comparison between radar estimated and rain gauge ...

Environmental Engineering and Management Journal

April 2012, Vol.11, No. 4, 723-731

http://omicron.ch.tuiasi.ro/EEMJ/

“Gheorghe Asachi” Technical University of Iasi, Romania

COMPARISON BETWEEN RADAR ESTIMATED AND RAIN GAUGE MEASURED PRECIPITATION IN THE MOLDAVIAN PLATEAU Sorin Burcea1,2 , Sorin Cheval1,3, Alexandru Dumitrescu1,4, Bogdan Antonescu1,5, Aurora Bell1, Traian Breza1 1

National Meteorological Administration, 97 Bucuresti-Ploiesti, 013686 Bucharest, Romania University of Bucharest, Faculty of Physics, 405 Atomistilor Str., Magurele, Ilfov, Romania 3 National R&D Institute for Environmental Protection, 294 Splaiul Independentei, 060031 Bucharest, Romania 4 Faculty of Geography,University of Bucharest, 1Nicolae Balcescu Blvd., 010041 Bucharest, Romania 5 University of Manchester, Centre for Atmospheric Science, School of Earth, Atmospheric and Environmental Sciences, Oxford Road, M13 9PL, Manchester, United Kingdom 2

Abstract Heavy rainfall events have produced significant damages and casualties in the Moldavian Plateau (Romania) in the last decades. Such phenomena are characterized by large spatial and temporal variations, and the forecast of their occurrence is thus very challenging. This study aims to compare the radar estimations and the rain gauge measurements, in order to improve the quantitative precipitation estimation (QPE) in the area of interest. The research uses data from the WSR-98D (Weather Surveillance Radar – 98 Doppler) S-band Doppler radar located at Bârnova, and from rain gauges within weather stations run by Meteo Romania (Romanian National Meteorological Administration). Considerable spatial distinctions and areas with good radar accuracy for QPE have been emphasized during the investigations. The output validation aimed to predict the rain gauge amounts using the radar information and the resulted adjustment parameters. The validation demonstrates that the Bârnova radar data are reliable within approx. 150 km radius, and the comparison with rain gauge measurements can foster consistently the QPE accuracy. Key words: comparison, Doppler radar, rain gauge, rainfall Received: December 2010; Revised final: July, 2011; Accepted: August, 2011

1. Introduction Atmospheric precipitation shows a large variation both in space and time, while numerous applications, such as flood forecast or agriculture, require robust information, capable to describe accurately the precipitation conditions over an area. Substantial efforts have been dedicated to improve the accuracy of estimations and measurements through diverse techniques and methodologies (Krajewski et al., 2010). The international scientific production on merging the rain gauge measurements and radar estimations is already vast and a few

revisions have been published recently (Gjertsen et al., 2003; Goudenhoofdt and Delobbe, 2009). Within the Romanian National Meteorological Administration (Meteo Romania), rain gauges and radar data are extensively used for quantitative precipitation estimations (QPE). Rain gauges measure the rainfall quantities close to the Earth’s surface, but their spatial representativeness can be very limited, mainly in the case of convective situations or complex topography. The measured amounts are influenced by systematic (i.e. wind, snowfalls) or unsystematic factors (i.e. station relocation, change of sensors), but they are generally considered satisfactory and used consequently.

Author to whom all correspondence should be addressed: e-mail: [email protected]; Phone: +4021 3183240 /int. 120

Burcea et al./Environmental Engineering and Management Journal 11 (2012), 4, 723-731

