QPESUMS User's Guide

Functional description

The algorithm, DATA QUALITY, removes bad data due to residual ground clutter and anomalous propagation (AP).

There are 2 modes of checking for AP. The first, "Reflectivity_mode," requires a reflectivity vertical continuity test using the hybrid scan. If there is a decrease in reflectivity from the first to the second tilt IN THE HYBRID SCAN of more than 90%, then the reflectivity is flagged as missing. It is important to keep in mind that the hybrid scan is used (as opposed to the lowest 2 elevation angles); this test MAY use elevation angles of 2.4 deg and 3.3 deg to perform this test. (SV comment: seems like this test could over-do removal since there very well could be a large decrease from 2.4 to 3.3 deg, even from .4 to 2.2 deg; only 1st 2 tilts should be used in stratiform rain/snow).

The second "Velocity_mode" checks for near-zero velocity. The gate spacing is 250 m for velocity and 1 km for reflectivity. A 4-bin average velocity is computed, each having one corresponding reflectivity value. If that average velocity is less than 2.5 m s-1, then the reflectivity is set to missing.

Either one or both modes can be selected.

Functional description

The algorithm, CONVECTIVE/STRATIFORM PRECIPITATION TYPING and rain rate computation, examines a volume of reflectivity to determine what precipitation type and precipitation rates are occurring at each range bin.

2.1

Convective vs. stratiform

First, convection is checked for at all elevation angles. The check for convection is based on reflectivity thresholds and bright band vertical depth dimensions. If the reflectivity in any bin for a vertical column of data bins is greater than 50 dBZ then precipitation is tagged as "convective," unless that bin is within the bright band layer.

If the upper reflectivity value is not exceeded, then the highest bin in the vertical column with reflectivity above 30 dBZ is found. If the second threshold value is found at or above the -10 deg C height, then the precipitation is tagged "convective."

2.2

Checking for other precipitation types

If the initial identification is that stratiform is present, i.e., not convective, then further typing is done. If the hybrid tilt reflectivity is located below the bright band layer, then the precipitation is tagged "good rain." If the hybrid tilt reflectivity is located within 1200 m distance above the bright band, then the precipitation is tagged as "good snow." (Click here for an example how good rain and snow are determined.)

Reflectivity measured within the bright band and at heights greater than 1200 m above the bright band top are deemed "bad rain" and "bad snow" respectively.

2.3

Rain rates

Functional Description

Section 2 above uses output from the bright band detection algorithm, which identifies a horizontally-uniform layer of contaminated reflectivity due to melting hydrometeors (see linked document).

3.1

Initialization

Check to see if there are previous BB detections.

3.2

Processing a volume scan

Only data within the range interval 10-30 km (for VCP 11) or 20-30 km (for VCP 21) are examined to determine the existence of a bright band. For each grid point, columns of reflectivity at each azimuth and range location are checked to see if there is sufficient data to determine the vertical extent of a potential bright band. If there is terrain blockage and the hybrid scan tilt angle is greater than or equal to 2.4 deg, then data in the vertical column are not used. The number of vertical columns used to locate the bright band is tracked by the variable "good_points." If there is an insufficient number of "good_points," there is no test for a bright band.

The column is then searched for the maximum reflectivity. If the maximum is greater than 30 dBZ, the data value and location are stored. The column reflectivity values are normalized by dividing each reflectivity by its column maximum. The heights at which normalized reflectivity decrease by 20%, above and below the maximum value, are then the bright band top and bottom, respectively. The tops and bottoms for each column are averaged and the standard deviation of the tops is calculated.

3.3

Checking for a bright band

The existence of a bright band is determined using two criteria. The first is the ratio of points with reflectivity greater than the threshold to the value of "good_points." The second is the standard deviation of the bright band tops found above. The existence of a bright band is declared if the ratio is greater than 10% and the standard deviation is less than 500 m.

If a bright band is detected then the top and bottom heights are averaged with the heights from the previous 30 minutes. These data are then used in the PRECIP TYPING algorithm.

Overview

(Note: Several of these routines are also used in the multi-sensor algorithm.)

