• Ali, A., Lebel T. , and Amani A. , 2005: Rainfall estimation in the Sahel. Part I: Error function. J. Appl. Meteor., 44, 16911706.

  • Anagnostou, E. N., Maggioni V. , Nikolopoulos E. I. , Meskele T. , Hossain F. , and Papadopoulos A. , 2009: Benchmarking high-resolution global satellite rainfall products to radar and rain-gauge rainfall estimates. IEEE Trans. Geosci. Remote Sens., 48, 16671683.

    • Search Google Scholar
    • Export Citation
  • Conner, M. D., and Petty G. W. , 1998: Validation and inter-comparison of SSM/I rain-rate retrieval methods over the continental United States. J. Appl. Meteor., 37, 679700.

    • Search Google Scholar
    • Export Citation
  • Dinku, T., Ceccato P. , Grover-Kopec E. , Lemma M. , Connor S. J. , and Ropelewski C. F. , 2007: Validation of satellite rainfall products over East Africa’s complex topography. Int. J. Remote Sens., 28, 15031526.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Gruber, A. V., and Levizzani V. , 2006: Assessment of global precipitation: A project of the Global Energy and Water Cycle Experiment (GEWEX) Radiation Panel. GEWEX, World Climate Research Program, WMO Rep., 61 pp.

  • Haylock, M. R., Hofstra N. , Klein Tank A. M. G. , Klok E. J. , Jones P. D. , and New M. , 2008: A European daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006. J. Geophys. Res., 113, D20119, doi:10.1029/2008JD010201.

    • Search Google Scholar
    • Export Citation
  • Hossain, F., and Huffman G. J. , 2008: Investigating error metrics for satellite rainfall data at hydrologically relevant scales. J. Hydrometeor., 9, 563575.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, 520.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 3855.

    • Search Google Scholar
    • Export Citation
  • Hughes, D. A., 2006: Comparison of satellite rainfall data with observations from gauging station networks. J. Hydrol., 327, 399410.

  • Joyce, J. R., Janowiak E. J. , Arkin P. A. , and Xie P. , 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487503.

    • Search Google Scholar
    • Export Citation
  • Kidd, C., Kniveton R. D. , Todd C. M. , and Bellerby J. T. , 2003: Satellite rainfall estimation using combined passive microwave and infrared algorithms. J. Hydrometeor., 4, 10881104.

    • Search Google Scholar
    • Export Citation
  • Krajewski, W. F., and Smith J. A. , 2002: Radar hydrology: Rainfall estimation. Adv. Water Resour., 25, 13871394.

  • Krajewski, W. F., Ciach G. J. , McCollum J. R. , and Bacotiu C. , 2000: Initial validation of the Global Precipitation Climatology Project monthly rainfall over the United States. J. Appl. Meteor., 39, 10711086.

    • Search Google Scholar
    • Export Citation
  • Kubota, T., and Coauthors, 2007: The Global Satellite Mapping of Precipitation (GSMaP) project. Proc. Seventh GPM Int. Planning Workshop, Bellesale Kudan, Tokyo, Japan, Ken’ichi Okamoto (Osaka Prefecture University), 14 pp. [Available online at http://www.eorc.jaxa.jp/GPM/ws7/pdf/7thDEC2007/7_am07.pdf.]

  • McCollum, J. R., Krajewski W. F. , Ferraro R. R. , and Ba M. B. , 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 10651080.

    • Search Google Scholar
    • Export Citation
  • McPhee, J., and Margulis S. A. , 2005: Validation and error characterization of the GPCP-1DD precipitation product over the contiguous United States. J. Hydrometeor, 6, 441459.

    • Search Google Scholar
    • Export Citation
  • Sapiano, M. R. P., and Arkin P. A. , 2009: An intercomparison and validation of high-resolution satellite precipitation estimates with 3-hourly gauge data. J. Hydrometeor, 10, 149166.

    • Search Google Scholar
    • Export Citation
  • Scheel, M. L. M., Rohrer M. , Huggel C. , Santos Villar D. , Silvestre E. , and Huffman G. J. , 2010: Evaluation of TRMM Multi-satellite Precipitation Analysis (TMPA) performance in the Central Andes region and its dependency on spatial and temporal resolution. Hydrol. Earth Syst. Sci. Discuss., 7, 85458586.

    • Search Google Scholar
    • Export Citation
  • Sorooshian, S., Hsu K.-L. , Gao X. , Gupta H. V. , Imam B. , and Braithwaite D. , 2000: Evaluation of PERSIANN system satellite–based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 20352046.

    • Search Google Scholar
    • Export Citation
  • Su, F., Hong Y. , and Lettenmaier D. P. , 2008: Evaluation of TRMM Multisatellite Precipitation Analysis (TMPA) and its utility in hydrologic prediction in the La Plata basin. J. Hydrometeor., 9, 622640.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., Peters-Lidard C. D. , Choudhury B. J. , and Garcia M. , 2007: Multitemporal analysis of TRMM-based satellite precipitation products for land data assimilation applications. J. Hydrometeor., 8, 11651183.

    • Search Google Scholar
    • Export Citation
  • Turk, F. J., and Miller S. D. , 2005: Toward improved characterization of remotely sensed precipitation regimes with MODIS/AMSR-E blended data techniques. IEEE Trans. Geosci. Remote Sens., 43, 10591069.

    • Search Google Scholar
    • Export Citation
  • van Engelen, A., Klein Tank A. , van de Schrier G. , and Klok L. , 2008: Towards an operational system for assessing observed changes in climate extremes. European Climate Assessment & Dataset Rep., 68 pp.

  • Xie, P., and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 25392558.

    • Search Google Scholar
    • Export Citation
  • Yamamoto, J. K., 2000: An alternative measure of the reliability of ordinary kriging estimates. Math. Geol., 32, 489509, doi:10.1023/A:1007577916868.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Topography of the study area and the two selected regions located in central and northwestern Europe. Region 1 is a low-elevation area, whereas region 2 is mountainous. Distribution of rain gauges (blue dots) over the study area is also shown.

  • View in gallery

    Maps of (left) mean gauge rainfall (mm day−1) and (right) the corresponding gauge error standard deviation over the European domain determined for the 8-yr period (2003–10) in (top) cold and (bottom) warm seasons.

  • View in gallery

    Time correlation maps (top two rows) between each satellite product and the reference and (bottom row) between the two satellite products for the (left) cold and (right) warm seasons.

  • View in gallery

    Mean error (mm day−1) maps of (left) CMORPH and (right) 3B42 V6 in the (top) cold and (bottom) warm seasons.

  • View in gallery

    Random error standard deviation (mm day−1) maps of (left) CMORPH and (right) 3B42 V6 in the (top) cold and (bottom) warm seasons.