Nevertheless, one should take into account that the gauge-based precipitation amounts almost always underestimate with a magnitude depending on the strength of the influencing factors (Legates, 1987; Adam and Lettenmaier, 2003; Ciach, 2003; Constantinescu et al., 2007). Cheval et al. (2011) report that, in June, the gauge based precipitation measurements can be biased by factors like wind and evaporation with 10-20%. Such values could explain the underestimation of the gauge-based precipitation compared to radar values. In their turn, the radar estimations have good spatial and temporal resolutions, although the lack of quantitative precision still characterizes the output, due to various sources of errors (radar miscalibration, attenuation; see Villarini and Krajewski (2010a) for a recent review). The qualitative output is yet valuable. Combining the rainfall input measured by ground sensors and radar estimations one can expect more reliable precipitation information. Previous studies have approached this topic (Huff, 1967; Wescott et al., 2008; Trapero et al., 2009) exploring different methods for this type of analysis, i.e. rain gauge amounts refer to point accumulations while radar-derived values correspond to a volumeaveraged rainfall rate (Zawadski, 1975; Villarini and Krajewski, 2008; Villarini et al., 2008). Ciach and Krajewski (1999) admit that many factors influence the relationships between radar-measured reflectivity, radar-predicted rainfall amounts, and rain gauge measurements, and they suggest the utility of studies on the given topic. The results may improve fundamentally the quality of the radar input (Trapero et al., 2009; Ciach et al., 2007) by increasing the QPE accuracy and facilitate the radar output postprocessing. At the same time, the radar-based rainfall climatology can add valuable information for precipitation mapping (DeGaetano and Wilks, 2009). Meteo Romania operates a weather radar network composed by 8 Doppler radar systems: five S-band WSR-98D units, two C-band EEC-2500C and one Gematronik METEOR 500C. Based on reflectivity, radial velocity and spectrum width a series of derived products are currently generated in order to support the nowcasting and warning activities. Sampling one convective season, Georgakakos and Spenser (2009) have recently investigated the relationships between radar data and real time rainfall over Romania and emphasized the benefits and limitations of such an approach. The EU FP6 Project HYDRATE (Hydrometeorological data resources and technologies for effective flash flood forecasting) developed through 2006-2009 has urged the integrated analysis of radar and rain gauges information, and this paper presents part of the results. The main objective of this paper is to evaluate the differences between radar-derived (RD) and gauge-measured precipitation amounts (G) in the Moldavian Plateau (Romania). The relationships between RD and G vary with a large number of factors (i.e. climate, radar characteristics), and the

724

quantitative associations may have very local relevance. Borga et al. (2002) point out that the gauge-based radar adjustment depends on the homogeneity of the radar estimations accuracy, so that one can assume that the local applicability prevails. Ulbrich and Lee (1999) find that the NEXRAD Z-R relation underestimates the stratiform rain up to 25% and overestimates the convective events up to 35% in the sub-tropical climate of South Carolina. By comparing rainfall events over five-year period in Texas, Jayakrishnan et al. (2004) document steady underestimation of the WSR-88D radars. Xie et al. (2005) report a substantial difference between the average precipitation (265.5 mm) from seven NEXRAD cells and the collocated gauge measurements (136.2 mm) in New Mexico. Applying a product-error-driven approach to obtain a realistic uncertainty model of the radar precipitation estimates, using data provided by the Oklahoma City radar, Ciach et al. (2007) found that radar tend to overestimate the total rainfall in the warm and hot seasons and to underestimate it at distances above 100 km from the radar in the cold season. Also, they show that the components of the proposed functional-statistical error representation can be approximated using simple analytical models. Applying an empirically based model in which the relation between true rainfall and radar rainfall could be described as the product of a systematic distortion function and a random component, Villarini and Krajewski (2010b) reached to results that are in agreement with the ones in Ciach et al. (2007). Moreover, Legates (2000) argued that the standard Z–R relationship tends to overestimate light precipitation and underestimate heavy precipitation. In our turn, we aim (1) to find quantitative expressions for the differences between RD and G in the Moldavian Plateau, (2) to assess the quality of radar measurements in relation to the features of the beam propagation and terrain, (3) and to evaluate the correlations between the two variables. The following section presents the dataset used and methodology applied for processing the data. Section 3 contains the results of the study and discussions regarding the results, while in the last section conclusions are presented. 2. Dataset and methodology The research focuses on daily (24 h) accumulations registered at weather stations during the 2003–2008 convective seasons (May–September) period. The output sustains the radar estimations, fostering the hydrological modeling, including flash flood forecast. Flash flood forecast systems are useful in assisting forecasters to issue flash flood warnings and watches over a certain hydrological area. These models require independent calibration for each region and it accounts for range effects, synoptic conditions, space–time resolutions, and the spatial and temporal dependences of the errors