Now that the precip rates and types have been determined in polar coordinates, they are now combined on a common grid to produce accumulations. Gridding also allows rates to be compared with satellite data to verify that there is really precip occurring and to be compared to lightning data to determine if convection is occurring. Development of additional tools is underway.

Functional Description

This routine determines which radar has the lowest hybrid scan bin (HSB) height for each grid point. A look up table contains the names of radars associated with each grid point. Some grid points may have 2 or more associated radars. Each radar is checked for availability. The name of the available radar having the lowest (HSB) height is saved for that grid point. If the lowest HSB is greater than 5000 m AGL, there will be no radar coverage for that grid point. The radar coverage map shows each grid point color coded according to which radar has the lowest HSB height.

Functional Description

Polar data for each radar with the lowest hybrid scan bin (HSB) height are remapped to a Cartesian grid. The remapped data are hybrid scan reflectivity, rain rate, and precipitation type. The data nearest to a grid point is assigned to that grid point.

Functional Description

Remapped Cartesian data for each radar are combined for the entire QPESUMS domain.

Functional Description

Model surface and satellite cloud top temperatures are used to eliminate non-precipitation echoes. The temperature difference is calculated. If the difference is less than 8 C (adaptable), then no clouds are believed to be present and the rate is set to zero.

Functional Description

If lightning occurs at a grid cell within a 5 minute period, the precip type at that grid cell is assigned as convective and the rate is recalculated. Lightning data for the QPESUMS grid are computed as a flash density (e.g., strikes in 5 minutes at each grid point). Lightning data for the previous 15 minutes are searched. A 100 mm hr^-1 cap is placed on convective rain rates.

Functional Description

There are various accumulation periods. The frequency of updating the accumulations depends on the period. One hour accumulations are updated every 5 minutes. Three, six, and 24 hour accumulations are updated once an hour. The 72 hour accumulation is updated every 24 hours.

Functional Description

This algorithm is designed to provide rain and snowfall (SWE) based on satellite IR data alone. Precipitation rates are derived from the latest satellite IR data using a climatological regression curve that relates IR temperatures and precipitation rates. This algorithm is used primarily for development purposes.

10.1

Reading in data

Satellite data are read in and scaled by multiplying each value by 10. A QC check is done whereby values not between 163K and 330K are set to missing.

10.2

Re-mapping

The satellite grid resolution is coarser than the QPESUMS grid. The satellite cloud-top temps are sampled on the common QPESUMS grid using a nearest neighbor approach.

10.3

Re-projection

Radar data are re-projected in a series of calculations that determine new spacing intervals and dimensions for the QPESUMS grid. The re-projection is used in the output of the NIDS formatted files which use different dimensions than the QPESUMS grid. If new satellite data are found within a 15 minute time window, NIDS and binary files of precip accumulations are output as specified in a configuration file.

10.4

Precipitation derived from climatological regression

Rain rates are calculated from the IR temperatures. The exponential curve or equation used was derived from > 160 hours of data consisting of radar-derived rain rates and cloud-top temperatures. Using precipitation type flags, a rain regression coefficient is used for rain (good, bad, or convective) and convection and a snow coefficient is used in snow (good or bad).

10.5

Accumulation

Precipitation accumulations are produced for the different time periods (see Section 9).

Functional Description

This algorithm prepares data points for use in a real-time regression calculation between satellite cloud top temperatures and good rain/snow. The user may choose if regressions will be built for individual radars or for the entire domain.

Separate equations are derived for rain and snow. Various quality control checks are done on the data range and correlation coefficients to ensure that the regression equation fits the data appropriately. (See Section 11.2.)

11.1

Generate real-time regression data

First, since there are more grid points in the QPESUMS grid than there are in the satellite grid, QPESUMS precipitation rates are averaged to correspond to the satellite grid scale. Averaging precipitation types is more complicated and relies on a priority system in lieu of averaging. If a convective type is found to be associated with a satellite grid point, then the satellite grid obtains convective type. If a satellite grid point contains both good rain and good snow, the satellite grid point assumes the majority type. If the grid point contains both bad rain and bad snow, the majority of bad snow or bad rain is assigned to the satellite grid point.