  • View in gallery

    Normalized mean error of (left) CMORPH and (right) 3B42 V6 evaluated (top) for different gauge rainfall thresholds and (bottom) for different gauge-elevation categories in the cold and warm season. The vertical bars represent the standard deviations. Sample sizes for each group are presented in boxes in different colors according to the season.

  • View in gallery

    As in Fig. 6, but for error standard deviation.

  • View in gallery

    (top) Normalized missed rainfall volume, (middle) normalized false alarm rainfall volume, and (bottom) their ratio for (left) CMORPH and (right) 3B42 V6 evaluated for different gauge-elevation categories in the cold and warm season. Vertical bars represent standard deviations. Sample sizes for each value are presented in boxes in different colors according to the season.

  • View in gallery

    Temporal variability of the normalized mean error (%) for 3-month rainfall accumulations of 3B42 V6 and CMORPH over regions (top) 1 and (bottom) 2.

  • View in gallery

    As in Fig. 9, but for CC (unitless).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 296 183 11
PDF Downloads 264 163 8

Evaluation of Global Satellite Rainfall Products over Continental Europe

View More View Less
  • 1 Department of Civil and Environmental Engineering, University of Connecticut, Storrs, Connecticut
© Get Permissions
Full access

Abstract

An extensive evaluation of two global-scale high-resolution satellite rainfall products is performed using 8 yr (2003–10) of reference rainfall data derived from a network of rain gauges over Europe. The comparisons are performed at a daily temporal scale and 0.25° spatial grid resolution. The satellite rainfall techniques investigated in this study are the Tropical Rainfall Measuring Mission (TRMM) 3B42 V6 (gauge-calibrated version) and the Climate Prediction Center morphing technique (CMORPH). The intercomparison and validation of these satellite products is performed both qualitatively and quantitatively. In the qualitative part of the analysis, error maps of various validation statistics are shown, whereas the quantitative analysis provides information about the performance of the satellite products relative to the rainfall magnitude or ground elevation. Moreover, a time series analysis of certain error statistics is used to depict the temporal variations of the accuracy of the two satellite techniques. The topographical and seasonal influences on the performance of the two satellite products over the European domain are also investigated. The error statistics presented herein indicate that both orography and seasonal variability affect the efficiency of the satellite rainfall retrieval techniques. Specifically, both satellite techniques underestimate rainfall over higher elevations, especially during the cold season, and their performance is subject to seasonal changes. A significant difference between the two satellite products is that TRMM 3B42 V6 generally overestimates rainfall, while CMORPH underestimates it. CMORPH’s mean error is shown to be of higher magnitude than that of 3B42 V6, while in terms of random error variance, CMORPH exhibits lower (higher) values than those of 3B42 V6 in the winter (summer) months.

Corresponding author address: Emmanouil Anagnostou, CEE, University of Connecticut, Environmental Engineering Program, 261 Glenbrook Road, Unit 2037, Storrs, CT 06269. E-mail: manos@engr.uconn.edu

Abstract

An extensive evaluation of two global-scale high-resolution satellite rainfall products is performed using 8 yr (2003–10) of reference rainfall data derived from a network of rain gauges over Europe. The comparisons are performed at a daily temporal scale and 0.25° spatial grid resolution. The satellite rainfall techniques investigated in this study are the Tropical Rainfall Measuring Mission (TRMM) 3B42 V6 (gauge-calibrated version) and the Climate Prediction Center morphing technique (CMORPH). The intercomparison and validation of these satellite products is performed both qualitatively and quantitatively. In the qualitative part of the analysis, error maps of various validation statistics are shown, whereas the quantitative analysis provides information about the performance of the satellite products relative to the rainfall magnitude or ground elevation. Moreover, a time series analysis of certain error statistics is used to depict the temporal variations of the accuracy of the two satellite techniques. The topographical and seasonal influences on the performance of the two satellite products over the European domain are also investigated. The error statistics presented herein indicate that both orography and seasonal variability affect the efficiency of the satellite rainfall retrieval techniques. Specifically, both satellite techniques underestimate rainfall over higher elevations, especially during the cold season, and their performance is subject to seasonal changes. A significant difference between the two satellite products is that TRMM 3B42 V6 generally overestimates rainfall, while CMORPH underestimates it. CMORPH’s mean error is shown to be of higher magnitude than that of 3B42 V6, while in terms of random error variance, CMORPH exhibits lower (higher) values than those of 3B42 V6 in the winter (summer) months.

Corresponding author address: Emmanouil Anagnostou, CEE, University of Connecticut, Environmental Engineering Program, 261 Glenbrook Road, Unit 2037, Storrs, CT 06269. E-mail: manos@engr.uconn.edu

1. Introduction

Measuring rainfall is of utmost importance for the quantification of the global water cycle and, consequently, for assessing climate variability. Quantifying rainfall variability and magnitude at high spatiotemporal scales would give further insight in assessing all those processes the global water cycle integrates (physical, chemical, and biological) to sustain ecosystems and influence climate. In terms of applications, accurate rainfall measurement is desirable as a means of predicting hydrological risks (floods and droughts) and for managing water uses (agriculture, energy, water supply, etc.)

Rainfall measurements are based on diverse sources including in situ meteorological stations, weather radar networks, estimates inferred from satellite observations, and outputs from numerical weather prediction models, all of which are characterized by both advantages and disadvantages (Sapiano and Arkin 2009; Xie and Arkin 1997). Arguably, rainfall observations from gauges and weather radars give the most accurate surface rainfall measurements, but are prone to certain ambiguities. On the one hand, rain gauges are point measurements and, thus, do not accurately represent the mean area rainfall. Moreover, rain gauges are primarily deployed over land, rendering the monitoring of rainfall over oceans practically impossible; there are, however, a few gauges located on islands and atolls. Another major issue associated with the use of rain gauges is the fact that they are often very sparse over important climatic regions, such as the tropics and mountains. Weather radar, on the other hand, provides a more spatially representative estimate, but is also subject to limitations. The conversion of the signal backscatter into rain rates is associated with significant variability due to variations in the rainfall drop size distribution, surface effects, mixed phase precipitation, and ground returns. Other issues include rain-path attenuation, beam blockage, beam filling, and beam overshoot effects (Gruber and Levizzani 2006; Krajewski and Smith 2002). Finally, radar and gauge monitoring systems are characterized as considerably demanding when it comes to system maintenance, and thus require significant financial investments, which in turn deters many developing countries from contributing to the extremely important task of observing and quantifying rainfall.

