THESIS ABSTRACT
Water management in the West is becoming increasingly important as heavily populated cities demand more water. Today, companies involved in dam management, such as the Salt River Project (SRP), rely on limited rain gauge measurements to aid in the prediction of streamflow from a watershed. Since contemporary and future hydrologic and atmospheric models will utilize WSR88D precipitation estimates, monitoring capabilities of both snow and rainfall in the West must improve.
Mixed-phase precipitation (rain at low elevations, snow at high elevations) associated with cool season, extratropical cyclones produces the greatest amount of seasonal runoff in central Arizona. Radar estimates of precipitation suffer the most during winter, especially over complex terrain. Thus, estimates of precipitation from the Phoenix (KIWA) radar are critically examined for three cases of widespread, stratiform precipitation over the Salt and Verde watersheds. Storm total accumulations as estimated by radar are shown to be extremely sensitive to brightband signatures in vertical profiles of reflectivity.
An automated method is developed to improve estimates of precipitation for these cases by blending data from two WSR-88Ds and one GOES satellite. The "multi-sensor" approach provides more accurate estimates of rainfall across the lower elevations and snowfall (liquid equivalent) over higher terrain. Rain gauges are used as an independent, ground-based source to assess the magnitude of improvements made. Improved estimates of precipitation will have a positive impact on capabilities of real-time, operational rainfall monitoring as well as on streamflow prediction.
CHAPTER VII
SUMMARY AND CONCLUSIONS
Over mountainous regions in the West, seasonal runoff results from a combination of rainfall at low elevations and snowmelt at high elevations. Efforts are underway to model streamflow given initial precipitation conditions, which result from extratropical cyclones. Hydrologic models of the current era and those of the near future will utilize PPS estimates from radar to initialize spatially distributed precipitation fields. Moreover, precipitation estimates from radar are the primary tool for operational, real-time rainfall monitoring in NWS offices. One goal of this study was to assess and understand the characteristics of operational PPS estimates generated by KIWA during the cool season.
Not only did KIWA systematically underestimate mean gauge accumulations, but it produced unrealistic spatial patterns of precipitation. It was shown how PPS estimates from the WSR-88D at KIWA are very sensitive to the radar's sampling of complex vertical profiles of reflectivity. Elevated reflectivity directly below the melting layer resulted in large, unreasonable precipitation estimates. The presence of a melting-layer brightband resulted in concentric rings of local maxima in the estimated accumulations around the radar. Furthermore, the magnitude of PPS estimates were found to decrease with range as beam heights increased above the surface. This range-dependency was a result of the radar beam sampling progressively less and less precipitation structure with height. Estimates from radar alone were found to be highly sensitive to the vertical structures of precipitation and largely unrepresentative of precipitation occurring at the surface. Thus, KIWA precipitation estimates have proven to be unreliable for accurate, precipitation monitoring during the cool season.
Storm total products from KFSX were shown, but not evaluated in detail. The ground elevation of the radar was always above the freezing level elevation during the cool season cases considered. Analysis of PPS accumulations revealed that, in all cases, KFSX severely underestimated the accumulations of snow water equivalent. This study addressed radar sampling problems and did not attempt to optimize many of the adaptable parameters used in the PPS. It is believed, however, that significant improvements can be made by changing coefficients and exponents in the default Z-R equation so that it applies to the liquid equivalency of snow (i.e., a Z-S equation).
Careful examination of the second and third events showed that precipitation estimates at several grid points were artificially inflated by anomalous beam propagation. Evidently, terrain features in the area are steep enough to block at least a portion of the lowest elevation angle from KIWA. The concomitant ground returns are not filtered in the clutter suppression component of the WSR-88D system. It is suggested that the NWS adjust their clutter filter on the KIWA radar to eliminate high intensity, non-meteorological returns.
The following conclusions were determined by examining radar PPS products during cool season precipitation events:
1) KIWA overestimates precipitation at ranges where the brightband is sampled.
2) Beyond ranges where the brightband is intercepted, KIWA underestimates precipitation totals.
3) The KIWA ground clutter filter does not properly suppress all high intensity returns that result from beams hitting the surface.
4) The Z-R equation used at KFSX must be optimized to better estimate the liquid equivalency of snow.
5) Neither radar adequately captures spatial gradients in the actual accumulations of precipitation. The use of these precipitation estimates in hydrologic models or as a precipitation-monitoring tool during the cool season is not recommended.
