• Accadia, C., , Casaioli M. , , Mariani S. , , Lavagnini A. , , and Speranza A. , 2003a: A diagnostic study of a Mediterranean cyclone using a regional model and SSM/I observations. Fifth Plinius Conf. on Mediterranean Storms, Ajaccio, France, European Geosciences Union, 159–164.

  • Accadia, C., , Mariani S. , , Casaioli M. , , Lavagnini A. , , and Speranza A. , 2003b: Sensitivity of precipitation forecast skill scores to bilinear interpolation and a simple nearest-neighbor average method on high-resolution verification grids. Wea. Forecasting, 18 , 918932.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Accadia, C., , Mariani S. , , Casaioli M. , , Lavagnini A. , , and Speranza A. , 2005: Verification of precipitation forecasts from two limited-area models over Italy and comparison with ECMWF forecasts using a resampling technique. Wea. Forecasting, 20 , 276300.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Alishouse, J. C., , Snyder S. , , Vongsathorn J. , , and Ferrero R. R. , 1990: Determination of oceanic total precipitable water from the SSM/I. IEEE Trans. Geosci. Remote Sens., 28 , 811816.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Amitai, E., , Nystuen J. A. , , Liao L. , , Meneghini R. , , and Morin E. , 2004: Uniting space, ground, and underwater measurement for improved estimates of rain rate. IEEE Trans. Geosci. Remote Sens., 1 , 811816.

    • Search Google Scholar
    • Export Citation
  • Arkin, P. A., , and Meisner B. , 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–1984. Mon. Wea. Rev., 115 , 5174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Baldwin, M. E., cited. 2000: QPF verification system documentation. [Available online at http://wwwt.emc.ncep.noaa.gov/mmb/ylin/pcpverif/scores/docs/mbdoc/pptmethod.html.].

  • Barnes, S. L., 1964: A technique for maximizing details in numerical weather map analysis. J. Appl. Meteor., 3 , 396409.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Barnes, S. L., 1973: Mesoscale objective analysis using weighted time-series observations. NOAA Tech. Memo. ERL NSSL-62, National Severe Storms Laboratory, Norman, OK, 60 pp. [NTIS COM-73-10781.].

  • Barrett, E. C., 1970: The estimation of monthly rainfall from satellite data. Mon. Wea. Rev., 98 , 322327.

  • Buzzi, A., , and Foschini L. , 2000: Mesoscale meteorological features associated with heavy precipitation in the southern Alpine region. Meteor. Atmos. Phys., 72 , 131146.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buzzi, A., , Fantini M. , , Malguzzi P. , , and Nerozzi F. , 1994: Validation of a limited area model in cases of Mediterranean cyclogenesis: Surface fields and precipitation scores. Meteor. Atmos. Phys., 53 , 5367.

    • Search Google Scholar
    • Export Citation
  • Casaioli, M., , Mariani S. , , Accadia C. , , Tartaglione N. , , Speranza A. , , Lavagnini A. , , and Bolliger M. , 2006: Unsatisfying forecast of a Mediterranean cyclone: A verification study employing state-of-the-art techniques. Adv. Geosci., 7 , 379386.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cressie, N. A. C., 1993: Statistics for Spatial Data. John Wiley and Sons, 900 pp.

  • Doelling, I. G., , Joss J. , , and Riedl J. , 1998: Systematic variations of Z–R relationships from drop size distributions measured in northern Germany during seven years. Atmos. Res., 48 , 635649.

    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., 2004: Monitoring the quality of operational and semi-operational satellite precipitation estimates: The IPWG validation/intercomparison study. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division and EUMETSAT. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Ebert1.pdf.].

  • Ebert, E. E., , and McBride J. L. , 2000: Verification of precipitation in weather systems: Determination of systematic errors. J. Hydrol., 239 , 179202.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., , Damrath U. , , Wergen W. , , and Baldwin M. E. , 2003: The WGNE assessment of short-term quantitative precipitation forecasts. Bull. Amer. Meteor. Soc., 84 , 481492.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., , Janowiak J. E. , , and Kidd C. , 2007: Comparison of near-real-time precipitation estimates from satellite observations and numerical models. Bull. Amer. Meteor. Soc., 88 , 4764.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraro, R. R., 1997: Special Sensor Microwave Imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 , 1671516735.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraro, R. R., , and Marks G. F. , 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12 , 755770.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fisher, R. A., 1925: Statistical Methods for Research Workers. Oliver and Boyd, 239 pp.

  • Gabella, M., , Michaelides S. , , and Perona G. , 2005: Preliminary comparison of TRMM and ground-based precipitation radar for a European test site. Int. J. Remote Sens., 26 , 9971006.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gabella, M., , Joss J. , , Perona G. , , and Michaelides S. , 2006: Range adjustment for ground-based radar, derived with the spaceborne TRMM Precipitation Radar. IEEE Trans. Geosci. Remote Sens., 44 , 126133.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gottschalck, J., , Meng J. , , Rodell M. , , and Houser P. , 2005: Analysis of multiple precipitation products and preliminary assessment of their impact on global land data assimilation system land surface states. J. Hydrometeor., 6 , 573598.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grams, J. S., , Gallus W. A. , , Koch S. E. , , Wharton L. S. , , Loughe A. , , and Ebert E. E. , 2006: The use of a modified Ebert–McBride technique to evaluate mesoscale model QPF as a function of convective system morphology during IHOP 2002. Wea. Forecasting, 21 , 288306.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hoffman, R. N., , Liu Z. , , Louis J-F. , , and Grassotti C. , 1995: Distortion representation of forecast errors. Mon. Wea. Rev., 123 , 27582770.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hollinger, J. P., 1989: DMSP Special Sensor Microwave/Imager Calibration/Validation. Final Rep., Vol. I., Space Sensing Branch, Naval Research Laboratory, Washington, DC, 186 pp. [Available from Naval Research Laboratory, Washington, DC 20375.].

  • Houze R. A. Jr., , , Brodzik S. , , Schumacher C. , , Yuter S. E. , , and Williams C. R. , 2004: Uncertainties in oceanic radar rain maps at Kwajalein and implications for satellite validation. J. Appl. Meteor., 43 , 11141132.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Iguchi, T., , Kozu T. , , Meneghini R. , , Awaka J. , , and Okamoto K. , 2000: Rain-profiling algorithm for the TRMM Precipitation Radar. J. Appl. Meteor., 39 , 20382052.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ikai, J., , and Nakamura K. , 2003: Comparison of rain rates over the ocean derived from TRMM Microwave Imager and Precipitation Radar. J. Atmos. Oceanic Technol., 20 , 17091726.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Imaoka, K., , and Spencer R. W. , 2000: Diurnal variation of precipitation over the tropical oceans observed by TRMM/TMI combined with SSM/I. J. Climate, 13 , 41494158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , Xie P. , , Joyce R. J. , , Chen M. , , and Yarosh Y. , 2004: Validation of satellite-derived rainfall estimates and numerical model forecasts of precipitation over the United States. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division and EUMETSAT. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Janowiak.pdf.].

  • Kalman, R. E., 1960: A new approach to linear filtering and prediction problems. Trans. ASME J. Basic Eng., 82D , 3545.

  • Kidd, C., 2004: Validation of satellite rainfall estimates over the mid-latitudes. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division, and EUMETSAT, 10 pp. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Kidd.pdf.].

  • Koch, S. E., , desJardins M. , , and Kocin P. J. , 1983: An interactive Barnes objective map analysis scheme for use with satellite and conventional data. J. Climate Appl. Meteor., 22 , 14871503.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozu, T., and Coauthors, 2001: Development of precipitation radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. IEEE Trans. Geosci. Remote Sens., 39 , 102115.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kummerow, C., , Barnes W. , , Kozu T. , , Shiue J. , , and Simpson J. , 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15 , 809817.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lagouvardos, K., , Kotroni V. , , Koussis A. , , Feidas C. , , Buzzi A. , , and Malguzzi P. , 2003: The meteorological model BOLAM at the National Observatory of Athens: Assessment of two-year operational use. J. Appl. Meteor., 42 , 16671678.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Malguzzi, P., , and Tartaglione N. , 1999: An economical second order advection scheme for numerical weather prediction. Quart. J. Roy. Meteor. Soc., 125 , 22912303.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mariani, S., and Coauthors, 2005: A limited area model intercomparison on the “Montserrat-2000” flash-flood event using statistical and deterministic methods. Nat. Hazards Earth Syst. Sci., 5 , 565581.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , and Palmer W. M. , 1948: The distribution of raindrops with size. J. Meteor., 5 , 165166.

  • Marzano, F. S., , Palmacci M. , , Cimini D. , , Giuliani G. , , and Turk F. J. , 2004: Multivariate statistical integration of satellite infrared and microwave radiometric measurements for rainfall retrieval at the geostationary scale. IEEE Trans. Geosci. Remote Sens., 42 , 10181032.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mazzetti, C., 2004: Multi-sensor rainfall measurements for flood forecasting. Ph.D. Dissertation, University of Bologna, 238 pp. [Available from Dept. of Earth Sciences and Environment, University of Bologna, Piazza di Porta San Donato 1, 40126 Bologna, Italy.].

  • Mazzetti, C., 2006: RAINMUSIC software manual. PROGEA S.r.l., Bologna, Italy, 69 pp. [Available from PROGEA S.r.l., Via Galliera 93, 40121 Bologna, Italy.].

  • Mesinger, F., 1996: Improvements in quantitative precipitation forecasting with the Eta regional model at the National Centers for Environmental Prediction: The 48-km upgrade. Bull. Amer. Meteor. Soc., 77 , 26372649.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michaelides, S. C., , Gabella M. , , Constantinides P. , , and Perona G. , 2004: Comparison between TRMM data and in situ measurement in the island of Cyprus. Proc. Second Voltaire Workshop, Ljubljana, Slovenia, VOLTAIRE Project, 47–52.

  • Panofsky, H. A., , and Brier G. W. , 1958: Some Applications of Statistics to Meteorology. The Pennsylvania State University, 224 pp.

  • Porcù, F., , Prodi F. , , Pinori S. , , Dietrich S. , , and Mugnai A. , 2003: On the reliability of current MW-IR techniques in quantitative precipitation measurements over Mediterranean basin. Proc. Fourth Plinius Conf. on Mediterranean Storms, Mallorca, Spain, European Geophysical Society, 161–170.