Measuring rainfall at a global scale is feasible only through the application of satellite remote sensing. Satellite-derived rainfall estimates provide spatial coverage over most of the earth’s surface (both over land and water). This capability addresses the need for near-global precipitation data, and has led to an ever-increasing number of satellite-based rainfall products that meet various needs (e.g., Kidd et al. 2003). Remote sensing of precipitation overland, however, has its shortcomings: satellite-based rainfall estimates rely on inferring rainfall from observing properties of cloud tops in visible (VIS) or infrared (IR) imagery, or from the scattering effects of raindrops or ice particles on microwave (MW) radiation (Sapiano and Arkin 2009). Both approaches (VIS–IR and MW) introduce errors and uncertainties. The VIS–IR sensors on board geostationary satellites are able to monitor the earth continually, providing us with quasi-global scans every 15 min (Sapiano and Arkin 2009); however, the relationship between cloud properties inferred from VIS or IR and surface precipitation is indirect and, therefore, their link is weak (Anagnostou et al. 2009; Sapiano and Arkin 2009; Joyce et al. 2004). In contrast to the VIS–IR sensors, MW instruments provide rainfall estimates with greater accuracy, since their observations are related to the hydrometeor content present within the atmospheric column. However, because of their low sampling frequency, they suffer from larger sampling errors when dealing with short-term rainfall estimates (e.g., Kidd et al. 2003).

At present, the products that are more capable of providing accurate and reliable precipitation estimates are the ones based on combinations of infrared and microwave observations (Ebert et al. 2007). Combining the high sampling rate of the geostationary satellites and the superior accuracy provided by several passive microwave (PMW) sensors has, therefore, been the subject of much work in recent years. As a result, a number of high-resolution rainfall products (resolutions at or below 0.25° and temporal resolution at or below 3 h) at a global scale are now available. Among these products is the Tropical Rainfall Measuring Mission (TRMM) 3B42RT and gauge-adjusted 3B42 (version 6) produced at the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC) from the TRMM Multisatellite Precipitation Analysis (TMPA; Huffman et al. 2007). The National Oceanic and Atmospheric Administration’s (NOAA) Climate Prediction Center morphing technique (CMORPH) is another global-scale satellite precipitation product (Joyce et al. 2004). Another example of such products is Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), which is derived solely from IR data on the basis of a neural network technique calibrated with coincident MW rainfall products (Sorooshian et al. 2000). Moreover, there are other precipitation products from several science teams worldwide, such as the Global Satellite Mapping of Precipitation (GSMaP) product (Kubota et al. 2007) and the rainfall product from the U.S. Naval Research Laboratory (Turk and Miller 2005).

Several recent studies have been conducted to evaluate the accuracy of various satellite rainfall products at different continental regimes of the earth. At present, the most extensive error studies focus on the continental United States (Hossain and Huffman 2008; McPhee and Margulis 2005; McCollum et al. 2002; Krajewski et al. 2000; Conner and Petty 1998). Error studies in other continental regions include the ones by Hughes (2006), who compared satellite-derived rainfall data with gauge data over four basins in South Africa; Su et al. (2008), who evaluated TRMM 3B42 V6 precipitation estimates for the La Plata basin in South America through comparison with gauged data; Ali et al. (2005), who evaluated satellite rainfall uncertainty over the Sahel region; and Dinku et al. (2007), who presented an extensive evaluation of 10 different satellite rainfall products using gauge data over East Africa’s complex topography. Moreover, Ebert et al. (2007) validated six satellite products over the United States, Australia, and northwestern Europe, and showed that satellite precipitation estimates agreed well with the gridded gauge data at monthly time scales, whereas the agreement was poor at daily time scales and particularly for high rain rates. Another study by Scheel et al. (2010) evaluated the ability of TMPA to estimate precipitation rates in the central Andes and the dependency of the estimate performance on changing spatial and temporal resolution. Finally, Tian et al. (2007) evaluated and compared TRMM 3B42 and CMORPH against ground-based rain gauge–only and gauge-corrected radar rainfall measurements.

The current study focuses on the evaluation of daily rainfall estimates over the European continent from two of the main global high-resolution satellite products (TRMM 3B42 and CMORPH). The evaluation is performed based on 8 yr (2003–10) of reference rainfall data derived from a European wide network of daily rain gauges used to depict variations and effects due to topography and climatic regimes. The study also includes two densely gauged regions (45 000 km2 each) for the time series analysis of the rainfall error statistics over a mountainous and a low-elevation terrain area. The reference dataset used in this study is associated with estimates of the gauge sampling uncertainty (Haylock et al. 2008), which is herein used qualitatively in validating the rain gauge dataset and supporting the discussion of the satellite rainfall error variability. This satellite rainfall error study should provide useful help to users who wish to apply satellite-derived precipitation estimates for hydrologic and climate studies in Europe.

In the next section we present the study area and describe the data, while in section 3 we describe the methodology used for the error analysis. In section 4 we present the results and in section 5 we summarize our findings and conclusions.

2. Study area and data

Continental Europe (below 50°N) is the study area. Europe has four dominant types of climate: maritime in the west, continental in the east and north, Mediterranean in the south, and a mountainous climate in the highlands. The study area, therefore, combines unique characteristics of complex terrain, coastal topography, and climatic variability; as a result, precipitation is not evenly distributed over its domain. The two densely gauged regions (6 × 12 grid cells each) represent low-elevation (region 1) and mountainous (region 2) terrain, selected for time series analysis of the satellite error. Figure 1 shows the study area, the rain gauge distribution, and the two densely gauged regions. Region 1 covers southwestern Germany and part of northeastern France (49.5°–48°N, 7°–10°E) while region 2 is located over the central–eastern part of Alps (47.5°–46°N, 9.5°–12.5°E).

Fig. 1.
Fig. 1.

Topography of the study area and the two selected regions located in central and northwestern Europe. Region 1 is a low-elevation area, whereas region 2 is mountainous. Distribution of rain gauges (blue dots) over the study area is also shown.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