Comparison between radar estimated and rain gauge measured precipitation in the Moldavian Plateau

(Villarini et al., 2010). Over the Romanian territory, the rainfall intensity is generally decreasing from East to West (Breza, 2008). The present investigation refers to radar data provided by the Bârnova Doppler weather radar, covering a large territory in the Eastern part of Romania, mainly the Moldavian Plateau (Fig. 1). The study area concentrates significant human assets and various natural environments, so that the impact of heavy rainfalls has always challenged the meteorological interest. The WSR-98D radars from the Romanian National Radar Network currently estimate the rainfall rate using the NEXRAD Z–R (reflectivityrainfall rate) relation: Z=300R1.4 (Fulton et al., 1998). Among the extensive collection of Z–R relations that exist in the literature for various types of rainfall (Ulbrich and Lee, 1999), this equation represents a compromise between the relations for stratiform and thunderstorm rain (Hunter, 1996; Villarini and Krajewski, 2010b). Meteo Romania collects rain gauge precipitation data from two sources (Fig. 1): one network is placed following WMO requirements, within climatic weather stations (Net1), and the other one consists of rain gauges operational inside private perimeters, reflecting fairly local conditions (Net2). A total number of 106 rain gauges are operational within radar coverage area.

northern and eastern borders. The climate is temperate, with definite continental influences, noticeable in the temperature and precipitation as well. Fig. 2 illustrates the spatial distribution of the rainfall intensities, based on the peak-over-threshold concept, which implies the analysis of all precipitation amounts above certain thresholds selected for different durations (Borga et al., 2005). The highest intensities occur in the Eastern and Southern regions, associated with the most intensive and frequent summer convection phenomena. The area of interest is largely characterized by high intensities for 24 h rainfalls, and a similar pattern preserves for lower durations (Fig. 3).

Fig. 2. Rainfall intensities for 24h duration and 100 years return period

Fig. 1. Spatial distribution of the rain gauges and Barnova radar coverage area (230 km). Range rings are 50 km apart

The Moldavian Plateau is situated in the eastern part of Romania, and it lies between 45.3° and 48.1°N and between 26.5° and 28.8°E. Its relief is quite smooth, tabular, with altitudes decreasing from north (700 m) to south (200 m). To the west, the Moldavian Plateau meets the Subcarpathians Hills and the Eastern Carpathians, with higher elevations (500-2100 m) and significant terrain fragmentation. The Romanian Plain neighbors the area of interest to the south, while the study does not pass the national

Fig. 3. Rain intensity for different durations and return periods in the Moldavian Plateau (Weather Station Iaúi)

The radar system used in the present study is located at Bârnova, in the Eastern part of Romania. The topography of the area was briefly described