  • Schumacher, C., , and Houze R. A. Jr., 2000: Comparison of radar data from the TRMM satellite and Kwajalein oceanic validation site. J. Appl. Meteor., 39 , 21512164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sohn, B., , and Smith E. , 2003: Explaining sources of discrepancy in SSM/I water vapor algorithms. J. Climate, 16 , 32293255.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tartaglione, N., , Mariani S. , , Accadia C. , , Speranza A. , , and Casaioli M. , 2005: Comparison of rain gauge observations with modeled precipitation over Cyprus using contiguous rain area analysis. Atmos. Chem. Phys., 5 , 21472154.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Todini, E., 2001: Bayesian conditioning of radar to rain-gauges. Hydrol. Earth System Sci., 5 , 225232.

  • VOLTAIRE, 2006: Validation of multisensor precipitation fields and numerical modeling in Mediterranean test sites. Final Rep., 135 pp. [Available from Dipartimento Tutela Acque Interne e Marine, APAT, Via Curtatone 3, I-00185 Rome, Italy.].

  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.

  • WWRP/WGNE Joint Working Group on Verification, cited. 2004: Recommendations for the verification and intercomparison of QPFs from operational NWP models. [Available online at http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/WGNE/QPF_verif_recomm.pdf.].

  • WWRP/WGNE Joint Working Group on Verification, cited. 2007: Forecast verification—Issue, methods and FAQ. [Available online at http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/verif_web_page.html.].

  • Wylie, D. D., 1979: An application of a geostationary satellites rain estimation technique to an extratropical area. J. Appl. Meteor., 18 , 16401648.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1995: An intercomparison of gauge observations and satellite estimates of monthly precipitation. J. Appl. Meteor., 34 , 11431160.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Geographical map of the eastern Mediterranean Sea area.

  • View in gallery

    (a) Geographical distribution over Cyprus of the 147 rain gauges stations (black circles), and location of the ground-based radar (white circle). The Achna and the Prodromos rain gauges and the Kykkos weather radar are indicated with arrows. Orography is indicated on a grayscale and the high massif visible in the figure is Troodos. (b) Localization of the two obscured sectors in the radar (white circle) data, which are caused by the presence of the Troodos high massif, in the southeast direction, and by the Tripylos peak in the northwest direction. The beam occultation at 0° elevation is emphasized using light gray. A darker grayscale is used for orography.

  • View in gallery

    Mean sea level pressure (MSLP) analysis at (a) 0000 UTC 5 Mar and (b) 0000 UTC 6 Mar 2003 over Europe (source: The Met Office). Cyprus and the conterminous area are indicated with a dashed box.

  • View in gallery

    Atmospheric forecast fields at 0600 UTC 5 Mar 2003: (a) 850-hPa and (b) 500-hPa geopotential height (m), (c) 10-m wind vectors, and (d) 2-m specific humidity (10−4 kg kg−1). The areas where specific humidity is larger than 9 × 10−3 kg kg−1 are indicated in gray.

  • View in gallery

    As in Fig. 4 but at 1800 UTC.

  • View in gallery

    Contours (mm) of precipitation modeled by BOLAM. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

  • View in gallery

    Contours (mm) of the rain gauge–based gridded analysis obtained using a two-pass Barnes scheme. Precipitation analysis is accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

  • View in gallery

    Contours (mm) of the radar–rain gauge composite using (a) OGRD and (b) RGRD. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

  • View in gallery

    As in Fig. 8 but for the radar–rain gauge composite obtained using the RainMusic software.

  • View in gallery

    TCWV verification result over the eastern Mediterranean Sea from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003. (left) The 0.27° (HR) BOLAM forecasts, (center) the forecast fields shifted as result of the Hoffman object-oriented method, and (right) the SSM/I TCWV fields. Data are masked over the area where the SSM/I-retrieved TCWV is available: (a)–(c) 0600 UCT 5 Mar, (d)–(f) 0700 UTC 5 Mar, (g)–(i) 0800 UTC 5 Mar, (j)–(l) 1800 UTC 5 Mar, (m)–(o) 1900 UTC 5 Mar, (p)–(r) 0500 UTC 6 Mar, and (s)–(u) 0600 UTC 6 Mar.

  • View in gallery

    (Continued)

  • View in gallery

    The 2D shift analysis, minimizing rmse, between SSM/I TCWV estimates and the BOLAM forecast from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003: (left) rmse and (right) the correlation. The best shift is plotted with a cross symbol with the displacement values in the longitudinal (E) and latitudinal (N) directions. Solid lines indicate statistically significant shifts, whereas dashed lines indicate statistically nonsignificant shifts: (a), (b) 0600 UCT 5 Mar; (c), (d) 0700 UTC 5 Mar; (e), (f) 0800 UTC 5 Mar; (g), (h) 1800 UTC 5 Mar; (i), (j) 1900 UTC 5 Mar; (k), (l) 0500 UTC 6 Mar; and (m), (n) 0600 UTC 6 Mar.

  • View in gallery

    (Continued)

  • View in gallery

    Contours (mm) of the forecast precipitation shifted w.r.t. (a) POGRD, (b) PRGRD, (c) RMOGRD, and (d) RMRGRD using MSE as the CRA pattern-matching criterion and sv equal to 9. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

  • View in gallery

    As in Fig. 12 but using the correlation as the CRA pattern-matching criterion.

  • View in gallery

    The 2D shift analysis, maximizing correlation, of CRA between the 0.09° BOLAM precipitation forecast and the rain gauge–radar composite (a) PRGRD and (b) RMRGRD using an isohyet equal to 5.0 mm (24 h)−1. Maximum values found during CRA are indicated with a cross symbol. The grayscale, from lighter to darker, indicates the progressive order in which these values are found. The suspicious final shift (top-left corner) is indicated with a black square and the displacement values in the longitudinal (E) and latitudinal (N) directions. The more reliable shift is indicated instead with a black circle. Solid lines indicate statistically significant shifts, whereas dashed lines indicate statistically nonsignificant shifts.

  • View in gallery

    Comparison during the event between hourly precipitation observed (dashed line, with square symbols) at the (a) Prodromos and (b) Achna stations and the forecast at the nearest BOLAM grid point (solid line, with diamond symbols).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 14 14 2
PDF Downloads 10 10 1

Multisensor Comparison and Numerical Modeling of Atmospheric Water Fields: A VOLTAIRE Case Study over Cyprus

View More View Less
  • 1 Agency for Environmental Protection and Technical Services, Rome, and Department of Mathematics, University of Ferrara, Ferrara, Italy
  • | 2 EUMETSAT, Darmstadt, Germany
  • | 3 Department of Physics, University of Camerino, Camerino, Italy
  • | 4 Agency for Environmental Protection and Technical Services, Rome, Italy
  • | 5 Department of Electronics, Politecnico di Torino, Turin, Italy
  • | 6 Meteorological Service, Nicosia, Cyprus
  • | 7 Department of Mathematics and Informatics, University of Camerino, Camerino, Italy
© Get Permissions
Full access

Abstract

This paper presents a study performed within the framework of the European Union’s (EU) VOLTAIRE project (Fifth Framework Programme). Among other tasks, the project aimed at the integration of the Tropical Rainfall Measuring Mission (TRMM) data with ground-based observations and at the comparison between water fields (precipitation and total column water vapor) as estimated by multisensor observations and predicted by NWP models. In particular, the VOLTAIRE project had as one of its main objectives the goal of assessing the application of satellite-borne instrument measures to model verification. The island of Cyprus was chosen as the main “test bed,” because it is one of the few European territories covered by the passage of the TRMM Precipitation Radar (PR) and it has a dense rain gauge network and an operational weather radar. TRMM PR provides, until now, the most reliable space-borne spatial high-resolution precipitation measurements.

Attention is focused on the attempt to define a methodology, using state-of-the-art diagnostic methods, for a comprehensive evaluation of water fields as forecast by a limited area model (LAM). An event that occurred on 5 March 2003, associated with a slow cyclone moving eastward over the Mediterranean Sea, is presented as a case study.

The atmospheric water fields were forecast over the eastern Mediterranean Sea using the Bologna Limited Area Model (BOLAM). Data from the Cyprus ground-based radar, the Cyprus rain gauge network, the Special Sensor Microwave Imager (SSM/I), and the TRMM PR were used in the comparison. Ground-based radar and rain gauge data were merged together in order to obtain a better representation of the rainfall event over the island. TRMM PR measurements were employed to range-adjust the ground-based radar data using a linear regression algorithm.

The observed total column water vapor has been employed to assess the forecast quality of large-scale atmospheric patterns; such an assessment has been performed by means of the Hoffman diagnostic method applied to the entire total column water vapor field. Subsequently, in order to quantify the spatial forecast error at the finer BOLAM scale (0.09°), the object-oriented contiguous rain area (CRA) analysis was chosen as a comparison method for precipitation. An assessment of the main difficulties in employing CRA in an operational framework, especially over such a small verification domain, is also discussed in the paper.

Corresponding author address: Dr. Stefano Mariani, Agency for Environmental Protection and Technical Services (APAT), Via Curtatone 3, 00185 Rome, Italy. Email: stefano.mariani@apat.it

Abstract

This paper presents a study performed within the framework of the European Union’s (EU) VOLTAIRE project (Fifth Framework Programme). Among other tasks, the project aimed at the integration of the Tropical Rainfall Measuring Mission (TRMM) data with ground-based observations and at the comparison between water fields (precipitation and total column water vapor) as estimated by multisensor observations and predicted by NWP models. In particular, the VOLTAIRE project had as one of its main objectives the goal of assessing the application of satellite-borne instrument measures to model verification. The island of Cyprus was chosen as the main “test bed,” because it is one of the few European territories covered by the passage of the TRMM Precipitation Radar (PR) and it has a dense rain gauge network and an operational weather radar. TRMM PR provides, until now, the most reliable space-borne spatial high-resolution precipitation measurements.

Attention is focused on the attempt to define a methodology, using state-of-the-art diagnostic methods, for a comprehensive evaluation of water fields as forecast by a limited area model (LAM). An event that occurred on 5 March 2003, associated with a slow cyclone moving eastward over the Mediterranean Sea, is presented as a case study.

The atmospheric water fields were forecast over the eastern Mediterranean Sea using the Bologna Limited Area Model (BOLAM). Data from the Cyprus ground-based radar, the Cyprus rain gauge network, the Special Sensor Microwave Imager (SSM/I), and the TRMM PR were used in the comparison. Ground-based radar and rain gauge data were merged together in order to obtain a better representation of the rainfall event over the island. TRMM PR measurements were employed to range-adjust the ground-based radar data using a linear regression algorithm.

The observed total column water vapor has been employed to assess the forecast quality of large-scale atmospheric patterns; such an assessment has been performed by means of the Hoffman diagnostic method applied to the entire total column water vapor field. Subsequently, in order to quantify the spatial forecast error at the finer BOLAM scale (0.09°), the object-oriented contiguous rain area (CRA) analysis was chosen as a comparison method for precipitation. An assessment of the main difficulties in employing CRA in an operational framework, especially over such a small verification domain, is also discussed in the paper.