The rain gauge data used in this study were downloaded from the European Climate Assessment & Dataset project (ECA&D; http://eca.knmi.nl), which started in 2003 as a follow up to the European Climate Assessment project (ECA), which started in 1998 and is now funded by the Royal Netherlands Meteorological Institute. The objective of ECA&D is to analyze the temperature and precipitation climate of World Meteorological Organization (WMO) region VI (Europe and the Middle East) (van Engelen et al. 2008). For this purpose, a sustainable system for data collection, archiving, quality control, and dissemination has been realized. The data are being gathered from 825 meteorological stations throughout Europe and the Mediterranean provided by the national meteorological services from over 40 countries (van Engelen et al. 2008). For the quality control of the gauge measurements, fixed procedures have been used in order to assess their quality and homogeneity. These daily gauge rainfall data were interpolated into a canonical grid with 0.25° spatial resolution covering a domain spanning from 35° to 50°N in longitude and 10°W to 30°E in latitude. The distribution of stations within the domain is inhomogeneous (as shown in Fig. 1), contributing to a spatially varying gauge-interpolation uncertainty. Areas with the densest network of rain gauges appear to be in northern Italy, along the Po valley, the southeastern part of the Alps, Switzerland, and parts of Germany. The number of stations within the two densely gauged regions is around 35 gauges, representing an average density of 1 gauge per 1285 km2. The gauge interpolation was performed using kriging based on an efficient method originally proposed by Yamamoto (2000). Kriging provides a measure of the expected mean at an interpolated point as well as the associated variance. Specifically, it interpolates to a point by calculating a weighted sum of neighboring observations, with the weights determined by the variogram model and the separation distances. This method was applied to each grid point for every day to arrive at the standard error for the daily anomaly. The standard error has units of mm. The uncertainty estimate was calculated from the magnitude interpolation, regardless of whether the occurrence model designated a wet or dry day. To calculate the final uncertainty at a grid square, the uncertainties from the monthly climatology and the daily anomaly in quadrature (i.e., the square root of the sum of the squares of the two uncertainties) were combined (Haylock et al. 2008).

The daily gridded standard error dataset was used in two ways for the purpose of this study. One such application was the estimation of the errors associated with the reference dataset, which was conducted utilizing the standard error estimates provided in the database. Moreover, the uncertainty estimates for the gauge data were used as a means of validating the reference dataset used in this study. Finally, the gauge standard error dataset was used in a qualitative way in the satellite product error estimation.

Two high-resolution global-scale satellite rainfall products are evaluated in this study. The first is TMPA, or 3B42, which is a gauge-adjusted (over land only) satellite product (Huffman et al. 2007). The product is the combination of two subproducts: the MW and the MW-calibrated IR. The final product has a fine spatial resolution (0.25°) and temporal resolution (3 hourly); it is available both as postanalysis (3B42 V6) [where the 3-hourly PMW–IR estimates are summed to monthly resolutions and adjusted using the monthly gauge data (Huffman et al. 1997)] and in real time (3B42 RT) (without the gauge correction). TMPA 3B42 covers all latitudes between 50°N and 50°S (with the exception of 3B42 RT—the coverage of which extends from 60°N to 60°S) and starts in 1998. The other global product evaluated in this study is CMORPH, in which precipitation estimates are derived from PMW-only satellite estimates, which are propagated by motion vectors derived from geostationary satellite IR data (Joyce et al. 2004). The original product is created on an 8-km grid at half-hourly time resolution. Three-hourly, 0.25° spatial-resolution products are also generated and are used in this study. CMORPH data are available between 60°N and 60°S starting in December 2002.

A grid-average gauge-elevation dataset is used in this paper to assess the topographic effect on the accuracy of satellite products. Moreover, a daily grid-average temperature dataset is used to differentiate rain from solid precipitation. For this purpose a temperature of 4°C is used as a threshold to secure the assessment of rainfall-only estimates, without the confounding effect of mixed phase precipitation.

3. Evaluation methodology

This paper focuses on the performance assessment of satellite rainfall products using the nominally more accurate grid-interpolated rain gauge rainfall estimates as the reference (ground truth). The study area is categorized into low (0–500 m), moderate (500–1000 m), high (1000–1500 m), and very high (>1500 m) 0.25° mean elevation areas, while all analyses have been performed on the basis of two distinct periods—cold season (November–April) and warm season (May–October)—to represent the seasonal and topographical effects on the satellite retrievals. All datasets used in this study (gauge data, TRMM, CMORPH, standard errors, and temperature) were scaled at the same spatial (0.25°) and temporal (daily) resolution.

The error analysis in this paper is separated into three distinct parts: 1) a qualitative approach for the evaluation of the two satellite algorithms over the entire study area, showing the various error statistics in the form of maps, so as to depict spatial variations and, consequently, topographical effects; 2) a more quantitative approach for the satellite technique-performance evaluation over the entire domain and only for those 0.25° areas (pixels in grid) that host at least one rain gauge; and 3) a time series analysis, only for those 0.25° areas (pixels in grid) that include at least one rain gauge, which depicts the variation of certain error statistics with time. The validation statistics selected for demonstration are correlation coefficient (CC), mean error (ME), normalized mean error (NME), random error standard deviation (RESD), normalized random error standard deviation (NRESD), normalized missed rainfall volume (NMRV), normalized false alarm satellite rainfall volume (NFASRV), and the ratio (R) of NMRV to NFASRV. The formulas that describe the aforementioned performance metrics are given below:
e1
e2
e3
e4
e5
e6
e7
where Sat represents the satellite product grid estimate (mm day−1), Gauge represents the corresponding rain gauge rainfall grid values (mm day−1), SD is the standard deviation, and n is the number of events that satisfy a specific condition, which may change according to the rainfall regime, elevation category, and the investigated satellite performance metric. Criteria that always apply regardless of the validation statistic are the following: 1) validation statistics are performed unconditionally (rain ≥ 0) for both satellite products and rain gauges, and 2) temperature must be greater than or equal to 4°C. All error statistics were calculated for sample sizes of at least 200 events. Areas with no available data appear white on the error maps.

By using the aforementioned criteria, we aim to provide a thorough and comprehensive assessment of the performances of the two satellite techniques, compared to ground rain gauge data for the greater European domain. The ME provides information about the differences between rainfall detected by the satellite products and the reference (rain gauges) and is used in the qualitative analysis. NME is the normalized ME and is used in both quantitative and time series analysis for the satellite technique-performance evaluation. The RESD is used to evaluate the variance of the estimation error; its normalized form (NRESD) is used in the quantitative analysis. The NMRV performance metric shows the reference (i.e., gauge) rainfall volume that the satellite retrieval missed throughout a specific period of time normalized by the total reference rainfall volume during that period; NFASRV represents the rainfall volume that was falsely detected by the satellite retrieval technique normalized by the total reference rainfall volume during the same period. The ratio NMRV/NFASRV provides useful information about the net balance of missed to false alarm rain fractions. Correlation coefficients are calculated between the reference data and each satellite product as well as between the two satellite products. Good performance of the satellite technique would imply a low ME accompanied by a low RESD, low NMRV, and low NFASRV. In the quantitative analysis the various error statistics are plotted against different gauge rainfall thresholds or elevation categories. This way, we attempt to show rainfall intensity- or elevation-dependency for the two different satellite retrieval techniques. The 8-yr period used for the evaluation in this paper provides sound statistics for both warm and cold seasons.