725


above and one can further mention that in the scanning area of the radar mountains are present as well. Although the minimum distance between radar site and mountains is about 115 km, the data quality can be affected by the beam-mountain interaction (Kucera et al., 2004; Burcea et al., 2010). Moreover, data quality is affected at long ranges by beam broadening and beam filling errors which can significantly underestimate de atmospheric precipitation (Villarini and Krajewski, 2010a). To adequately sample the atmosphere, the WSR-98D employs scanning strategies or volume coverage patterns (VCPs). A VCP is a series of 360 degree sweeps of the antenna at pre-determined elevation angles completed in a specified period of time. The WSR-98D radar system from Bârnova performs a volum scan using 9 elevation angles and completes 11 azimuthal scans in 6 minutes. After sampling the atmospheric volume over the coverage area and processing the received signal, all raw base data are sent to radar product generator (RPG) where, by applying radar algorithms, the final radar products can be disseminated and displayed by various users. In the present study, data used to generate the rainfall accumulations is the base reflectivity, measured along the convective seasons (MaySeptember) through 2003-2008. The radar reflectivity is sampled by volume scans over a polar grid with 1 km spatial resolution, 1 degree in azimuth and 6 minutes temporal resolution of data. Small scale (local) inhomogeneity contributes to the radar measurements and introduces bias to measurements inside the sample volume (Zhang et al., 2002). In the case of the inhomogeneity of random media, radar can only measure the spatial variation with a scale larger than the size of a sample volume (1km by 1°). Default radar algorithms process the data in order to mitigate the ground clutter. The values of reflectivity factor (Z, expressed in units of decibelZ) greater than 53 dBZ were capped to this value, in order to eliminate the hail contamination (Fulton et al., 1998), and values smaller than 5 dBZ were not included in order to eliminate noise. The radar dataset was converted from the native format into binary grid files which is then used to obtain the derived rainfall rate by using the default Z-R relation (Z=300R1.4) of the Bârnova WSR-98D. Further, the 6 minutes radar rainfall rates were integrated into 24 hours estimated radar precipitation accumulations and compared with the corresponding amounts from 42 Net1 rain gauges in the coincident pixels, at 1 km resolution. Both RD and G sum up information between 0530 and 0530 UTC, and only weather stations with G 1 mm were considered. Besides, only events when at least one station in the area of interest registered more than 20 mm were selected. Finally, a number of 267 rainfall events fulfilled the radar and rain gauge criteria and were selected for this analysis. Their regional averages were generally under 10 mm precipitation (70.6%

726

from cases), and in 38.6% of the cases at least one station registered 50 mm or more. The differences between RD and G were investigated based on two objectives functions: the ratio RD/G (BIAS) and the Root Mean Square Factor (RMSf) (Jayakrishnan et al., 2004; Trapero et al., 2009; Goudenhoofdt and Delobbe, 2009): BIAS (dB) = 10log(RD/G)

RMSF

ª1 N ª §R D exp « ¦ «ln¨¨ i « N 1 1 «¬ © Gi ¬

(1)

·º ¸» ¸ ¹»¼

2

º » » ¼

12

(2)

The BIAS and RMSf were calculated for each rainfall event and station, and N is the number of events. Basically, BIAS values close to zero and low RMSf values indicate good RD-G fitting. The Pearson and Spearman coefficients were used to evaluate the existing correlation and association between RD and G. The four indices were averaged for each station and mapped over the area of interest using the Kriging interpolation method. The outputs were validated by calculating the differences between the precipitation predicted (Gp) based on the mean BIAS and RMSf and the precipitation amount measured (Gm) at rain gauges during each event, using the methodology described in section 3.5. For validation, 18 rain gauges from the Net1 and 38 rain gauges from Net2 were used, all selected from the regions with the best BIAS, RMSf, Pearson and Spearman scores. Enclosing inherent limitations, the method assumes that the relationships based on comparisons between gauges and radars are valid for other locations. 3. Results and discussion The outputs refer to detecting the areas where the values of BIAS, RMSf, Pearson and Spearman coefficients reveal a good rainfall estimate within Bârnova radar coverage area. Some considerations about the factors generating the spatial pattern are also included. 3.1. BIAS Negative values prevail over the studied area (Fig. 4). We speculate that this is due to the use of a Z-R relation that underestimates the convective rains (Legates, 2000) or due to the presence of errors from radar miscalibration. Also, contributions from noise or ground clutter may induce a bias in the Z values. Near the radar location and over the mountainous chain in the western and southwestern parts of the radar coverage, the BIAS is negatively large. Range dependent biases are present in NEXRAD precipitation estimates due to the effects of range on beam size (broadening) and height of the sampling volume (radar beam overshoots the weather system) (Smith et al., 1996).