Corresponding author address: Dr. Stefano Mariani, Agency for Environmental Protection and Technical Services (APAT), Via Curtatone 3, 00185 Rome, Italy. Email: stefano.mariani@apat.it

1. Introduction

Precipitation over the Mediterranean area is of great interest and importance for the weather, climate, meteorological, and hydrological communities. Focusing on NWP model verification, the quality of quantitative precipitation forecasts (QPFs) is indeed considered by several operational meteorological centers to be a general indicator of the capability of a NWP model to produce a good forecast (e.g., WWRP/WGNE Joint Working Group on Verification 2004; Ebert et al. 2003; Mesinger 1996), because rainfall strongly depends on atmospheric motion, moisture content, and physical processes.

Verification studies and diagnostics of precipitation and humidity have seldom been performed over the Mediterranean region and in those few cases, only over land (e.g., Accadia et al. 2005; Lagouvardos et al. 2003), because of the sparseness of available in situ observations. Small islands or coastal locations covered by ground-based radars are an exception, because radars can also provide precipitation estimates over the conterminous waters. Satellite-borne instruments provide a possible alternative, at least in terms of coverage. The estimation process can be performed through some algorithms starting from available observations obtained in the infrared (IR) and microwave (MW) spectral regions (Marzano et al. 2004). Satellite precipitation estimates from passive microwave data are currently validated in some research and operational meteorological centers by means of in situ observations (ground radars and rain gauges), assessing the existing uncertainties (e.g., Janowiak et al. 2004; Kidd 2004; Ebert 2004; Ebert et al. 2007).

However, this kind of satellite precipitation estimation could be less accurate than the rainfall estimates made from ground-based instruments (Gottschalck et al. 2005), because passively observed IR and MW radiances are only indirectly related to rainfall (Porcù et al. 2003). Moreover, IR precipitation estimates are sensitive to colder IR temperatures and can confuse cold cloud tops for precipitation (e.g., Wylie 1979; Arkin and Meisner 1987), whereas MW precipitation estimates are hampered by poor temporal sampling. The precipitation radar (PR) on board the Tropical Rainfall Measuring Mission (TRMM) satellite allows, instead, an estimate of precipitation (less indirect than the ones obtained by other satellite sensors) that can be compared with rain gauge observations (e.g., Amitai et al. 2004). In fact, it provides, at present, the most reliable space-borne rainfall measurements, because it was designed to provide three-dimensional maps of storm structures (information online at http://trmm.gsfc.nasa.gov/).

The availability of TRMM PR measurements over the Mediterranean Sea and the island of Cyprus, in particular, was at the core of the European VOLTAIRE project—Fifth Framework Programme (VOLTAIRE 2006). The project had as one of its main objectives the integration of TRMM data with ground-based observations, and in particular the use of TRMM PR observations for the improvement of the precipitation analysis fields and its use in NWP verification study. Moreover, as discussed later, alternative observations from space, like total column water vapor (TCWV), can be used to perform a model evaluation over otherwise in situ data-void regions.

The island of Cyprus, located in the eastern Mediterranean Sea (see Fig. 1), was an optimal test site for the project’s activities. Cyprus is one of the few European lands covered by TRMM. Moreover, this island is also instrumented with a dense rain gauge network and a weather radar (see Fig. 2), which also allows for measuring precipitation over the surrounding waters. The island of Cyprus plays a role similar to that of the Kwajalein atoll in the Marshall Islands of the western tropical Pacific Ocean, which represents the TRMM oceanic tropical ground validation site (e.g., Schumacher and Houze 2000; Houze et al. 2004).

The VOLTAIRE project provided the opportunity to assess the use of multisensor data and state-of-the-art verification techniques by performing diagnostic studies and comparing atmospheric water fields modeled by the Bologna Limited Area Model (BOLAM) with the ones estimated using rain gauge, radar, and satellite measurements. Relatively heavy rain events over the Cyprus area were the object of several case studies (VOLTAIRE 2006).

The aim of this work is to present the initial definition of a forecast assessment methodology developed within the VOLTAIRE project. The focus is mainly on the “water fields,” thus verifying the quality of BOLAM forecasts of precipitation and humidity fields against available observations from ground radar and rain gauges, and from satellite-borne instruments. In particular, a simple correction scheme is applied to Cyprus’s ground radar rainfall estimates, using also the TRMM PR rainfall estimates. The assessment of a relatively heavy precipitation event that occurred over Cyprus on 5 March 2003 is presented as an example of the methodology.

Because the TRMM PR overpasses the island of Cyprus once or twice in a day, a comparison with simultaneous precipitation forecasts is problematic. The main obstacle to the use of the satellite data over short time ranges (1 day) lies in the poor temporal sampling. In fact, good agreement between precipitation estimated from space and in situ measurements can be found when point measurements are spatially averaged over the radar pixels (or over a coarser grid) and the comparison is performed on long time scales, for instance, on a monthly scale (e.g., Barrett 1970; Ikai and Nakamura 2003; Imaoka and Spencer 2000). Such a comparison has been performed over the Cyprus region by Michaelides et al. (2004) and Gabella et al. (2005). Furthermore, the comparison between precipitation observed by satellite sensors and forecast precipitation is also problematic because satellite sensors give almost instantaneous information, while forecast precipitation is usually accumulated over a few hours or on a daily basis.

The above-described difficulty does not hold for the comparison of the TCWV. In fact, this field can be directly compared with model outputs because it is not the result of any accumulation in time. Comparison of TCWV forecast fields with water vapor fields retrieved from orbiting platforms, such as the Special Sensor Microwave Imager (SSM/I), is expected to provide some insight into the quality of the meteorological simulations. It is evident that a correct forecast of the 2D TCWV structure is a prerequisite for a correct prediction of space–time rainfall patterns, keeping in mind that the link between water vapor and precipitation is not straightforward. Moreover, TCWV is spatially smoother than precipitation, partially reducing the requirements of the spatial comparison. Indeed, it is nontrivial to provide an accurate estimate of the precipitation fields because of its high variability in space and in time. Such observed variability might be difficult to forecast by meteorological models for any number of reasons (e.g., initialization of humidity fields or the simplified physics and parameterization schemes employed). So, it is not surprising that the modeled precipitation shows some differences (in space and time) with respect to the observed rainfall. For instance, such differences might penalize a forecast, misplacing the position of the precipitation patterns, especially when verifying at high spatial and temporal resolutions (WWRP/WGNE Joint Working Group on Verification 2007), and resulting in a double-penalty effect.1 For a case study approach, because of the small size of the statistical sample, it could be misleading to perform precipitation verification only by means of standard nonparametric statistical methods (Wilks 1995).

Thus, in this study a contiguous rain area (CRA) analysis (Ebert and McBride 2000) has been employed, in order to provide a quantitative estimation of the most relevant qualitative features that characterize the difference between the forecast and observed precipitation patterns. This method was chosen for its ability to quantify the displacement of the forecast precipitation pattern with respect to the observed one. Another diagnostic verification method, developed by Hoffman et al. (1995), was applied to water vapor. The matching criterion is similar to CRA, showing the detection and characterization of TCWV analysis errors in terms of position errors.

The paper is organized as follows. A brief meteorological description of the event is presented in section 2. Forecast and observed data used in the study are described in section 3. Methodologies applied in the diagnostic verification are briefly reported in section 4. Section 5 illustrates the results of the multisensor comparison and a discussion of the results. Finally, conclusions are reported in section 6.

2. Meteorological event description

The rain event took place over the eastern Mediterranean Sea (Fig. 1) from 4 to 6 March 2003. It was associated with a cyclonic circulation, which moved from the western to the eastern Mediterranean Sea. Moderate rainfall occurred over Cyprus during the morning of 5 March. When the cold front (see Fig. 3) passed during the afternoon of 5 March, thunderstorms occurred. Rainfall hit the island of Cyprus, particularly its southern part; with an observed rain gauge maximum of about 40.0 mm in 24 h.

A description of the event as forecast by BOLAM is given in detail as follows. A cyclonic circulation was present over the Aegean Sea, north of Crete, at 0600 UTC 5 March (Fig. 4a). The cyclone was in its final phase, as can be seen by comparing the 500- and 850-hPa geopotential charts (Figs. 4a and 4b), which clearly show that a westward tilt of the lows and highs with altitude was almost absent. The pressure gradient between the eastern high pressure and the cyclone placed near the Greek coasts caused a strong ageostrophic wind at lower levels. Wind blew along the coastline of the Antalya Gulf and impinged upon the western part of the Taurus Mountains (Fig. 4c). A warm low-level jet, within the cyclone warm sector, approached the coast of Lycia. The flow was associated with a region of high specific humidity values (Fig. 4d). At 1800 UTC 5 March, the low pressure center was located between Crete and Cyprus (Figs. 5a and 5b). Two low-level jets were present over the eastern Mediterranean Sea, embedded in the overall cyclonic circulation (Fig. 5c). In one of these jets, wind vectors followed the coastline of Egypt and Israel and then rotated cyclonically toward Cyprus. The other jet, north of Cyprus, followed the coastline of Turkey and was directed in the Gulf of Antalya, impinging upon the Taurus Mountains. At this stage, the southerly circulation brought moist surface air from the northern coast of Africa toward the coasts of Cyprus and Turkey (Fig. 5d). At 0600 UTC 6 March, the cyclone approached Cyprus (not shown). The low-level jet (previously present over the Gulf of Antalya) merged into the cyclonic circulation. The southerly flow of this cyclonic circulation hit Cyprus. Orographically enhanced heavy precipitation occurred on Cyprus’s southwest side (Fig. 6), as the flow impinged upon the Troodos high massif.

3. Forecasts and observations

a. The BOLAM model

The NWP model used in this work is the BOLAM (Buzzi et al. 1994), which was developed at the Bologna branch of the Istituto di Scienze dell’Atmosfera e del Clima–Consiglio Nazionale delle Ricerche (ISAC–CNR). For a thorough description of the BOLAM model, its parameterization, and its main characteristics, readers may refer to Malguzzi and Tartaglione (1999), Buzzi and Foschini (2000), and Mariani et al. (2005).

