydrology, American Meteorological Society, Long Beach, CA, pp 373-376. Hultstrand, D.M., T.W. Parzybok, E.M. Tomlinson and W.B. Kappel, 2008. “Advanced Spatial and Temporal Rainfall Analyses for Use in Watershed Models”, In U.S. Geological Survey Scientific Investigations Report: Proceedings of the Third Interagency Conference on Research in the Watersheds, Estes Park, CO, September 8-11 2008. Hunter, Richard D. and Ross K. Meentemeyer, 2005. “Climatologically Aided Mapping of Daily Precipitation and Temperature”. Journal of Applied Meteorology, October 2005, Vol. 44, pp. 1501-1510. Hunter, Steven M., 1999. “Determining WSR-88D Precipitation Algorithm Performance Using The Stage III Precipitation Processing System”, Next Generation Weather Radar Program, WSR-88D Operational Support Facility, Norman, OK. Lakshmanan, V. and M. Valente, 2004. “Quality control of radar reflectivity data using satellite data and surface observations”, 20th Int’l Conf. on Inter. Inf. Proc. Sys. (IIPS) for Meteor., Ocean., and Hydr., Amer. Meteor. Soc., Seattle, CD-ROM, 12.2. Martner, B.E, and Vladimir Dubovskiy, 2005: Z-R Relations from Raindrop Disdrometers: Sensitivity To Regression Methods And DSD Data Refinements. 32nd Radar Meteorology Conference, Albuquerque, NM, October, 2005. Parzybok, Tye W., and Edward M. Tomlinson, 2006. “A New System for Analyzing Precipitation from Storms”, Hydro Review, Vol. XXV, No. 3, 58-65. Parzybok, Tye W., Douglas M. Hultstrand, Edward M. Tomlinson, Ph.D. and William D Kappel, 2009. “How Six Recent Extreme Pacific Northwest Storms Compare to Historical Storms in HMR 57”, Western Regional ASDSO Conference, Coeur d'Alene, Idaho, May 3-6, 2009. Rudolph, Richard M, Ellen Faulkner, Terry Hampton, Edward Tomlinson, 2004: Old Problems and New Technologies: Spatial Analysis Adds Value to Dam Safety Studies, proceedings of Association of Dam Safety Officials Annual Conference, Technical Session I, Phoenix, Arizona. Tokay, A., P.G. Bashor, and V.L. McDowell, 2010: Comparison of Rain Gauge Measurements in the Mid-Atlantic Region. J. Hydrometeor., 11, 553-565. Tomlinson, E.M., W.D. Kappel, T.W. Parzybok, B. Rappolt, 2006. “Use of NEXRAD Weather Radar Data with the Storm Precipitation Analysis System (SPAS) to Provide High Spatial Resolution Hourly Rainfall Analyses for Runoff Model Calibration and Validation”, ASDSO Annual Conference, Boston, MA. Tomlinson, Edward and Tye Parzybok, 2004: Storm Precipitation Analysis System (SPAS), proceedings of Association of Dam Safety Officials Annual Conference, Technical Session II, Phoenix, Arizona. Tomlinson, Edward M., Ross A. Williams, and Tye W. Parzybok, September 2003: Site-Specific Probable Maximum Precipitation (PMP) Study for the Great Sacandaga Lake / Stewarts Bridge Drainage Basin, Prepared for Reliant Energy Corporation, Liverpool, New York. U.S. Weather Bureau, 1946: Manual for Depth-Area-Duration analysis of storm precipitation. Cooperative Studies Technical Paper No. 1, U.S. Department of Commerce, Weather Bureau, Washington, D.C., 73pp.
Most popular related searches