4. Discussion of results

a. Qualitative analysis

The seasonal maps of the mean reference rainfall (in mm day−1) over the 8-yr period (2003–10) are shown in Fig. 2 (left panels); this was calculated by first locating for every grid cell the time periods at which the rain gauges indicate precipitation at liquid form (T ≥ 4°C). Figure 2 (right panels) also shows the seasonal maps of the gauge error standard deviation (mm day−1), which was estimated at every pixel for the aforementioned time periods of the years 2003–10. As shown in the figure, rainfall is greater over the mountainous regions (Alps and Pyrenees) compared to low-elevation areas, and this difference becomes more significant during the warm season (the misleading low mean rainfall over the Alps is due to the unaccounted solid precipitation that prevails over the area in the cold season). Areas of high elevation seem to receive rainfall that is two- to threefold increased, relative to that over the low-terrain areas. Moreover, rainfall in the cold season exhibits higher values over the main mountain ranges, the western Iberian Peninsula, parts of France, and most of the remaining Mediterranean region. In the warm season, rainfall shows a dipole behavior, with higher values in the central and northern study area and extremely low values in the Mediterranean region. The gauge error standard deviation (SD) is shown to be higher in the cold season over Spain as well as over the central and northeastern part of the study domain, whereas in the warm season, high gauge error SD values are shown mostly over the southern Mediterranean. Gauge error SD relative to the mean gauge rainfall is generally lower over higher terrain and, therefore, the gauge contribution is generally higher over mountainous areas. The Pyrenees and Carpathian Mountains show a distinctly lower relative gauge error compared to low-elevation areas in close vicinity, which is more evident in the cold season. The Alps, however, show a more distinct pattern during the warm season, where the relative gauge error standard deviation is significantly lower compared to that of the surrounding areas. Furthermore, gauge uncertainty is higher over areas with low station density (cf. Fig. 1). Parts of central Europe, northwestern Spain, as well as the Po valley in northern Italy clearly show that densely gauged areas are characterized by low gauge error SDs. Arguably, lack of gauges leads to an unavoidable error in the interpolation method and, thus, certain grid cells appear to have high gauge error SDs. By and large, the gauge errors shown here cannot be considered as insignificant. However, this qualitative analysis characterizes the representativeness of the rain gauge data as a “reference” dataset for the satellite validation study.

Fig. 2.
Fig. 2.

Maps of (left) mean gauge rainfall (mm day−1) and (right) the corresponding gauge error standard deviation over the European domain determined for the 8-yr period (2003–10) in (top) cold and (bottom) warm seasons.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

Next, we present seasonal maps of the correlation coefficient between each of the satellite products and the reference, as well as between the two satellite products (Fig. 3). The correlation coefficient was estimated for every grid cell of the study domain, determining how close the rainfall estimates of one are to those of the other product in time; one single correlation value was determined for every grid cell. It is evident that there are no significant differences between the gauge–3B42 V6 and gauge–CMORPH correlations in the cold season, while in the warm season 3B42 V6 seems to agree better with the reference data relative to CMORPH. Furthermore, within each pair there seem to be notable seasonal effects, which are more profound in the case of 3B42 V6; the warm-season correlations are significantly higher than those of the cold season for 3B42 V6, whereas the warm-season-related increase of the correlation values in the case of CMORPH is mostly evident over the central region of the study domain. Tian et al. (2007), who evaluated TRMM 3B42 V6 and CMORPH against ground-based rain gauge-only and gauge-corrected Doppler radar measurements over the contiguous United States (CONUS) have also found that both satellite products correlate to ground-based measurements better in summer than in winter. For both satellite products, correlation coefficient values are moderate to high, ranging from 0.3 to 0.8, where the lowest values are exhibited over higher elevations (Alps and Pyrenees); this discrepancy in the correlation values due to topography is clearly more evident during the cold months of the year for both 3B42 V6 and CMORPH. In the warm season, 3B42 V6 is characterized by an average correlation value of 0.65, while the corresponding value for CMORPH is 0.55. In the cold season, the average correlation value for both satellites is 0.5.

Fig. 3.
Fig. 3.

Time correlation maps (top two rows) between each satellite product and the reference and (bottom row) between the two satellite products for the (left) cold and (right) warm seasons.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

The moderate values of correlation between the two satellite products and the gauges can be attributed partly to the weak spatial representativeness of the rain gauges. Gauges and satellites differ considerably in terms of scale, and this difference can have a significant impact on the correlation values. Furthermore, the interpolation of point measurements of rainfall leads to uncertainties, which in turn affect the accuracy. Moreover, errors in the retrievals are inherent in satellite products and this amplifies the difference from the rain gauges, reducing the correlation.

On the contrary, 3B42 V6 and CMORPH seem to correlate well in both seasons, although the correlation is significantly higher in the warm season, where correlation coefficient values are above 0.8 for the majority of the grid cells. It should be noted that in both seasons, but mainly in the cold season, correlation between the two satellite products over the mountains is significantly lower. Specifically, the greater area of the Alps exhibits a very low correlation coefficient relative to the surrounding areas. During the cold months of the year, central parts of the study domain as well as coastal regions in northern Spain and France display a low correlation coefficient as well. This is also true in the warm season, but the significance of this phenomenon is lower. Furthermore, the two satellites seem to agree well over Portugal and a big part of southeastern France, regardless of the season.

In Fig. 4, we show the seasonal maps of ME for both satellite products, presented in mm day−1. ME maps depict how the mean error varies in space for both 3B42 V6 and CMORPH in the cold and warm seasons. Shades of red indicate a positive error (overestimation), whereas shades of blue are used over areas with a negative error (underestimation). It is evident that 3B42 V6 tends to overestimate throughout the year; this overestimation becomes more significant in magnitude during the cold months, but more prominent spatially during the warm season (where the mean error ranges from −1.5 to 1.5 mm day−1). Overall, there seems to be no obvious seasonal cycle in the mean error for 3B42 V6, which is in agreement with the findings of Huffman et al. (2007), who conducted an early validation of TMPA against ground-based gauges. Conversely, CMORPH underestimates in the cold season, whereas during the warm months the underestimation is limited to the central and northern part of Europe, as well as the west-facing coastal areas near the Atlantic Ocean. These findings agree with those of Tian et al. (2007). Their analyses showed that CMORPH biases (mean differences) are season dependent, with mostly overestimation in the summer and underestimation in the winter, while 3B42 V6 biases over CONUS were significantly low in the summer. A closer look at Fig. 4 reveals that in the case of 3B42 V6 the overestimation becomes significantly (up to fivefold) more intensified over the mountainous areas (Alps and Pyrenees) during the cold months of the year, while in the warm season (during which most of the convective rain takes place) topography does not seem to have any major effect. These observations are in disagreement with those of Scheel et al. (2010), who evaluated the TMPA performance over the high terrain of central Andes. Their results conclude that both variance and bias are higher in the wet season, which occurs during the warm months of the year [December–February (DJF)], because of increased precipitation amounts. However, the very different climatological conditions between the two study domains seem to play a major role in this discrepancy; the Andes are subject to dry winters [June–August (JJA)] whereas the major mountain ranges in Europe have relatively wet winters with high amounts of rain and snow. The accuracy of CMORPH also seems to be impacted by the topography of the area in the cold season, but not during the warm summer months. More specifically, although CMORPH underestimates rainfall in the cold season, high-elevation regions (Alps, Pyrenees, and Carpathian Mountains) are subject to a significant overestimation. During the warm season there does not seem to be any specific trend driven by topography. Within each satellite product one can also notice certain seasonal effects; in the case of 3B42 V6, the seasonal effect is minor and is the intensification of the overestimation over the mountainous areas during the cold months. However, CMORPH performance shows more significant seasonal effects as the underestimating trend that prevails in the cold season clearly subsides during the warm months and partially gives way to a noteworthy overestimation. Arguably, overland microwave retrievals during the cold season are plagued by deficiencies, which become more profound at high elevations characterized by colder surfaces impacting the satellite measurements. Hence, seasonal effects should not be neglected, since the poor performance of the satellite retrievals over mountainous terrain seems to be dependent on the season. Finally, the overall error for 3B42 V6 is lower than that for CMORPH. Sapiano and Arkin (2009), who examined the performance of CMORPH and TMPA (among other satellite products) against gauge estimates, also found that TMPA had the lowest biases, which is expected as TMPA has the obvious advantage of the gauge bias correction (applied as a monthly mean correction).