The average BIAS decreases with the distance from the radar location, and the standard deviation increases more dramatically after 150-160 km, due to the presence of topography and range-dependency of the radar measurements (Fig. 5). A recent study by Krajewski et al. (2011) presents the development of a method that describes the range-dependent error in radar-rainfall estimates. The method is a simple parametric model that represents the vertical profile of reflectivity structure with respect to the radar beam height. Using four years of radar data, the range-dependent error model shows very good agreement with the observed differences between radar rainfall estimates and gauge measurements.

radar, while the relation RD/G apparently becomes less significant after 150-160 km (Figs. 6 and 7). The regression coefficient, r, between mean RMSf and distance from the radar location (r = 0.64) suggests possible causal links between the two variables.

Fig. 6. Spatial distribution of the mean RMSf

Fig. 4. Spatial distribution of the mean BIAS Fig. 7. Mean RMSf variation with the distance from the radar location

3.3. Pearson correlation coefficients Radar-derived (RD) and gauge-measured (G) amounts are relatively high correlated (r 0.7) within approx. 150 km radius around the Bârnova radar (Fig. 8). The Pearson correlation becomes less significant outside this buffer, while its values constantly diminish with the distance (Fig. 9). Fig. 5. Variation of the mean BIAS (ave) and standard deviation (stdev) with the distance from the radar location

3.2. RMSf The spatial pattern anticipated by the BIAS is also emphasized by the RMSf spatial distribution. The lowest values, indicative of good RD-G fitting, occur at a certain distance from the radar. Mean RMSf does not exceed 3 dB before 150 km from the

3.4. Spearman’s rank correlation coefficients They define the association between RD and G, and they were calculated and mapped in order to substantiate the correlation levels accredited by the Pearson correlation. Spatial distribution of the Spearman coefficients suggests a significant decrease at around 150 km from radar location, once the measurements are influenced by the beam rangedependency, and by the presence of Carpathian Mountains.

727


Fig. 8. Spatial distribution of the Pearson correlation coefficients

Fig. 10. Spatial distribution of the Spearman rank correlation coefficients

Fig. 9. Pearson correlation coefficients: RD/G ratio vs. distance from radar to raingauge

Fig. 11. Spearman’s rank correlation coefficients: R/G ratio vs. distance from radar to raingauge

3.5. Validation The results were further tested on a dataset covering the period May-September 2009, at rain gauges from the area with significant relationships between RD and G, in terms of BIAS, RMSf and correlation factors. Thus, for 18 Net1 and 38 Net2 rain gauges placed between 17 and 156 km from the Bârnova radar, we predicted the rain amount of each event (Gp), based on the Eqs. (3) and (4):

G p BIAS

G p RMSf

728

R 10

BIAS 10

R RMSf

(3)

(4)

BIAS and RMSf used in the above equations were expressed in dB units. For the network outside 156 km, the validation is superflous since the correlation RD-G is poor. Both the absolute (mm) and relative (%) average differences between Gp(BIAS) and the measured rain amounts (Gm) are quite low (Fig. 12), indicating the potential to use the average BIAS values for adjusting the radar QPE. The highest 5 values of the differences belong to Net2 rain gauges situated in depressions or shaded by hills. After eliminating these outliers, the average absolute difference Gp(BIAS) – Gm calculated for the whole dataset is 1.6 mm, and the relative difference is 13.7%. Generally, the predicted amounts underestimate the rain gauge precipitation, but overestimation can occur in certain spots and moments. The validation has also referred to the RMSf (Fig. 13). The magnitude of the average differences


between predicted and measured amounts has similar values with the BIAS, and the same outliers can be observed. By removing them, the average absolute difference Gp(RMSf) – Gm for the whole dataset becomes 0.2 mm, and the relative difference 1.3%. It is important to emphasize that the results refer to average values, and temporal variations of the estimation characteristics are possible.