A satellite-radar pair is accepted if there is a) no missing satellite temperature, or b) either good rain or good snow associated with the pair. Any grid cells that are deemed convective are not used in the regression. This technique applies to stratiform rain and snow only. Valid pairs are used in the regression technique if they are less than one hour old. Data older than one hour are purged and no longer used.

11.2

Update regression function

Several quality control criteria must be met when deriving real-time satellite-radar regressions. When regressing two variables, stable relationships may be found when there is a) a sufficient number of satellite-radar pairs, b) variety in the satellite-radar pairs. The technique described herein assesses these criteria by initially placing each satellite-radar pair in 1 K temperature "bins". A temperature bin is said to have enough data if there are five or more pairs that fall into the bin. The radar rain (or snow) rates are then averaged for each temperature bin having five or more pairs. The second criterion is evaluated by calculating the number of temperature bins that have five or more pairs. If ten or more bins satisfy this criterion, then there is sufficient variety of satellite-radar pairs and a regression may be derived.

A regression is fit to the data using a decaying exponential equation. This equation was found to fit the data most accurately. Two parameters that describe the steepness of the curve and the y-intercept are optimized using a least-squares fit. Finally, a correlation coefficient describing the "goodness of fit" to the data is calculated. If this value is 0.5 or greater, then the real-time regression is used. Otherwise, the technique reverts to a climatological regression for both rain and snow.

11.3

Correlation coefficients for cloud top temperature/precipitation rate regressions

The algorithm 'Multisensor QPE' derives a regression between satellite cloud top temperature (K) and radar derived precipitation rates (mm/hr) using data from areas identified as "good" rain or "good" snow.

If there are not enough temperature bins in the regression (a valid temperature bin contain > 5 precipitation rates) then a check is done to see if the there is a 'flat' distribution of data points. If the maximum 'mean' precipitation rate for bins is less than the 1mm/hr threshold, the distribution is considered flat and the correlation coefficient is set to -4 meaning that the climatological regression is used. If the maximum 'mean' precipitation rate associated with any cloud top temperature bin is greater than 1mm/hr then the correlation coefficient is set to -3 and the most recent good regression (".latest") is used.

If there are enough data points (i.e. sufficient number of cloud top temperature bins and associated precipitation rates) then the following regression relationships are calculated to determine whether the regression is good enough to be used.

The equation below describes the relationship of the regression curves:

R = alpha * exp ( beta * CTT )

Where R is the precipitation rate (mm/hr), CTT is cloud top temperature (K) and alpha and beta are regression function coefficients.

Regression function coefficients (alpha and beta) and the correlation coefficient (CCOEFF) are calculated for each regression every 5 minutes.   In order to derive the regression, mean precipitation rates for each CTT bin must be calculated. If the maximum mean precipitation rate corresponding to a CTT bin is less than 1mm/hr then the CCOEF is instantly set to '-2' and the regression is not used and a climatological regression is used instead.

If the alpha and beta coefficients produce a precipitation rate less than 1mm/hr from the following equation:

Rate = alpha * exp ( beta * 200K )

the regression is given a correlation coefficient of -2 and again a climotological regression is applied for that time.

If the alpha and beta calculated for a regression are not found to be in the following range then the CCOEF is set to zero and the most recent good regression is applied instead

0 < alpha < 1 * 10³⁰

-1 * 10³⁰ < beta < 0

If the regression passes all the checks listed above then the CCOEFF is calculated for the regression. If the correlation coefficient is greater than 0.5 the regression is used. If the correlation coefficient is below 0.5 then the regression is not used and the most recent good regression is applied instead.

See figures below for examples of regression with each of the correlation coefficients discussed above (0, 2, > 0.5 and < 0.5). The red line is the mean precipitation rate and the blue line is the regression fit to the data.

Figure 1. CCOEFF=0; alpha and/or beta are out of range

Figure 2. CCOEFF=-2. Alpha and beta produce rates < 1mm/hr at 200K or maximum mean precipitation rate < 1mm/hr

Figure 3. CCOEF = -3. There are not enough temperature bins with greater than 5 precipitation rates to fit a regression with and the maximum "mean" precipitation rate > 1mm/hr so '.latest' good regression is used.