Fig. 4.
Fig. 4.

Mean error (mm day−1) maps of (left) CMORPH and (right) 3B42 V6 in the (top) cold and (bottom) warm seasons.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

In Fig. 5, we show the seasonal maps of RESD (in mm day−1) for both products. A point to note from Fig. 5 is that in the cold season 3B42 V6 exhibits higher RESD relative to CMORPH, whereas during the warm months there are no significant differences between the two satellite products. Within the same satellite technique, seasonal effects are also apparent; in the case of CMORPH the RESD is greater in the warm season, with an average value of 6 mm day−1. Higher values are evident over mountainous regions (mainly the Alps and the Carpathian Mountains) in both seasons for CMORPH. 3B42 V6 RESD values are generally higher compared to those for CMORPH. The seasonal effect is significant for the RESD values of this satellite product as well, as the regions characterized by high values in the warm season are subject to even higher values in the cold months of the year. However, relief seems to strongly affect the RESD for the retrieval technique of both satellite techniques. Mountainous terrain is consistently characterized by higher RESD values, especially in the cold season. Overall, the two satellite products exhibit a different behavior in terms of the RESD error statistic. RESD values are generally high in the cold season for 3B42 V6, while they are high in the warm season for CMORPH. In both cases (3B42 V6 and CMORPH), though, mountainous terrain is characterized by high RESD values. However, gauge representativeness errors due to the low gauge density in these regions may have contributed to the enhanced RESD values. Finally, a comparison between the rain gauge error maps (right panels of Fig. 2) and the RESD maps in Fig. 5 leads to the conclusion that gauge error is significantly lower than the RESD values of both remote sensing products at the corresponding space–time resolution, which justifies the use of the rain gauge grid-interpolated product as a reference dataset. A point to note here is that RESD is an error statistic that does not contain the high satellite bias, and thus it is considered a rather precise estimate of the satellite’s error variance.

Fig. 5.
Fig. 5.

Random error standard deviation (mm day−1) maps of (left) CMORPH and (right) 3B42 V6 in the (top) cold and (bottom) warm seasons.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

b. Quantitative analysis

A more quantitative approach for the evaluation of the satellite technique performance is presented here, assessing four different statistics (NME, NRESD, NMRV, and NFASRV, together with the ratio of NMRV to NFASRV) for the entire European domain. All statistics used here were assessed only for those 0.25° grid sites of the study domain that host at least one rain gauge within their area. In Fig. 6 we show the NME values for the two satellite products determined for different gauge rainfall thresholds (upper panels) and elevation categories (lower panels) for each season. The normalized mean error is essentially the mean error of the satellite products divided by the mean gauge rainfall. In both seasons, rainfall intensity affects the NME for both satellite products, as there is a significant decrease in the NME values with increasing rainfall intensity, especially at high rain rates (≥16 mm day−1). For both satellite products the lowest NME values occur in the cold season, except for the very high rainfall rates, where the warm season exhibits the lowest NME. Both satellite techniques are characterized by negative NME values for all rainfall thresholds, except for the unconditional case (≥0) where the NME is positive in the warm season for both products; and in the cold season for 3B42 V6 only. However, NME values are for both seasons lower for CMORPH relative to those for 3B42 V6. There does not seem to be any significant seasonal effect on the NME for either satellite product; although the NME values for the two seasons are almost identical in the unconditional case (≥0) for both satellite techniques, the two products differ in terms of the trends the NME values take for the two seasons. Specifically, the discrepancy in the NME values between the two seasons decreases with rainfall threshold in the case of CMORPH, whereas this discrepancy increases for 3B42 V6. By and large, both satellite techniques are prone to underestimation with increasing rainfall rates in both seasons. For both satellite products the standard deviation of the NME values increases with rainfall threshold, but this effect is stronger for the 3B42 V6 estimates. When plotted against gauge-elevation categories, the cold-season estimates of both CMORPH and 3B42 V6 exhibit an increasing NME with elevation, which is mostly notable for the high altitudes (>1500 m). Specifically, for both satellite products, NME values in the winter are either negative or slightly above zero for elevations lower than 1500 m, and strongly positive for elevations exceeding 1500 m. Furthermore, 3B42 V6 is characterized by higher NME values than those for CMORPH in the cold season. During the warm months the mean of NME values for both satellite products is close to zero and slightly decreasing with elevation. The seasonal effects are apparent for CMORPH, where in low-to-moderate elevations the technique tends to underestimate rainfall in the cold season relative to the warm season, while only at high elevations cold season exhibits overestimation. The 3B42 V6 shows similar NME values of the cold- and warm-season estimates for low–moderate altitudes and a strong divergence for the high altitudes (>1500 m) where cold-season NME values are significantly higher than the warm-season values. Another point to note is the high standard deviations for the NME values at high-elevation categories for both satellite products, indicating high spatial variability of NME in high-elevation terrain.

Fig. 6.
Fig. 6.