precipitation measurements in the Moldavian Plateau (Romania), using data covering 6 convective seasons (2003-2008). It aims to provide a general overview of the benefits resulting from integrated analysis of radar and rain gauge information and to assess the relationships between the two types of data in the area of interest. The research highlights a few limitations of such an approach, and it can motivate more thorough studies in the Moldavian Plateau, a heavy precipitation and flash flood prone region. Table 1. Statistics of the rain gauge measured amounts, BIAS- and RMSf-derived precipitation, before and after eliminating the outliers

Average Maximum Minimum Standard deviation

Fig. 12. Average absolute and relative differences between BIAS-derived and measured rain amounts

Fig. 13. Average absolute and relative differences between RMSf-derived and measured rain amounts

The small differences between the predicted and measured amounts, as well as the overall statistic parameters of the rain gauge measurements, on one side, and BIAS and RMSf derived precipitation, on the other side, support the utility of the approach. The elimination of the outliers marked in Fig. 12 leads to better statistics for most of the considered parameters (Table 1). The significant changes of the maximum values and standard deviations of the derived precipitation in the context of quite stable statistics of the measured amounts before and after ouliers correction suggest radar limitations and advocate more detailed studies. 4. Conclusions This study is the first comparison between the Bârnova weather radar estimations and rain gauge

Measured Before After 11.9 11.8 17.1 17.1 8.2 8.2 1.7 1.6

BIAS-derived Before After 11.2 10.2 27.2 19.8 7.1 7.1 4.2 2.2

RMSf-derived Before After 12.8 12.0 25.8 19.1 7.4 7.4 3.9 2.6

The datasets refer mainly to convective events, and the initial data have been filtered to avoid strong artificial biases as much as possible. The convective events selection was based on quantitative criteria, such as amount threshold. Starting with this approach, it can be assumed that combining a clear distinction between convective and stratiform precipitations could supply interesting details. We investigated the differences between radar and rain gauge precipitation (a) and the statistical correlations between the two variables (b). The output was then validated and the results are useful for further research in the area. One could notice that the reliable radar QPE is limited to 150-160 km around the radar location, as demonstrated by the BIAS, the RMSf and the RD-G correlation coefficients as well. Generally, the Bârnova radar information likely underestimates the gauge precipitation. The underestimation increases steadily with the distance from radar because of the influence of land curvature on the radar beam (Austin, 1987; Villarini and Krajewski, 2010a). The decrease is more rapid to the west, once the topography becomes more complex. The BIAS and the RMSf values confirm that the radar beam-terrain interaction and range dependency are capable to cause severe underestimation and nondetection of precipitation in mountainous areas (Young et al., 1999; Kucera et al., 2004). Gjertsen et al. (2003) argue that the daily scale can be reasonably used to analyze the accuracy of the precipitation estimate algorithms. While the 24 h temporal resolution is suitable for illustrating the general pattern of the relationships between radar and rain gauge precipitation, finest scale approaches can reveal details to improve the available hydrological input. Currently, the hourly precipitation values are available at a limited number of stations in the Moldavian Plateau, so that may influence the initiation of regional scale studies.

729


Our findings are consistent with previous studies, emphasizing that both the differences and correlations between radar data and rain gauge amounts have rather local significance than general relevance over the studied area (Georgakakos and Spenser, 2009). For this reason, they should be utilized accordingly. We admit that the output could change quantitatively while longer datasets become available, but the qualitative pattern is most likely stable. At the same time, the results might be improved if similar analysis extends eastward, in the Republic of Moldova, covering completely the area where observations provided by Bârnova weather radar are available. Acknowledgements This research was completed in the framework of the EU FP6 Project HYDRATE (Hydrometeorological data resources and technologies for effective flash flood forecasting), Contract no: 037024, 2006-2009. This work was supported by project: POSDRU/88/1.5/S/61150 “Doctoral Studies in the field of life and earth sciences”, project co-financed through Sectorial Operational Program for the Development of Human Resources 2007-2013 from European Social Fund.