The 0.5°-resolution, 60-hybrid level, European Centre for Medium-Range Weather Forecasts (ECMWF) analyses were first horizontally interpolated onto the 0.27° model domain (outer domain) and then vertically interpolated onto 40 sigma levels. The boundary conditions were provided by the 6-h ECMWF forecasts. The 0.27° BOLAM domain covers the entire Mediterranean region, including part of the eastern Atlantic. For this study, outputs of the 0.27° BOLAM model are one-way nested onto a domain covering the eastern Mediterranean region with a grid spacing of 0.09° (inner domain). Initial conditions were provided by the ECMWF analysis at 1200 UTC 4 March 2003. Boundary conditions coming from the ECMWF 6-hourly forecasts were imposed. The low-resolution BOLAM provided the initial and boundary conditions for the 0.09° BOLAM; the initial condition was the 0.27° BOLAM 12-h forecast at 0000 UTC 5 March 2003.

For the precipitation comparison, the high-resolution precipitation forecasts were accumulated on a daily basis, from 0600 UTC 5 March to 0600 UTC 6 March 2003, according to the accumulation time of the rain gauge network. Furthermore, the model output was interpolated from the native rotated grid onto a latitude–longitude grid (the verification grid), with a horizontal grid spacing of 0.09°, using the remapping procedure.2 This grid-to-grid transformation, which is operationally employed at the National Centers for Environmental Prediction (NCEP; Baldwin 2000), is preferable to a simple bilinear interpolation scheme, especially when comparing precipitation, because it conserves to a desired degree of accuracy the total forecast of the native grid (Accadia et al. 2003b; WWRP/WGNE Joint Working Group on Verification 2004). Contours of the forecast precipitation are shown in Fig. 6.

The 0.27° TCWV model outputs (about 30 km) were used because TCWV estimates retrieved from SSM/I data have similar spatial resolution. A postprocessing procedure, which sums up over the sigma levels the specific humidity weighted with the layer thickness, has been applied to obtain the BOLAM TCWV. A comparison was subsequently performed between the TCWV calculated for each of the SSM/I satellite passages over the eastern Mediterranean Sea and the corresponding BOLAM TCWV (see Table 1).

b. SSM/I data

The SSM/I sensor is a passive microwave radiometer (linearly polarized) that provides brightness temperatures at 19.35, 22.235, 37.0, and 85.5 GHz. This instrument performs a conical scan with an angle of 53.1° at the surface and a swath of about 1400 km (Hollinger 1989). The sensor is on board the Defense Meteorological Satellite Program (DMSP) platforms, which are in a polar sun-synchronous orbit. The SSM/I data used in this study are freely available from the Comprehensive Large Array-data Stewardship System (CLASS) of the National Oceanic and Atmospheric Administration (NOAA). Data from three DMSP platforms were available (F-13, F-14, and F-15), providing extensive and dense coverage of the Mediterranean Sea.

From SSM/I brightness temperatures, it is possible to retrieve, among many other geophysical parameters, TCWV and precipitation. However, the latter estimate was not considered in the comparison, because of the insufficient temporal sampling. As mentioned above, the SSM/I has a vertical polarization channel at 22.235 GHz. This channel frequency is at the peak of a weak water vapor absorption line. Over the ocean, this allows the retrieval of TCWV. Retrievals over land are impossible because of the high and varying emissivity of land surfaces. The TCWV mainly corresponds to lower-level water vapor (i.e., 700 hPa and below). TCWV estimates are computed only where precipitation is absent. The precipitation algorithm from Ferraro and Marks (1995) and Ferraro (1997) is used as a screening procedure. The regression algorithm of Alishouse et al. (1990) is used to estimate TCWV. Sohn and Smith (2003) pointed out that TCWV algorithms might not be completely appropriate for regional studies. Biases are likely because of a reduced representation of all possible climatological situations in the original training set. In fact, Accadia et al. (2003a) have found (at least over a 20-day time period) that TCWV estimates over the Mediterranean Sea made by using the same algorithm have a bias greater than 1 kg m−2 when compared against radiosondes. Thus, a computation of the bias is performed to evaluate the relative importance of this error in the frame of the Hoffman analysis.

c. TRMM PR data

Kummerow et al. (1998) offer a comprehensive description of the TRMM sensor packages. A complete description of the Ku-band TRMM PR sensor can be found in Kozu et al. (2001). TRMM PR data are attenuation-corrected radar reflectivities obtained at 13.8 GHz with the TRMM 2A25 algorithm, which produces the best estimate of the rain profile close to the ground level (Iguchi et al. 2000). In particular, in the present manuscript, the authors use the radar reflectivity calculated for the lowest TRMM PR pulse volume, the so-called NearSurfZ. The TRMM PR data are not directly involved in the verification, but they are used to range-adjust the ground radar data (see section 3e).

d. Rain gauge data

The Cyprus rain gauge network data, available within the framework of VOLTAIRE, were used for comparison with the rainfall modeled by BOLAM. The Meteorological Service of Cyprus manages this dense network of 147 rain gauges (Fig. 2a) located over the western part of the island (about 5895 km2).

A gridded analysis of the observed precipitation field was produced by means of a two-pass Barnes scheme (Barnes 1964, 1973; Koch et al. 1983). This technique assigns to a rain gauge observation a Gaussian weight, which is a function of distance between the rain gauge and grid-box center. A first pass is performed to produce a first-guess precipitation analysis, whereas the second pass increases the amount of detail from the previous one. For the second pass the convergence parameter is set to 0.2, while the average data spacing has been set to 0.2°. This setting is consistent with the constraint that the ratio between the grid size and the average data spacing lies between 0.3 and 0.5, as pointed out by Barnes (1973). To avoid the excessive rainfall spreading, introduced by the analysis scheme on grid points far from the actual locations of rain gauges, those grid points that do not have any rain gauge within a radius of 0.15° have been considered as data-void points in the precipitation gridded analysis. The rain gauge–based gridded analysis was accumulated over a 24-h time interval, from 0600 UTC 5 March to 0600 UTC 6 March 2003 (see Fig. 7).

e. Ground-based radar data

Within the VOLTAIRE framework, data from the Cyprus ground-based radar (Gabella et al. 2005, section 2) were also available. This radar, named Kykkos (because it is situated near the Kykkos medieval monastery) and belonging to the Meteorological Service of Cyprus, is situated in the northwestern, mountainous region of the island (34.98°N, 32.73°E; 1310 m MSL; Fig. 2a). The presence of the Troodos high massif to the southeast and of the Tripylos peak to the northwest produces two shadowed sections for which no data are available (see Fig. 2b).

In this study, rainfall estimates based on original ground-based radar data (OGRD) are obtained by using a single power law transformation. Let Z be the Kykkos ground-based radar reflectivity (mm6 m−3) and R the rainfall intensity (mm h−1). Then, the exponential fit proposed in the famous paper by Marshall and Palmer [(1948), their Eqs. (1) and (3)] was used for deriving an empirical power law:
i1520-0434-23-4-674-e1
This empirical relationship was not so different from the one reported by Doelling et al. (1998):
i1520-0434-23-4-674-e2
which was obtained using 7 yr of measurements in central Europe. Thus, for robustness of the latter relationship, Eq. (2) has been employed to obtain the OGRD rainfall estimates. However, it should be noted that while the exponent parameter in the above relationship has some effect on the results, the multiplicative parameter is compensated for by the adjustment technique described below. Radar rain intensities are combined over each rainfall period to get a first guess of the accumulated rain depth estimations.

In addition to OGRD estimates, range-adjusted values, which have been derived using the TRMM PR radar echoes as a reference, have been used. The range-adjustment technique attempts to compensate for the dramatic decrease of the radar resolution with range: the scattering volume, in fact, increases with the square of the distance. The OGRD measurements are acquired, in the present study, at distances between 10 and 120 km. Because of this large ratio of distances, the scattering volume changes by a factor of over 100. On the contrary, the scattering volume of TRMM PR has almost similar size in all the locations, because its measurements from the top originate from similar distances (402 km at the nadir and ∼420 km off nadir). This advantage of TRMM PR stimulated the idea of using TRMM radar to estimate the influence of the sampling volume of the ground-based radar.

Gabella et al. (2006) established a range-adjustment relationship by comparing TRMM PR precipitation estimates with those obtained by the ground radar, during some rain events in 2002 and 2003. For both radars, the average (in azimuth) reflectivities 〈OGRD〉2π and 〈TRMM PR〉2π in a 10-km circular ring have been computed. Because of their size, these two variables show similar behavior, except for the previously mentioned decrease in the sensitivity of the ground-based radar with distance. Moreover, averaging over the large area of the rings reduces deviations caused by rain cells of high intensity. The ratio F = 〈OGRD〉2π/〈TRMM PR〉2π is then statistically described by using a weighted regression between log(F) and the logarithm of the distance (km), log(D):
i1520-0434-23-4-674-e3
where a0 and a1 are the regression coefficients and D0 (=40 km) is a coefficient for the distance normalization. Thus, a two-coefficient scheme based on TRMM PR was applied to OGRD in order to compensate for the influence of sampling volume variations in ground-based radar. As a consequence, two different kinds of ground-based radar data were used in the precipitation verification: OGRD and the range-adjusted ground-based radar data (RGRD). Both OGRD and RGRD were 24-h accumulated, from 0600 UTC 5 March to 0600 UTC 6 March 2003, and are available on a 0.01° grid. Then, after applying the remapping procedure, they were made available on the 0.09° latitude–longitude verification grid in order to be compared with BOLAM. Values greater than 300 mm (24 h)−1, derived from residual clutter, were considered as having no data.

Radar data are useful in order to improve the rain gauge–based gridded analysis. As indicated in the previous subsection, rain gauges were available on the western part of the island. Thus, in order to obtain a better observational analysis over the island and its surrounding sea, it was decided to merge the radar data (both OGRD and RGRD) with the rain gauge data.

Two approaches were used. The first one is based on a linear combination of the two datasets, the radar data and the Barnes precipitation analysis, although only over the 0.09° verification grid points where both datasets are available, about 70 points in total (less that 20% of the considered points). If one of the two quantities is not available, the expression should instead return the other. This means that over the sea and over the eastern part of Cyprus only the radar data are considered, whereas over the radar shadow areas in Cyprus only the precipitation analysis is considered. Thus, the formula applied at each grid point in order to acquire the final observational analysis, P, is the following:
i1520-0434-23-4-674-e4
where GP is the rain gauge precipitation obtained using the Barnes analysis, RP is the radar precipitation estimated using either OGRD or RGRD, and the weights wp and wr are the linear combination coefficients. Figures 8a and 8b show the 24-h final observational analyses using the two radar datasets, hereafter indicated as POGRD and PRGRD, respectively. The relationship in Eq. (4) has the advantage of being simple and easy to implement. It may have the disadvantage of giving a higher weight to the instruments that measure more precipitation. However, in this case a higher weight was given to the Barnes precipitation analysis over almost all of the 70 points, coherently with the fact that rain gauge data, where available, provide better quantitative rainfall information, even if related to single point measurements and limited in space. This is supported by the fact that the correlation between the Barnes rain gauge analysis and the rain gauge–radar composite is stronger than the one obtained between the Barnes rain gauge analysis and the radar field alone. Namely, the correlation between the Barnes analysis and RGRD is equal to 0.36, whereas the correlation between the Barnes analysis and the PRGRD composite is indeed equal to 0.84.