Normalized mean error of (left) CMORPH and (right) 3B42 V6 evaluated (top) for different gauge rainfall thresholds and (bottom) for different gauge-elevation categories in the cold and warm season. The vertical bars represent the standard deviations. Sample sizes for each group are presented in boxes in different colors according to the season.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

The next error statistic used in this study is NRESD. In Fig. 7 we show the NRESD values of the two satellite products for different gauge rainfall thresholds (upper panels) and gauge-elevation categories (lower panels) of both seasons. In terms of CMORPH, the pattern of NRESD versus rainfall threshold is the same—namely, there is no apparent dependence of NRESD on the rainfall intensity category, except for the very high rainfall rates where the warm season exhibits an increase in the NRESD values. The 3B42 V6 satellite product, on the other hand, exhibits a tendency for NRESD to increase at rainfall thresholds above 16 mm day−1. Conversely, elevation seems to have strong effects on the NRESD for both satellite retrieval techniques. In the case of CMORPH, NRESD increases with elevation in the cold season, whereas the opposite is true during the warm season. The 3B42 V6 product, on the other hand, displays similar patterns in both seasons up to 1500 m altitude—namely, the trend for the NRESD values is to decrease with elevation. Above 1500 m the trend for NRESD in the warm season is to further decrease with elevation, while in the cold season to moderately increase with elevation. A point to note is that in the warm season both satellite products exhibit a similar NRESD decreasing trend with elevation; in the cold season, NRESD values are higher for 3B42 V6 than those for CMORPH and their respective patterns are opposite. Comparing the vertical bars on the graphs indicates higher spatial variability in the 3B42 V6 NRESD versus elevation patterns in the cold season. In the warm season the spatial variability in the NRESD versus elevation patterns is higher for the CMORPH product.

Fig. 7.
Fig. 7.

As in Fig. 6, but for error standard deviation.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

The last error metrics used for this part of our analysis are NMRV, NFASRV, and their ratio, all of which are presented in Fig. 8 for the two satellite products and for both seasons. The NMRV (upper panels of Fig. 8) is calculated by summing the gauge-measured rainy events that were not captured by the satellite and dividing them by the total amount of gauge-measured rainfall. Similarly, the NFASRV (middle panels) is the summation of the satellite-measured rainy events for which the reference (rain gauges) indicated no rain, divided by the total amount of gauge-measured rainfall. Finally, the ratio of the NMRV and NFASRV (lower panels) is shown on a logarithmic scale. All three statistics are plotted against gauge-elevation categories in order to show how topography affects the performance or accuracy of the two satellite retrieval techniques. NMRV is shown to decrease with elevation in both CMORPH and 3B42 V6 and in both seasons; the decreasing trend is stronger in the 3B42 V6, which is associated with higher NMRV values in elevations up to 1500 m, and within each satellite product—stronger in the cold season. Overall, CMORPH exhibits lower NMRV values than 3B42 V6; while between seasons, cold season exhibits higher NMRV values than warm season for both satellite estimates at low or moderate elevations. At high elevations the warm season exhibits higher NMRV values for both 3B42 V6 and CMORPH, although the difference is more significant in the case of 3B42 V6. In terms of NFASRV, we note that the warm season exhibits significantly higher values than the cold season for both satellite products, regardless of the elevation. Comparing the two satellite estimates, NFASRVs are higher for 3B42 V6 (for all elevations) in both seasons. For both satellite products, warm-season NFASRV exhibits very little variability among different elevation categories (except for the case of 3B42 V6 where NFASRV increases at very high elevations), while in the cold season it tends to increase up to elevations of 1500 m, above which it decreases. The ratio of NMRVs to NFASRVs shows similar patterns for the two satellite products. Specifically, we note a strong decreasing trend with elevation for both seasons, indicating that at higher altitudes the false alarm rainfall volumes are significantly higher than the missed rainfall volumes. Overall, the warm season exhibits higher NFASRVs than NMRVs for CMORPH at all elevations, while the opposite (same) is true for the warm season for 3B42 V6 for low-to-moderate (high) elevations. In the cold season, CMORPH and 3B42 V6 exhibit higher (lower) NMRVs than NFASRVs at low-to-moderate (high) elevations.

Fig. 8.
Fig. 8.

(top) Normalized missed rainfall volume, (middle) normalized false alarm rainfall volume, and (bottom) their ratio for (left) CMORPH and (right) 3B42 V6 evaluated for different gauge-elevation categories in the cold and warm season. Vertical bars represent standard deviations. Sample sizes for each value are presented in boxes in different colors according to the season.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

c. Time series analysis

The temporal evolution of two statistics (NME and correlation coefficient) is investigated for the two selected regions of the central–northern study domain. Region 1 extends over southwestern Germany and part of northeastern France and comprises mostly low-elevation terrain. Region 2 is located on the central–eastern part of the Alps and is highly mountainous. Both areas are of equal size (45 000 km2) and are characterized by an even distribution of rain gauges. Both performance indicators investigated in this section are calculated based on those 0.25° grid boxes that host at least one rain gauge within their domain. Each graph shown in Figs. 9 and 10 represents the change of the investigated statistic in time for the entire 8-yr period (2003–10) calculated for 3-month intervals: June–August, September–November, December–February, and March–May.

Fig. 9.
Fig. 9.

Temporal variability of the normalized mean error (%) for 3-month rainfall accumulations of 3B42 V6 and CMORPH over regions (top) 1 and (bottom) 2.

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

Fig. 10.
Fig. 10.

As in Fig. 9, but for CC (unitless).

Citation: Journal of Hydrometeorology 13, 2; 10.1175/JHM-D-11-086.1

Figure 9 presents the temporal evolution of the NME (shown as percentage) for the two satellite products in region 1 (upper panel) and region 2 (lower panel). The NME of 3B42 V6 over the low-elevation area (region 1) seems to fluctuate around zero, with the lowest values exhibited in the coldest months (DJF) and the highest ones in the warmest season (JJA); the majority of the values range from −25% to 25%. The NME of CMORPH follows a similar trend with the highest underestimation occurring in the cold months. CMORPH, however, exhibits higher extreme values both in cases of overestimation (positive NME) and underestimation (negative NME). In general, NME values for CMORPH are lower than those for 3B42 V6. Over mountainous terrain (region 2), the performance of the two satellite techniques differs substantially. While 3B42 V6 shows a general overestimation, CMORPH seems to mostly underestimate in a season-dependent fashion, where the underestimation becomes more significant in the cold seasons (up to 80%) and less in the summer. A point to note is the behavior of CMORPH’s NME over mountainous terrain (region 2); specifically, the estimates exhibited a significant overestimation in the summer of 2003, while in the remaining summers the satellite technique underestimated rainfall with an increasing trend in recent years. In terms of 3B42 V6 the NME does not show a clear seasonal dependence over high terrain.