References Adam J.C., Lettenmaier D.P., (2003), Adjustment of global gridded precipitation for systematic bias, Journal of Geophysical Research, 108(D9) 4257, doi: 10.1029/ 2002JD002499. Austin P.M., (1987), Relation between Measured Radar Reflectivity and Surface Rainfall, Monthly Weather Review, 115, 1053-1070. Borga M., Vezzani C., Dalla Fontana G., (2005), Regional Rainfall Depth–Duration–Frequency Equations for an Alpine Region, Natural Hazards 36, 221–235. Borga M., Vezzani C., Dalla Fontana G., Tonelli F., Moore R.J., Andrieu H., (2002), Long-term assessment of bias adjustment in radar rainfall estimation, Water Resources Research, 38, doi:10.1029/2001WR000555. Breza T., (2008), Maximum intensities of rainfall, Romanian Climatologica Characterization, (in Romanian), Academiei Publishing House, Bucharest, Romania. Burcea S., Antonescu B., Bell A., (2010), Data quality of the Romanian WSR-98D weather radar systems, Proceedings, 6th European Conf. on Radar in Meteorology and Hydrology: Advances in Radar Technology, Sibiu, Romania, 322–325, On line at: http://www.erad2010.org/pdf/oral/tuesday/quality2/02 _ERAD2010_0369.pdf. Cheval S., Baciu M., Dumitrescu A., Breza T., Legates D.R, Chendes V., (2011), Climatologic adjustments to monthly precipitation in Romania, International Journal of Climatology, 31, 704-714. Ciach G.J., Krajewski W.F., Villarini G., (2007), Producterror-driven uncertainty model for probabilistic precipitation estimation with NEXRAD data, Journal of Hydrometeorology, 8, 1325-1347. Ciach G.J., Krajewski W.F., Villarini G., (2003), Local random errors in tipping-bucket rain gauge measurements, Journal of Atmospheric and Oceanic Technology, 20, 752-759. Ciach G.J., Krajewski W.F., Villarini G., Krajewski W.F., (1999), Radar–Rain Gauge Comparisons under