The second approach employs the multisensor combination tool known as RainMusic (Todini 2001; Mazzetti 2004, 2006). This software tool has been developed by the University of Bologna within the framework of the European Union (EU) funded Multi-Sensor Precipitation Measurements Integration, Calibration and Flood Forecasting (MUSIC) project (information online at http://www.geomin.unibo.it/hydro/Music/index2.htm). The project aims to provide reliable observational analysis by combining precipitation estimated by rain gauges, weather radars, and satellites using innovative algorithms based on block-kriging (e.g., Cressie 1993) and Bayesian combination by means of a Kalman filter (Kalman 1960). For this work, only rain gauges and ground-based radar data, both OGRD and RGRD, have been considered, because the satellite data were not available on a 24-h basis. The RainMusic merging was performed as follows. First, a block-kriging rain gauge analysis (referred to as BK) was produced over a 0.01° rectangular subdomain of the verification area, which matches the contours of the island of Cyprus. This was done to avoid introducing spurious block-kriging values over the points far away from rain gauges. Then, the Bayesian combination was performed only over the subdomain points where both BK and radar data were available. For the points outside the 0.01° subdomain, only radar data were considered. Finally, the two observational fields obtained using both OGRD and RGRD were remapped over the 0.09° verification grid. Contours of these two gridded fields, referred to as RMOGRD and RMRGRD, are shown in Figs. 9a and 9b, respectively.

4. Methodologies

State-of-the-art diagnostic methods were chosen for their ability to quantify and qualify the forecast error precisely. They give quantitative support to the standard “eyeball” verification of model forecasts against observed atmospheric water fields. Checking the coherence of the modeled and observed TCWV fields, an objective measure of the agreement between the model and observations is provided over almost the whole simulation domain, at horizontal scales of the order of hundreds of kilometers. Diagnostics of the forecast precipitation are performed here at smaller scales, around and below 100 km.

a. TCWV comparison

The method presented by Hoffman et al. (1995), which provides a quantitative measure of the pattern matching between two gridded fields, is quite similar to CRA (see below) and is based on rmse minimization. The main difference is that the entire forecast field is shifted in latitude and longitude, while searching for an rmse minimum. This method allows for decomposition of the forecast error into three components: displacement, amplitude, and residual error. Because distortion error is not considered (see Hoffman et al. 1995), the residual error can be interpreted as a pattern error. The TCWV rmse between the model forecast and the SSM/I analysis was computed for a set of small displacements, looking for the displacement associated with the lowest rmse.

In the current setting, the TCWV BOLAM forecasts were shifted (w.r.t. the SSM/I TCWV) from −2.7° to 2.7° in both latitude and longitude. To achieve a reasonable pattern match between the BOLAM TCWV and that observed, a quality control test is performed by verifying that a minimum correlation is exceeded for each displacement (or shift) in the x and y directions. This value, defined as the minimum correlation statistically different from zero at the 95% confidence level, depends on the effective number of independent comparing samples, which is a function of the number of grid points where the analysis is performed, and the autocorrelation of both the observed and forecast fields (Xie and Arkin 1995). Thus, the F test (Panofsky and Brier 1958) is chosen for assessing whether the shifts in the x and y directions are statistically significant.

b. Precipitation comparison using the CRA analysis

The object-oriented CRA analysis is based on pattern matching of two contiguous areas defined as the observed and forecast precipitation areas (or entities) delimited by a predetermined isohyet: the CRA rain-rate contour. For our case study, because the event’s magnitude (although intense for Cyprus) did not reach particularly high levels, it was decided to consider two rain-rate contours, one set to 0.5 mm (24 h)−1 and the other one set to 5.0 mm (24 h)−1.

The pattern matching is obtained by translating in the x and y directions the forecast rainfall features over the observed ones, until a best-fit criterion is satisfied. Two criteria were used for quantifying the spatial forecast error: the minimization of the mean square error (MSE) and the maximization of the correlation (COR). In general, over a large CRA domain the two criteria produce the same results (Ebert and McBride 2000), whereas over too small a domain, such as Cyprus, the results may be different (Tartaglione et al. 2005). Moreover, the use of the MSE criterion in a limited spatial domain may lead to suspicious results, because the minimization may be obtained by shifting the forecast field (and/or the forecast maxima) out of the domain (Grams et al. 2006).

The analysis was applied considering three shifting values (sv): 5, 9, and 13; that is, the forecast field was shifted from −sv × 0.09° to sv × 0.09°, both in latitude and in longitude. It should be noted that, over a relatively small verification area, larger shift values might produce incorrect pattern matching, because of the possible presence of many precipitation patterns.

Ebert and McBride (2000, section 2.3) proposed a decomposition of the displacement forecast error into three components: displacement, pattern, and volume errors. Such decomposition may not be suitable when the correlation criterion is applied, because the best pattern match may be associated with incorrectly negative displacement error. Thus, the alternative decomposition proposed by Grams et al. (2006, section 2e) is applied when the correlation maximization is used.

Also, for this method, a quality check of the shift in the x and y directions (obtained by using both MSE and COR as pattern-matching criterion) was performed using the aforementioned F test at the 95% confidence level. Only statistically significant displacements were considered for the study.

5. Multisensor and modeling comparison

a. TCWV comparison results

First, the TCWV verification results were analyzed. For the Hoffman verification, the 0.27° BOLAM TCWV forecasts were verified against TCWV retrieved from the various SSM/I observations. Available overpasses over the eastern Mediterranean Sea from 0600 UTC 5 March to 0600 UTC 6 March 2003 are listed in Table 1, together with the associated forecasts, within a time window of 0.5 h. Only overpasses with valid retrievals (with no corruption of brightness temperature fields, no rain contamination, and only over sea) are considered. Observations were mainly available at about dawn and/or dusk times (see Table 1).

The TCWV comparison is graphically presented in Fig. 10, while Fig. 11 shows the rmse and correlation spatial variation plots. A quantitative overview is provided in Table 2. As a reference, positive (negative) values in x and y directions indicate that forecasts should be displaced eastward (westward) and northward (southward) w.r.t. the original position in order to achieve a best fit with observations. The results show a marked difference between the situation on the morning of 5 March and the rest of the time period considered. From 0600 to 0800 UTC 5 March the comparison evidences a systematic reduction in the rmse, and the correlation of the shifted fields is above 0.85 (see Table 2), indicating that the original 0.27° BOLAM forecasts are shifted slightly southwestward with respect to the SSM/I-retrieved fields. Specifically, the shifts seem to be mainly due to a small forecast displacement error in the BOLAM model, because the comparison between TCWV retrieved by SSM/I and that estimated by the ECMWF analysis did not show such a significant shift (not shown). This difference might be explained by considering the smoothness of the ECMWF analysis (and, consequently, the smoothness of the SSM/I TCWV field averaged over a 0.5° grid) with respect to the comparison of BOLAM TCWV forecasts on a finer grid. These results indicate good agreement between modeled and observed TCWV fields in this time range; thus, the large-scale dynamical representation of the cyclonic system can be considered to be satisfactory on the morning of 5 March. However, the results obtained for the remaining overpasses show a rather different picture. All overpasses but one (at 0600 UTC 6 March) indicate a relatively large westward shift. It is interesting to note that now the observed TCWV fields are “flatter” than in the morning of 5 March, as reflected by the generally lower correlation values (see Figs. 10 and 11 and Table 2).

It is noticeable from Figs. 10j and 10p that during the evening of 5 March and on the following morning BOLAM predicts a much drier region at the western rim of the considered area compared with the observations. The forecast TCWV fields are eastward compared with the observations; thus, a westward shift is found. This can be interpreted as a sign of a faster forecast of eastward movement as time proceeds. This is true when considering almost all of the extent in longitude. As shown in Table 3, the biases are larger at 0500 UTC 6 March, when observations on the westernmost edge are available. On the other hand, the bias is quite small when considering the following hour, when only observations around Cyprus are available. However, the BOLAM model seems to fail in forecasting TCWV features similar to the observations (see Fig. 10), despite the relative intensity of the event under study. The error decomposition in terms of amplitude, displacement, and pattern errors (Table 3) indicates that the pattern error together with the displacement error explain a large amount of the total error, while the volume error (bias after displacement) is quite small. For the last instant considered (0600 UTC 6 March), the pattern error is indeed the most relevant error component.

b. Rainfall comparison results

The precipitation verification is performed at the BOLAM’s finer scale (about 10 km) and over a domain limited to Cyprus and its surrounding waters using the CRA analysis. Precipitation fields are accumulated over a 24-h time period. As described before, four precipitation analyses were used when verifying the BOLAM forecast (Figs. 8 and 9). Differences among these fields are due to the application of the range-adjustment technique based on TRMM PR data (POGRD versus PRGRD; RMOGRD versus RMRGRD) and due to the different merging method used (POGRD versus RMOGRD; PRGRD versus RMRGRD). The latter difference can be quantified in a root-mean-square difference of about 9 mm (24 h)−1 for POGRD versus RMOGRD and 15 mm (24 h)−1 for PRGRD versus RMRGRD, and a correlation of about 0.6 for both comparisons. By visually comparing each of the four precipitation analyses (Figs. 8 and 9) with the BOLAM forecast (Fig. 6), it is evident that the precipitation forecast needs, in any case, to be shifted eastward and/or northeastward to obtain a better spatial match with the observations. This is particularly true over Cyprus, where the rain area to the southwest and the no-rain area to the northwest are better spatially matched by shifting the precipitation forecasts. The visual inspection is not so helpful over the sea, because the observed peaks, probably due to convective activity (confirmed by lightning observations; not shown), were not correctly forecast by the 0.09° hydrostatic BOLAM.

The CRA is then applied to quantify and qualify (in terms of error sources) such perceived displacement forecast errors. Moreover, the results can be used to assess the reliability improvement given by using radar data in the observational gridded field.