Figure 4. CCOEF = -4 . There are not enough temperature bins to fit a regression and the maximum "mean" precipitation rate for temperature bins is less than 1mm/hr so a climatological regression is used. (In this case there are not enough rates i.e. >5 in ANY bin so "mean" rates cannot be calculated)

Figure 5. CCOEFF=0.77. Good regression, used in real time

Figure 6. CCOEFF=0.32. Bad regression, not used

11.4

Regression application

The parameters derived in the regression process are used to supply rain and snow rates to the cloud-top temperature map only for those grid cells that have bad rain or bad snow precipitation types. The calibrated satellite precip rates are thus applied to grid cells where the quality of radar data is in question due to overshooting and/or bright band contamination.

For the entire grid, hourly RUC data are used to determine a rain/snow line. At a bad snow grid point, the snow regression is used, while a rain regresison is used at bad rain grid points. The user can specify in a configuration file whether the regression is derived from satellite and individual radars or with satellite and all radars combined.

Functional Description

The gauge-only analysis incorporates all available rain gauge observations in a domain. Initially, a simple quality control (QC) step is applied to the data. The QC routine is capable of checking the magnitude, temporal consistency, and spatial consistency of each rain gauge report. Currently, only the magnitude check is applied to the data such that gauge accumulations are ignored if the accumulations exceed 8"/hour (203 mm/hr). If the reporting interval of the gauges is more frequent than one hour, then the accumulations are aggregated to produce hourly accumulations. The point estimates are analyzed on the 1x1 km QPESUMS common grid using a Barnes objective analysis scheme. The two parameters optimized in the gridding process are the weighting function and the radius of influence. Initial values for these parameters are chosen according to the distribution of the gauge spacing for a given gauge network. Further refinement of the parameters is accomplished by examining both widespread and scattered rainfall events. Expert analysis is used to determine the quality of the gridded gauge product and thus the final values for the Barnes objective analysis parameters. The hourly accumulations are aggregated to produce long-term accumulation products.

Functional Description

The multisensor and radar-only products are both adjusted using a mean field bias correction based on the comparison of the nearest grid point values to the gauge locations. This adjustment technique is intended to remove domain-wide biases on an hourly basis that may be due to improper Z-R equations, overestimation from hail contamination, or underestimation from virga. The mean field bias is computed as follows:

where is the bias factor, R_i is the hourly multisensor or radar-only accumulation, G_i is the hourly gauge accumulation and N is the number of gauges reporting for that hour. Some checks are placed on the bias value to ensure that no unreasonable biases are computed. If the mean of the gauge accumulations (G_i) over the entire domain is zero and the mean of the estimator (R_i) at the gauge locations is zero, then the bias factor is set to a value of one. If the mean of the gauge accumulations (G_i) is zero and the mean of the estimator (R_i) at the gauge locations is nonzero, then the bias factor is set to a value of zero. Otherwise, the entire grid of radar-only and multisensor hourly rainfall is divided by the bias factor to produce the radar-mean field bias adjusted and multisensor-mean field bias adjusted products. Like all other QPE products, these are aggregated to produce long-term accumulations.

Functional Description

The multisensor and radar-only products are adjusted on an hourly basis using a spatially nonuniform bias adjustment technique. This adjustment is intended to remove nonuniform biases that may be due to improper Z-R relationships, range-dependency in QPEs from reflectivity profiles that decrease with height, and contamination from hail, birds, ground clutter, chaff, and other echoes from nonweather targets. First, the difference between the gauges and the estimators is computed at each gauge location (e.g., G-R). A local bias field is then computed for the 1x1 km QPESUMS common grid using the point (G-R) values. The parameters used in the Barnes objective analysis scheme discussed in section 12.0 are used here to analyze the point (G-R) values at the gauge locations to the entire grid. The weighting parameter and radius of influence are dependent on the gauge spacing with a given gauge network. Finally, the local bias field is added to the multisensor and radar-only hourly products to yield the radar-local bias adjusted and multisensor-local bias adjusted QPE products.

return to top