For comparison purposes, estimates were generated by a rainfall estimation technique using IR satellite data alone. Currently, automated IR-based techniques do not exist that are suitable for estimating mixed-phase, stratiform precipitation during the cool season. The GOES Precipitation Index was thus modified to estimate precipitation at spatial and temporal resolutions comparable to radar. The thresholding technique of cloud top temperatures apparently produced better distributions of rainfall totals than radar-only estimates, but performed poorly in the snow regions. These satellite estimates were believed to be contaminated by high, cold clouds, which yielded little precipitation. All case studies consistently revealed an overestimation of precipitation by the satellite only technique. Thus, estimates from satellite alone are inaccurate.
This study developed a new precipitation estimation scheme, which incorporates data from multiple remote sensors. PPS estimates from KIWA and KFSX are blended together with IR satellite imagery to estimate both rain and snow at high spatial and temporal resolutions. All estimates are made completely independent of gauge reports.
The multi-sensor technique made significant improvements over estimates made from individual remote sensors (radar and satellite). The spatial distribution of rainfall amounts are better represented by blending data. The blending technique, however, did not perform as well in the snow regions. Many of the spatial snow gradients were not captured. The multi-sensor technique assumes that KFSX reasonably estimates snow water equivalent at low levels near the radar. Inaccurate radar estimates over these small regions are reflected in the multi-sensor estimates. Future work will entail testing several different Z-S equations for optimal radar PPS performance in snow.
By utilizing all information, such as the height of the brightband, precipitation types (snow versus rain) can be segregated in a physically meaningful way. The multi-sensor scheme improves precipitation-monitoring capabilities during cool season, extratropical cyclones. Basin-wide mean precipitation may now be estimated well enough for inclusion in streamflow prediction and atmospheric numerical models. Moreover, the multi-sensor estimates may be useful for real-time rain and snowfall monitoring in NWS offices.
The following conclusions were determined by examining multi-sensor estimates during cool season precipitation events:
1) Mean domain-wide estimates from blended radar and satellite data produced the best overall agreement with co-located gauges.
2) The spatial patterns of rainfall were represented best by the multisensor scheme.
3) Multi-sensor estimates are not contaminated by features common in radar reflectivity fields like the brightband or heavy accumulations from anomalous beam propagation.
4) The adaptive characteristics of the technique screen out contamination by cirrus clouds. These rainfall-deficient clouds often result in an overestimation by simple, IR-only techniques.
6) The multi-sensor approach separates raining from snowing regions in a physically meaningful way by identifying and utilizing the height of the brightband.
7) The developed technique will continue to be tested so that it may be applicable in real-time rainfall and snowfall monitoring operations.
Tests will be conducted in the future to develop the multi-sensor scheme as a technique employed in operations. These proposed algorithm enhancements are summarized as follows:
1) Employ a Z-S relationship for KFSX PPS estimates. Better snow water equivalent estimates near the radar will yield better multisensor estimates in the entire snowing region.
2) Account for spatial variations in freezing level and radar brightband heights to more accurately separate snow from rain. Field studies are being planned to assess the behavior of these heights and how they relate to snow levels in complex terrain.
3) Instead of using a mean brightband height, account for the fact that freezing level heights, brightband heights, and snow level altitudes fall with time during cool season precipitation events. Observations from the proposed field programs and numerical model output will enhance the time and spatial resolution of the changing freezing level heights.
4) When correlating hourly KBVA PPS estimates with mean hourly CTTS, incorporate all reflectivity data at the lowest level measured below the brightband (i.e., use data from the teffain-based hybrid scan). This will allow many more grid values to be used in the regressions. This will yield more stable correlation equations, which should be more representative of the relationship between the two variables.
5) The current methodology correlates mean CTTs and radar PPS estimates each hour. Each cumulative regression includes all data from previous hours in the event. Perhaps the adaptive characteristics of the technique would be refined by including data only from the previous 3 hours. The regressions would thus be derived from a moving 3-hour window of mean CTTs and radar PPS estimates.
Future research will also determine if this methodology has applications in
other mountainous geographic regions. Radar sitings exist in complex terrain
in many locations in the IW. This work may have implications for improving precipitation
estimation at several WSR-88D radar sites. Furthermore, brightband contamination
may also be a serious problem over flat terrain during the cool season. Tests
will be conducted to see if the developed scheme can improve precipitation monitoring
over these flatlands as well.