In a previous study by Tartaglione et al. (2005), problems concerning the CRA application to this event were discussed. In that work, rain gauge observations alone were used for comparison with BOLAM. Results showed that the verification domain (corresponding to the grid points where the Barnes rain gauge analysis was available) was too small for an automated unsupervised application of the CRA technique (E. Ebert 2005, personal communication), suggesting that some care has to be taken when evaluating the CRA results over small areas in order to distinguish reliable results from suspicious, unphysical ones. Over a small area, there is in fact the possibility of not completely enclosing the observed entity of interest to be compared with the modeled entity. As summarized in Table 4, the results of the comparison between the Barnes rain gauge–based observational analysis (GP) and the forecast precipitation using MSE (hereafter GP MSE) are strongly dependent on the maximum allowed shift (GP MSE 5/9 steps versus GP MSE 13 steps). These results are also very different from the results obtained by using correlation matching (hereafter GP CORR), which does not change as a function of the shifting value and which seems more physically reliable (see discussion on visual verification above). Volume and pattern errors play a major role in this case.

Let us now focus on the CRA results obtained by combining the ground-based radar and the rain gauge gridded analysis using both Eq. (4) and the RainMusic tool. A CRA rain-rate contour equal to 0.5 mm (24 h)−1 is first used to isolate the forecast patterns and the observed precipitation. Results obtained by using both POGRD and RMOGRD (observational composite based on the nonadjusted radar data) become more and more different when changing the CRA criterion and the maximum shifting value sv. In addition, when using the MSE criterion, the results seem to be less physically reliable than the ones achieved by the correlation criterion, because the modeled rainfall is shifted too far w.r.t. the observations. This is corroborated by the visual verification performed by comparing the gauge-nonadjusted radar composite POGRD and RMOGRD (Figs. 8a and 9a) against both the original nonshifted (Fig. 6) and the shifted forecast fields obtained by using the MSE criterion (Figs. 12a and 12c) and by using the correlation criterion (Figs. 13a and 13c). When using PRGRD and RMRGRD (Figs. 8b and 9b), reliable pattern matching is obtained (except for the RMRGRD CORR 13 case; see Table 4), indicating that the forecast field should be moved eastward and/or northeastward (0.27°N, 0°–0.09°E for PRGRD and 0.36°N, 0° for RMRGRD) to obtain the best match with the observations (Figs. 12b, 12d, 13b and 13d). The pattern error provides the largest contribution to the forecast error (between 80% and 90%). These results, which are not so sensitive to the chosen criterion (Table 4), are in line with the GP CORR displacement, apart for the magnitude of the forecast error components.

As previously reported, some of the results are considered to be suspicious (see values in boldface in Table 4), because they tend to shift the forecast field, and in particular the higher rain values present in the forecast, out of the verification domain. Such a statement is confirmed by checking the bias error (see Table 4). These suspicious results are more evident when using a precipitation threshold equal to 5.0 mm (24 h)−1, making the CRA analysis difficult to interpret (see Table 5). Two examples can help to clear up the situation: the CRA 2D shift analysis, maximizing correlation, between the BOLAM forecast field and both PRGRD (Fig. 14a) and RMRGRD (Fig. 14b). By plotting the maximum correlation coefficients found during CRA over these analyses, it is evident that the “suspicious” final shift in the top-left corner is a localized isolated maximum, whereas the relative second maximum in the middle-right part of the panels (0°, 0.27°E for PRGRD and 0°, 0.36°E for RMRGRD) is a more robust result. A more complex matching procedure, based on the correlation maximization conditioned to the MSE minimization, is able to automatically select the secondary maximum.

Finally, it is worth noting that when ground-based radar data are included in the precipitation analysis, the correlation is lower than the precipitation analysis based only on rain gauges. The MSE is, instead, always higher, reaching the maximum values for RMRGRD (Tables 4 and 5). A lower correlation coefficient may be explained by the fact that radar data introduce into the observational analysis a spatial structure that is present in neither the forecast field nor the gauge analysis. An increase of the number of outlying data points may also contribute to this situation. The increase of the MSE, with respect to GP data, shows that forecast precipitation, in particular over the sea, is overall overestimated (see original bias values in Tables 4 and 5).

c. Comparison discussion

Results of the TCWV verification for the morning of 5 March and the CRA precipitation verification indicate a consistent eastward and/or northeastward shift (even if associated with different displacement values). Such agreement is not present when considering the TCWV verification for the other SSM/I overpasses. TCWV verification is focused on relatively large-scale fields, while the precipitation verification is performed on a small-scale basis. Moreover, satellite humidity fields have generally smoother gradients than do the rainfall patterns, so the precipitation verification is more demanding in terms of the spatial variability.

Space–time moisture localization gives at least a constraint on the precipitation occurrence, although the link between TCWV and rainfall patterns is not straightforward, because many physical processes act to mediate this relationship. For instance, when orographic precipitation occurs, as in our case (see section 2), the orography pattern is much more effective than the moisture distribution in determining the rainfall location. At any rate, spatial- and temporal-scale nonhomogeneities must be taken into account to explain the verification results’ discrepancies. SSM/I data refer to almost instantaneous values, whereas precipitation is accumulated over 24 h. Hence, its verification is less demanding in terms of the temporal variability.

Focusing our attention on the northeastward rainfall displacement, this can be interpreted as a delay in the passage of the forecast moist air band over Cyprus, as shown by the TCWV verification during the early phases of the event (morning of 5 March). This can be tentatively checked by including in the event analysis the temporal evolution of the rain rate over two gauge stations3 (Prodromos and Achna; see Fig. 2a) and the nearest grid points (Fig. 15). A few-hours temporal lag between the observed and modeled rain is actually present in the comparison: the modeled rain precipitated later with respect to the observed result. Note that this lag is short enough to be fully masked by the 24-h accumulation of precipitation.

By comparing the BOLAM pressure and wind fields with the ECMWF analyses (not shown), good agreement is found for the cyclone size and position, but notable differences are present in the evolution of the inner structures as testified to by the TCWV comparison. For instance, the closed cyclonic wind path visible west of Cyprus in Fig. 5c is absent in the analysis (not shown). Such a difference might be explained by considering that the ECMWF analysis is based, over the Mediterranean Sea, on sparse data and that high-resolution models (like BOLAM) have instead the tendency to enhance the mesoscale variability (Casaioli et al. 2006). Nevertheless, the above feature is more likely connected to the sharp pressure minimum visible at 850 hPa (Fig. 5a), which, in the BOLAM simulation, moves eastward faster than does the entire cyclone, as is also suggested by the TCWV verification in the later stage of the event (see section 5a).

6. Conclusions

In this paper, a multisensor comparison methodology based on diagnostic techniques is proposed in order to evaluate the performance of a LAM water fields’ forecasts. A precipitation event occurred on 5 March 2003 over the eastern Mediterranean and was used as an exemplificative case study. In particular, the forecast quality has been assessed by estimating the spatial displacement of the water forecast fields w.r.t. the available observations. Several different instruments (ground and satellite based) have been used in order to estimate the rainfall over and around the island of Cyprus, and the water vapor over the eastern Mediterranean Sea. The approach was motivated by the fact that verification is often performed with knowledge of the precipitation over limited areas (mainly land areas), while satellite observations have a larger spatial coverage (which is an advantage over extensive water areas) and can be used to recalibrate ground-based observations. Moreover, satellite-based instruments are major contributors to our knowledge of other atmospheric variables, such as total column water vapor. However, the available SSM/I observations have limited temporal sampling. This limit can be overcome by using water vapor information from geostationary platforms like Meteosat second-generation satellites, with 6.2- and 7.3-μm channels [or the equivalent Geostationary Operational Environmental Satellite (GOES) platforms] for verification purposes.

As shown in the previous section, water fields can be shifted in directions that are not the same as those of the baric fields and this can make the evaluation of the shift errors of models particularly problematic. A better temporal sampling could help to better pin down these discrepancies.

As to the precipitation comparison, when observational analysis is limited within a small domain, the results of the CRA analysis may strongly depend on the maximum shifting value allowed. Merging ground-based radar data with the observational analysis extends the area covered by observations and stabilizes the results, in the sense that the best CRA matching result is not influenced by the choice of the maximum shift (sv). Adding further observations, the stability of the CRA verification can be improved. This is not surprising because the observational coverage extension helps to enclose completely the observed entity to be compared with the forecast entity. Nevertheless, suspicious unphysical results (see in particular Table 5) may be obtained if the CRA method is applied in an automatic, unsupervised way. However, such suspicious displacements may be detected by performing quality check tests, and in some case corrected (see Fig. 14) by applying a double criterion based on both correlation and MSE. Future studies are envisaged in this direction, in particular on the application of the CRA method to a small domain. For instance, the implementation of such a conditioned criterion in an automated verification is currently under investigation. Some of the authors are also studying the possibility of using a skill score measure for pattern matching, and eventually for analyzing the relationship between the CRA 2D shift region in which the correlation (MSE) is greater (less) than the original correlation (MSE) value and the CRA results obtained using the correlation (MSE) criterion.

This study also investigated the pros and cons of the usage of state-of-the-art object-oriented methods, such as CRA, for the comparison of modeled rainfall fields with multisensor observations. For instance, it is evident from results both presented here and in Tartaglione et al. (2005) and Grams et al. (2006) that a supervised use of the CRA analysis is strongly recommended in order to distinguish reliable results from the unreliable, suspicious ones. Moreover, Ebert and McBride (2000) suggest that when these methods are used on a series of case studies (or equivalently a long time series), it is also possible to diagnose systematic spatial forecast errors, an operation that is currently performed only by a few meteorological services.

For the selected event, results can dramatically change when the correlation maximization is employed as a pattern-matching criterion, instead of the MSE minimization (see, e.g., the GP, PORGD, and RMORGRD results in Table 4). Moreover, some of the results, obtained using both the MSE and correlation criteria, seem to not be physically reliable. Unfortunately, it is not possible to give an a priori estimate of the impact of the observational errors on the CRA results, because it is not straightforward to determine, a priori, the observational field that gives the best estimate of the “ground truth” precipitation field. The uncertainty of the precipitation analysis can be very high. On one hand, rain gauges give a point measurement that needs to be spatially interpolated (e.g., using the Barnes analysis or block-kriging) in order to have a gridded precipitation field (but limited to land areas). On the other hand, radar gives a spatially retrieved measure of precipitation (in this case also over conterminous sea areas) that needs to be corrected because of its viewing geometry, for instance by using the TRMM PR range adjustment, in order to compensate for the decreased spatial resolution with range.

Bearing in mind these limitations, it was decided to consider the PRGRD and the RMRGRD gridded fields as the most reliable precipitation fields. In fact, the CRA results obtained by using MSE and CORR, both for thresholds of 0.5 and 5.0 mm (24 h)−1 (after considering the correction of the suspicious results), are in general similar to each other, with a difference of one or two grid points for the displacement in the latitudinal and longitudinal directions.