730

Observational Uncertainties, Journal of Applied Meteorology, 38, 1519-1525. Constantinescu G.S., Krajewski W.F., Ozdemir C.E., Tokyay T., (2007), Simulation of airflow around rain gauges: Comparison of LES with RANS, Advances in Water Resources, 30, 43-58. DeGaetano A.T., Wilks D.S., (2009), Radar-guided interpolation of climatological precipitation data. International Journal of Climatology, 29, 185-196. Fulton R.A., Breidenbach J.P., Seo D.J., Miller D.A., O’Bannon T., (1998), The WSR-88D Rainfall Algorithm, Weather Forecast, 13, 377–395. Georgakakos K.P., Spenser C., (2009), Analysis of real time rainfall gauge and radar data over Romania, HRC Technical Communication ROFFG-2009-01, Hydrologic Research Center, San Diego, CA, USA, 26 May 2009. Goudenhoofdt E., Delobbe L., (2009), Evaluation of radargauge merging methods for quantitative precipitation estimates, Hydrology and Earth System Sciences, 13, 195–203. Gjertsen U., Salek M., Michelson D.B., (2003), Gaugeadjustment of radar-based precipitation estimates – a review, On line at: http://www.smhi.se/cost717/COST 171, WDF_02_200203_3. Huff F.A., (1967), The adjustment of radar estimates of storm mean rainfall with rain gage data, Journal of Applied Meteorology, 6, 52–56. Hunter S.M., (1996), WSR-88D radar rainfall estimation: Capabilities, limitations and potential improvements, National Weather Digest, 20, 26–38. Jayakrishnan R., Srinivasan R., Arnold J.G., (2004), Comparison of raingage and WSR-88D Stage III precipitation data over the Texas-Gulf basin, Journal of Hydrology, 292, 135-152. Krajewski W.F., Vignal B., Seo B.C., Villarini G., (2011), Statistical model of the range dependent error in radarrainfall estimates due to the vertical profile of reflectivity, Journal of Hydrology, 402, 306-316. Krajewski W.F., Villarini G., Smith J.A., (2010), Radarrainfall uncertainties: Where are we after thirty years of effort?, Bulletin of the American Meteorological Society, 91, 87-94. Kucera P.A., Krajewski W.F., Young C.B., (2004), Radar beam occultation studies using GIS and DEM technology: An example study of Guam, Journal of Atmospheric and Oceanic Technology, 21, 995-1006. Legates D.R., (1987), A climatology of global precipitation, Publications in Climatology, 40(1), 86. Legates D.R., (2000), Real-time calibration of radar precipitation estimates, Professional Geographer, 52, 235–246. Smith J.A., Seo D.J., Baeck M.L., Hudlow M.D., (1996), An intercomparison study of NEXRAD precipitation estimates, Water Resources Research, 32, 2035–2045. Trapero L., Bech J., Rigo T., Pineda N., Forcadell D., (2009), Uncertainty of precipitation estimates in convective events by the Meteorological Service of Catalonia radar network, Atmospheric Research, 93, 408–418. Ulbrich C.W., Lee L.G., (1999), Rainfall Measurement Error by WSR-88D Radars due to Variations in Z–R Law Parameters and the Radar Constant, Journal of Atmospheric and Oceanic Technology, 16, 1017– 1024. Villarini G., Krajewski W.F., Ntelekos A.A., Georgakakos K.P., Smith J.A., (2010), Towards probabilistic forecasting of flash floods: The combined effects of


uncertainty in radar-rainfall and flash flood guidance, Journal of Hydrology, 394, 275-284. Villarini G., Krajewski W.F., (2010a), Review of the different sources of uncertainty in single-polarization radar-based estimates of rainfall, Survey in Geophysics, 31, 107-129. Villarini G., Krajewski W.F., (2010b), Sensitivity studies of the models of radar-rainfall uncertainties, Journal of Applied Meteorology and Climatology, 49, 288-309. Villarini G., Krajewski W.F., (2008), Empirically-based modeling of spatial sampling uncertainties associated with rainfall measurements by rain gauges, Advances in Water Resources, 31, 1015-1023. Villarini G., Mandapaka P.V., Krajewski W.F., Moore R.J., (2008), Rainfall and sampling errors: A rain gauge perspective, Journal of Geophysical Research, 113, D11102, doi:10.1029/2007JD009214. Wescott N.E., Vernon Knapp H., Hilberg S.D., (2008), Comparison of gage and multi-sensor precipitation

estimates over a range of spatial and temporal scales in the Midwestern United States, Journal of Hydrology, 351, 1-12. Young C.B., Nelson B.R., Bradley A.A., Smith J.A., Peters-Lidard C.D., Kruger A., Baeck M.L., (1999), An evaluation of NEXRAD precipitation estimates in complex terrain, Journal of Geophysical Research, 14(D16), 19691-19703. Xie H., Zhou X., Vivoni E.R., Hendrickx J.M.H., Small E.E., (2005), GIS-based NEXRAD Stage III precipitation database: automated approaches for data processing and visualization, Computer Geosciences, 31, 65–76. Zawadzki I., (1975), On radar-raingauge comparison, Journal of Applied Meteorology, 14, 1430–1436. Zhang G., Vivekanandan J., Brandes E., (2002), Effects of random inhomogeneity on radar measurements and rain rate estimation, IEEE Transactions in Geoscience and Remote Sensing, 40, 223-227.

731