The northeastward shift has also been obtained as a result of the Hoffman TCWV verification, performed over the eastern Mediterranean Sea, for the first three SSM/I overpasses. This seems to indicate for the present case study a common shift of the BOLAM forecast water fields during the first 12 h. The discrepancy observed in the following overpasses seems to indicate that after an initial coherence between TCWV and the precipitation fields, the quality of the forecast seems to decrease in the next 12 h, as evidenced by the TCWV observations and suggested from the hourly precipitation comparison shown in Fig. 15. Thus, a relatively good precipitation forecast assessed on a 24-h basis could be the result of a forecast still affected by displacement, pattern, and volume errors.

Acknowledgments

This work was developed within the framework of UE project VOLTAIRE (Fifth Framework Programme; EVK2-2002-CT-00155). The authors are grateful to ECMWF for the data for the BOLAM initial and boundary conditions, to NOAA for the SSM/I data by means of the CLASS System, and to NASA for the TRMM data used to adjust the ground-based radar data. We thank Dr. Cinzia Mazzetti (PROGEA S.r.l., Bologna, Italy) for the RainMusic Software and her support in applying the method. Our thanks go also to all VOLTAIRE partners that were involved in the overall discussions about comparisons between the radar and model fields. We also wish to thank Prof. Giovanni Perona (Politecnico di Torino, Turin, Italy), Ing. Giuseppina Monacelli [Agenzia per la Protezione dell’Ambiente e per i Servizi Tecnici (APAT), Rome, Italy], and Prof. Mariarosaria Padula (University of Ferrara) for their support. Finally, Dr. Elisabeth Ebert (Bureau of Meteorology Research Centre, Melbourne, Victoria, Australia), Dr. Andrea Buzzi (ISAC–CNR, Bologna, Italy), Dr. Alexandre Lanciani (APAT), and three anonymous reviewers have made several helpful comments and suggestions to improve the paper.

REFERENCES

  • Accadia, C., , Casaioli M. , , Mariani S. , , Lavagnini A. , , and Speranza A. , 2003a: A diagnostic study of a Mediterranean cyclone using a regional model and SSM/I observations. Fifth Plinius Conf. on Mediterranean Storms, Ajaccio, France, European Geosciences Union, 159–164.

  • Accadia, C., , Mariani S. , , Casaioli M. , , Lavagnini A. , , and Speranza A. , 2003b: Sensitivity of precipitation forecast skill scores to bilinear interpolation and a simple nearest-neighbor average method on high-resolution verification grids. Wea. Forecasting, 18 , 918932.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Accadia, C., , Mariani S. , , Casaioli M. , , Lavagnini A. , , and Speranza A. , 2005: Verification of precipitation forecasts from two limited-area models over Italy and comparison with ECMWF forecasts using a resampling technique. Wea. Forecasting, 20 , 276300.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Alishouse, J. C., , Snyder S. , , Vongsathorn J. , , and Ferrero R. R. , 1990: Determination of oceanic total precipitable water from the SSM/I. IEEE Trans. Geosci. Remote Sens., 28 , 811816.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Amitai, E., , Nystuen J. A. , , Liao L. , , Meneghini R. , , and Morin E. , 2004: Uniting space, ground, and underwater measurement for improved estimates of rain rate. IEEE Trans. Geosci. Remote Sens., 1 , 811816.

    • Search Google Scholar
    • Export Citation
  • Arkin, P. A., , and Meisner B. , 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–1984. Mon. Wea. Rev., 115 , 5174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Baldwin, M. E., cited. 2000: QPF verification system documentation. [Available online at http://wwwt.emc.ncep.noaa.gov/mmb/ylin/pcpverif/scores/docs/mbdoc/pptmethod.html.].

  • Barnes, S. L., 1964: A technique for maximizing details in numerical weather map analysis. J. Appl. Meteor., 3 , 396409.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Barnes, S. L., 1973: Mesoscale objective analysis using weighted time-series observations. NOAA Tech. Memo. ERL NSSL-62, National Severe Storms Laboratory, Norman, OK, 60 pp. [NTIS COM-73-10781.].

  • Barrett, E. C., 1970: The estimation of monthly rainfall from satellite data. Mon. Wea. Rev., 98 , 322327.

  • Buzzi, A., , and Foschini L. , 2000: Mesoscale meteorological features associated with heavy precipitation in the southern Alpine region. Meteor. Atmos. Phys., 72 , 131146.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buzzi, A., , Fantini M. , , Malguzzi P. , , and Nerozzi F. , 1994: Validation of a limited area model in cases of Mediterranean cyclogenesis: Surface fields and precipitation scores. Meteor. Atmos. Phys., 53 , 5367.

    • Search Google Scholar
    • Export Citation
  • Casaioli, M., , Mariani S. , , Accadia C. , , Tartaglione N. , , Speranza A. , , Lavagnini A. , , and Bolliger M. , 2006: Unsatisfying forecast of a Mediterranean cyclone: A verification study employing state-of-the-art techniques. Adv. Geosci., 7 , 379386.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cressie, N. A. C., 1993: Statistics for Spatial Data. John Wiley and Sons, 900 pp.

  • Doelling, I. G., , Joss J. , , and Riedl J. , 1998: Systematic variations of Z–R relationships from drop size distributions measured in northern Germany during seven years. Atmos. Res., 48 , 635649.

    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., 2004: Monitoring the quality of operational and semi-operational satellite precipitation estimates: The IPWG validation/intercomparison study. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division and EUMETSAT. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Ebert1.pdf.].

  • Ebert, E. E., , and McBride J. L. , 2000: Verification of precipitation in weather systems: Determination of systematic errors. J. Hydrol., 239 , 179202.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., , Damrath U. , , Wergen W. , , and Baldwin M. E. , 2003: The WGNE assessment of short-term quantitative precipitation forecasts. Bull. Amer. Meteor. Soc., 84 , 481492.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebert, E. E., , Janowiak J. E. , , and Kidd C. , 2007: Comparison of near-real-time precipitation estimates from satellite observations and numerical models. Bull. Amer. Meteor. Soc., 88 , 4764.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraro, R. R., 1997: Special Sensor Microwave Imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 , 1671516735.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraro, R. R., , and Marks G. F. , 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12 , 755770.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fisher, R. A., 1925: Statistical Methods for Research Workers. Oliver and Boyd, 239 pp.

  • Gabella, M., , Michaelides S. , , and Perona G. , 2005: Preliminary comparison of TRMM and ground-based precipitation radar for a European test site. Int. J. Remote Sens., 26 , 9971006.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gabella, M., , Joss J. , , Perona G. , , and Michaelides S. , 2006: Range adjustment for ground-based radar, derived with the spaceborne TRMM Precipitation Radar. IEEE Trans. Geosci. Remote Sens., 44 , 126133.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gottschalck, J., , Meng J. , , Rodell M. , , and Houser P. , 2005: Analysis of multiple precipitation products and preliminary assessment of their impact on global land data assimilation system land surface states. J. Hydrometeor., 6 , 573598.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grams, J. S., , Gallus W. A. , , Koch S. E. , , Wharton L. S. , , Loughe A. , , and Ebert E. E. , 2006: The use of a modified Ebert–McBride technique to evaluate mesoscale model QPF as a function of convective system morphology during IHOP 2002. Wea. Forecasting, 21 , 288306.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hoffman, R. N., , Liu Z. , , Louis J-F. , , and Grassotti C. , 1995: Distortion representation of forecast errors. Mon. Wea. Rev., 123 , 27582770.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hollinger, J. P., 1989: DMSP Special Sensor Microwave/Imager Calibration/Validation. Final Rep., Vol. I., Space Sensing Branch, Naval Research Laboratory, Washington, DC, 186 pp. [Available from Naval Research Laboratory, Washington, DC 20375.].

  • Houze R. A. Jr., , , Brodzik S. , , Schumacher C. , , Yuter S. E. , , and Williams C. R. , 2004: Uncertainties in oceanic radar rain maps at Kwajalein and implications for satellite validation. J. Appl. Meteor., 43 , 11141132.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Iguchi, T., , Kozu T. , , Meneghini R. , , Awaka J. , , and Okamoto K. , 2000: Rain-profiling algorithm for the TRMM Precipitation Radar. J. Appl. Meteor., 39 , 20382052.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ikai, J., , and Nakamura K. , 2003: Comparison of rain rates over the ocean derived from TRMM Microwave Imager and Precipitation Radar. J. Atmos. Oceanic Technol., 20 , 17091726.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Imaoka, K., , and Spencer R. W. , 2000: Diurnal variation of precipitation over the tropical oceans observed by TRMM/TMI combined with SSM/I. J. Climate, 13 , 41494158.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Janowiak, J. E., , Xie P. , , Joyce R. J. , , Chen M. , , and Yarosh Y. , 2004: Validation of satellite-derived rainfall estimates and numerical model forecasts of precipitation over the United States. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division and EUMETSAT. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Janowiak.pdf.].

  • Kalman, R. E., 1960: A new approach to linear filtering and prediction problems. Trans. ASME J. Basic Eng., 82D , 3545.

  • Kidd, C., 2004: Validation of satellite rainfall estimates over the mid-latitudes. Proc. Second Int. Precipitation Working Group Workshop, Monterey, CA, U.S. Naval Research Laboratory, Marine Meteorology Division, and EUMETSAT, 10 pp. [Available online at http://www.isac.cnr.it/~ipwg/meetings/monterey/pdf/Kidd.pdf.].

  • Koch, S. E., , desJardins M. , , and Kocin P. J. , 1983: An interactive Barnes objective map analysis scheme for use with satellite and conventional data. J. Climate Appl. Meteor., 22 , 14871503.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kozu, T., and Coauthors, 2001: Development of precipitation radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. IEEE Trans. Geosci. Remote Sens., 39 , 102115.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kummerow, C., , Barnes W. , , Kozu T. , , Shiue J. , , and Simpson J. , 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15 , 809817.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lagouvardos, K., , Kotroni V. , , Koussis A. , , Feidas C. , , Buzzi A. , , and Malguzzi P. , 2003: The meteorological model BOLAM at the National Observatory of Athens: Assessment of two-year operational use. J. Appl. Meteor., 42 , 16671678.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Malguzzi, P., , and Tartaglione N. , 1999: An economical second order advection scheme for numerical weather prediction. Quart. J. Roy. Meteor. Soc., 125 , 22912303.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mariani, S., and Coauthors, 2005: A limited area model intercomparison on the “Montserrat-2000” flash-flood event using statistical and deterministic methods. Nat. Hazards Earth Syst. Sci., 5 , 565581.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , and Palmer W. M. , 1948: The distribution of raindrops with size. J. Meteor., 5 , 165166.

  • Marzano, F. S., , Palmacci M. , , Cimini D. , , Giuliani G. , , and Turk F. J. , 2004: Multivariate statistical integration of satellite infrared and microwave radiometric measurements for rainfall retrieval at the geostationary scale. IEEE Trans. Geosci. Remote Sens., 42 , 10181032.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mazzetti, C., 2004: Multi-sensor rainfall measurements for flood forecasting. Ph.D. Dissertation, University of Bologna, 238 pp. [Available from Dept. of Earth Sciences and Environment, University of Bologna, Piazza di Porta San Donato 1, 40126 Bologna, Italy.].

  • Mazzetti, C., 2006: RAINMUSIC software manual. PROGEA S.r.l., Bologna, Italy, 69 pp. [Available from PROGEA S.r.l., Via Galliera 93, 40121 Bologna, Italy.].

  • Mesinger, F., 1996: Improvements in quantitative precipitation forecasting with the Eta regional model at the National Centers for Environmental Prediction: The 48-km upgrade. Bull. Amer. Meteor. Soc., 77 , 26372649.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michaelides, S. C., , Gabella M. , , Constantinides P. , , and Perona G. , 2004: Comparison between TRMM data and in situ measurement in the island of Cyprus. Proc. Second Voltaire Workshop, Ljubljana, Slovenia, VOLTAIRE Project, 47–52.

  • Panofsky, H. A., , and Brier G. W. , 1958: Some Applications of Statistics to Meteorology. The Pennsylvania State University, 224 pp.

  • Porcù, F., , Prodi F. , , Pinori S. , , Dietrich S. , , and Mugnai A. , 2003: On the reliability of current MW-IR techniques in quantitative precipitation measurements over Mediterranean basin. Proc. Fourth Plinius Conf. on Mediterranean Storms, Mallorca, Spain, European Geophysical Society, 161–170.

  • Schumacher, C., , and Houze R. A. Jr., 2000: Comparison of radar data from the TRMM satellite and Kwajalein oceanic validation site. J. Appl. Meteor., 39 , 21512164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sohn, B., , and Smith E. , 2003: Explaining sources of discrepancy in SSM/I water vapor algorithms. J. Climate, 16 , 32293255.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tartaglione, N., , Mariani S. , , Accadia C. , , Speranza A. , , and Casaioli M. , 2005: Comparison of rain gauge observations with modeled precipitation over Cyprus using contiguous rain area analysis. Atmos. Chem. Phys., 5 , 21472154.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Todini, E., 2001: Bayesian conditioning of radar to rain-gauges. Hydrol. Earth System Sci., 5 , 225232.

  • VOLTAIRE, 2006: Validation of multisensor precipitation fields and numerical modeling in Mediterranean test sites. Final Rep., 135 pp. [Available from Dipartimento Tutela Acque Interne e Marine, APAT, Via Curtatone 3, I-00185 Rome, Italy.].

  • Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.

  • WWRP/WGNE Joint Working Group on Verification, cited. 2004: Recommendations for the verification and intercomparison of QPFs from operational NWP models. [Available online at http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/WGNE/QPF_verif_recomm.pdf.].

  • WWRP/WGNE Joint Working Group on Verification, cited. 2007: Forecast verification—Issue, methods and FAQ. [Available online at http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/verif_web_page.html.].

  • Wylie, D. D., 1979: An application of a geostationary satellites rain estimation technique to an extratropical area. J. Appl. Meteor., 18 , 16401648.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, P., , and Arkin P. A. , 1995: An intercomparison of gauge observations and satellite estimates of monthly precipitation. J. Appl. Meteor., 34 , 11431160.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Geographical map of the eastern Mediterranean Sea area.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 2.
Fig. 2.

(a) Geographical distribution over Cyprus of the 147 rain gauges stations (black circles), and location of the ground-based radar (white circle). The Achna and the Prodromos rain gauges and the Kykkos weather radar are indicated with arrows. Orography is indicated on a grayscale and the high massif visible in the figure is Troodos. (b) Localization of the two obscured sectors in the radar (white circle) data, which are caused by the presence of the Troodos high massif, in the southeast direction, and by the Tripylos peak in the northwest direction. The beam occultation at 0° elevation is emphasized using light gray. A darker grayscale is used for orography.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 3.
Fig. 3.

Mean sea level pressure (MSLP) analysis at (a) 0000 UTC 5 Mar and (b) 0000 UTC 6 Mar 2003 over Europe (source: The Met Office). Cyprus and the conterminous area are indicated with a dashed box.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 4.
Fig. 4.

Atmospheric forecast fields at 0600 UTC 5 Mar 2003: (a) 850-hPa and (b) 500-hPa geopotential height (m), (c) 10-m wind vectors, and (d) 2-m specific humidity (10−4 kg kg−1). The areas where specific humidity is larger than 9 × 10−3 kg kg−1 are indicated in gray.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 5.
Fig. 5.

As in Fig. 4 but at 1800 UTC.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 6.
Fig. 6.

Contours (mm) of precipitation modeled by BOLAM. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 7.
Fig. 7.

Contours (mm) of the rain gauge–based gridded analysis obtained using a two-pass Barnes scheme. Precipitation analysis is accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 8.
Fig. 8.

Contours (mm) of the radar–rain gauge composite using (a) OGRD and (b) RGRD. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 9.
Fig. 9.

As in Fig. 8 but for the radar–rain gauge composite obtained using the RainMusic software.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 10.
Fig. 10.

TCWV verification result over the eastern Mediterranean Sea from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003. (left) The 0.27° (HR) BOLAM forecasts, (center) the forecast fields shifted as result of the Hoffman object-oriented method, and (right) the SSM/I TCWV fields. Data are masked over the area where the SSM/I-retrieved TCWV is available: (a)–(c) 0600 UCT 5 Mar, (d)–(f) 0700 UTC 5 Mar, (g)–(i) 0800 UTC 5 Mar, (j)–(l) 1800 UTC 5 Mar, (m)–(o) 1900 UTC 5 Mar, (p)–(r) 0500 UTC 6 Mar, and (s)–(u) 0600 UTC 6 Mar.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 10.
Fig. 10.

(Continued)

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 11.
Fig. 11.

The 2D shift analysis, minimizing rmse, between SSM/I TCWV estimates and the BOLAM forecast from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003: (left) rmse and (right) the correlation. The best shift is plotted with a cross symbol with the displacement values in the longitudinal (E) and latitudinal (N) directions. Solid lines indicate statistically significant shifts, whereas dashed lines indicate statistically nonsignificant shifts: (a), (b) 0600 UCT 5 Mar; (c), (d) 0700 UTC 5 Mar; (e), (f) 0800 UTC 5 Mar; (g), (h) 1800 UTC 5 Mar; (i), (j) 1900 UTC 5 Mar; (k), (l) 0500 UTC 6 Mar; and (m), (n) 0600 UTC 6 Mar.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 11.
Fig. 11.

(Continued)

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 12.
Fig. 12.

Contours (mm) of the forecast precipitation shifted w.r.t. (a) POGRD, (b) PRGRD, (c) RMOGRD, and (d) RMRGRD using MSE as the CRA pattern-matching criterion and sv equal to 9. Precipitation is 24-h accumulated from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 13.
Fig. 13.

As in Fig. 12 but using the correlation as the CRA pattern-matching criterion.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 14.
Fig. 14.

The 2D shift analysis, maximizing correlation, of CRA between the 0.09° BOLAM precipitation forecast and the rain gauge–radar composite (a) PRGRD and (b) RMRGRD using an isohyet equal to 5.0 mm (24 h)−1. Maximum values found during CRA are indicated with a cross symbol. The grayscale, from lighter to darker, indicates the progressive order in which these values are found. The suspicious final shift (top-left corner) is indicated with a black square and the displacement values in the longitudinal (E) and latitudinal (N) directions. The more reliable shift is indicated instead with a black circle. Solid lines indicate statistically significant shifts, whereas dashed lines indicate statistically nonsignificant shifts.

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Fig. 15.
Fig. 15.

Comparison during the event between hourly precipitation observed (dashed line, with square symbols) at the (a) Prodromos and (b) Achna stations and the forecast at the nearest BOLAM grid point (solid line, with diamond symbols).

Citation: Weather and Forecasting 23, 4; 10.1175/2007WAF2007032.1

Table 1.

The SSM/I overpasses from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003, and the associated forecasts.

Table 1.
Table 2.

Comparison results of the SSM/I TCWV fields against both the original nonshifted BOLAM forecasts and the shifted (as a result of the Hoffman method) BOLAM forecasts in term of continuous verification statistics. The best shift reported is obtained by minimizing the rmse and considering the significance test (sig. test) constraint. For each comparison, the number of comparing grid points, the rmse value, its value w.r.t. the TCWV mean field (relative rmse), the Pearson correlation coefficient (corr.), and its confidence interval are indicated. The correlation confidence interval is calculated using the technique proposed by Fisher (1925).

Table 2.
Table 3.

TCWV bias for nonshifted fields, bias after the shift, and percentage error decomposition for the instants considered.

Table 3.
Table 4.

CRA verification for 24-h rainfall from 0600 UTC 5 Mar to 0600 UTC 6 Mar 2003, with a CRA rain-rate contour equal to 0.5 mm (24 h)−1. The observational type column (obs type) indicates the type of precipitation field used in the comparison. Results that change as a function of the selected shifting value (sv = 5, 9, or 13) are coherently indicated. MSE decomposition is applied according to the pattern-matching criterion used (see section 4b). For completeness, the CRA results obtained using the Barnes rain gauge analysis alone (GP; Tartaglione et al. 2005) have been reported as well. Suspicious displacements are indicated in boldface.

Table 4.
Table 5.

As in Table 4 but with a CRA rain-rate contour equal to 5.0 mm (24 h)−1 and only for an sv equal to 9.

Table 5.

1

The double-penalty effect arises when an event is simulated, but it is misplaced with respect to the original position. Thus, the result is that the forecast is penalized twice: once for missing the event in the correct position and once for producing a false alarm where the event is not observed.

2

Remapping is performed to transfer the forecast field from the native grid to the verification grid as follows. First, each verification grid box is subdivided into n × n subboxes (i.e., 5 × 5 subboxes). Then, to each subgrid point is assigned the value of the nearest native grid point. Finally, the average of these subgrid point values produces the remapped value of the verification grid point. Readers may refer to section 3b in Accadia et al. (2003a) for the primary details of the procedure.

3

Available data from most of the Cyprus rain gauge stations are accumulated on a 24-h basis. Cumulative daily precipitation amounts are indeed read daily at 0600 UTC by professional or trained part-time staff of the Meteorological Service of Cyprus. Only six rain gauge stations provide shorter-time accumulated rainfall values.

Save