The second statistic used in the time series analysis is CC. Figure 10 shows the temporal variability of CC values for the two satellite techniques determined over 3-month intervals (as in Fig. 9) in the 8-yr period over region 1 (top panel) and region 2 (lower panel). As seen from the figure, the temporal patterns of the CC are similar for both satellite products over both low- and high-elevation areas. In the case of low-elevation terrain the correlations seem to be consistently positive. By and large, the correlation values for both satellite retrieval techniques over region 1 range from 0.2 to 0.87, while those over region 2 range from −0.3 to 0.9. There seems to be no specific seasonal effect for either satellite product over the low-elevation region. Both techniques, however, show a clear seasonal effect on their correlations with the reference data over mountainous terrain; the higher correlation values are shown primarily in the fall [September–November (SON)] and to a lesser extent in the summer (JJA), while the lowest CCs are exhibited by both satellites in the coldest months of the year (DJF). Overall, the change of CC values in time for both 3B42 V6 and CMORPH is almost identical with the seasons, clearly impacting the temporal patterns.

5. Conclusions

An evaluation of near-continent-wide precipitation daily estimates from 8 yr (2003–10) of two satellite products was conducted using daily rain gauge rainfall data. All rainfall products were available on a 0.25° regular latitude–longitude grid. The satellite rainfall techniques whose quality was investigated are TRMM 3B42 (V6) and CMORPH. The study area of this work is the greater European domain, focusing on the central and southern (Mediterranean region) part. Various validation statistics were used to assess the performance of the aforementioned techniques, through a qualitative evaluation over the entire study area, a more quantitative approach in an attempt to give insight into quantifying the uncertainty for the two satellite techniques, and finally a time series analysis over two regions located in northeastern Italy (high-elevation terrain) and southwestern Germany (low-elevation terrain).

Mean gauge rainfall was found to be affected by both topography and season. A large part of the European domain is characterized by higher rainfall in the warm season; this is due to the convective rain that occurs over Europe (especially the western part of it) throughout the warm months of the summer. A significant orographic effect on the mean daily rainfall was also evident, as all high-elevation areas exhibited a two- to threefold increase of rainfall; the Alps in the cold season seem to receive less rainfall relative to the surrounding areas because snow and other solid-form types of precipitation were excluded from the study. Gauge-interpolation uncertainty was also assessed in this paper. The mean gauge error standard deviation yielded higher values for the warm season; this is mainly due to the convective nature of rainfall, which increases gauge-interpolation uncertainty. Overall, however, gauge sampling error variances determined across the study area are significantly lower than the RESD values of the two satellite products, indicating that these estimates constitute a good reference dataset for use in satellite evaluation studies.

The study showed moderate-to-high values of correlation between the two satellite products and the gauges; moderate values were attributed in part to the weak spatial representativeness of the gauges and partly on the satellite retrieval error variability. The study showed a strong magnitude dependence on correlation. Higher daily rainfall accumulations exhibit higher correlation values, but this is only valid for low-elevation areas; mountainous areas, despite their high mean rainfall, are characterized by low correlation coefficients. The two satellite products correlate well (especially in the warm season) over most of the study area, despite their sampling differences. Mountainous regions also exhibited low correlation values between the two satellite products. Taking into account the findings of both qualitative and quantitative statistical analyses presented in this study, we conclude the following:

  1. TRMM 3B42 V6 exhibits overestimation of rainfall in both seasons, with that during the cold season being more significant over the mountainous regions and generally over areas characterized by high mean daily rainfall. This overestimation yields high RESD values, which become significantly higher in the cold season. The version of CMORPH evaluated in this paper was found to significantly underestimate over low-elevation areas and overestimate over high-terrain areas in the cold season, while during the warm months its performance shows a dipole behavior; underestimation is prominent in the central and northern study-domain, while overestimation is more common over Mediterranean regions.

  2. Performance of both satellite products tends to change significantly over mountainous terrain during the cold season. This phenomenon indicates an important orographic effect which takes place in the winter months and is primarily attributed to snow/cold surface contamination. On the other hand, during the warm-season months, relief does not impact the performance of the two satellite products.

  3. The error variance of both satellite estimates is subject to seasonal changes and is generally higher over the mountains. CMORPH’s RESD values are higher in the summer, while those for 3B42 V6 are higher in the winter. Overall, in terms of RESD, CMORPH exhibits higher accuracy in the winter relative to 3B42 V6. The opposite is shown in terms of ME, where the 3B42 V6 product is associated with lower values than those of CMORPH.

  4. Increasing rainfall intensity leads to an increasing underestimation for both satellites, a phenomenon which is even more profound in the warm season. Elevation also seems to impact the mean satellite error, as the performance of both products weakens at high elevations. This is also verified by the fact that at elevations greater than 1500 m the NME increases by 100%. The error variability seems to be elevation-dependent for CMORPH (although in different fashions for different seasons), whereas 3B42 V6 shows a decreasing-with-elevation pattern for its error variability in the warm season, and no specific trend during the cold months.

  5. The normalized missed rainfall volume decreases with elevation for both satellite techniques and both seasons. At moderate-to-high elevations however, that fraction of rainfall volume is even smaller in the cold season. Both satellite products have a tendency to falsely indicate rain, and that tendency increases only at very high elevations in the warm season, while in the cold months, the NFASRV increase is evident only up to 1500 m. NFASRV is higher in the warm months for both 3B42 V6 and CMORPH regardless of the elevation. 3B42 V6 exhibits a more balanced (relative to 3B42 V6) performance throughout the year especially over moderate-elevation areas, where NFASRV and NMRV tend to cancel each other’s effect (ratio close to zero).

  6. The NME for 3B42 V6 shows season-dependency only over low elevations, whereas the NME for CMORPH depends on seasons over both low and high terrain. In general, for both satellite products over low-elevation areas the underestimation becomes more significant in the winter, while the overestimation prevails in the summer. This is not the case with mountainous areas, where the seasonal effect on the NME is not clear, especially for 3B42 V6.

  7. The two satellite products exhibit season-dependent correlations with the reference data over high-elevation regions, where the correlations become low or moderate in the winter and high in the fall or summer. Over low-elevation areas however, the correlations are not characterized by such great ranges of values, being constantly positive and with no specific seasonal effect.

In summary, we have shown that seasonal changes and topographical effects greatly influence the performance of satellite rain retrieval techniques. Europe is a complex study area, as it is associated with a long coastline, relief extremes, and a directional trend of its mountain ranges. The results, therefore, presented herein may not be valid for other regions, as precipitation processes differ regionally. Future continuation should include other high-resolution global-scale satellite rainfall products. Moreover, assessing the performance of satellite rainfall products of high temporal resolution (3 h) would also give an invaluable insight into quantifying their uncertainty. Evaluating the efficiency of these satellites over other densely gauged areas within the complex European domain would also provide useful knowledge. Furthermore, reference datasets of higher accuracy and validity would be very useful for the purpose of such studies. Radar-derived precipitation data can be jointly used along with rain gauges as a more reliable and accurate reference dataset for comparisons against satellite products. Finally, investigating the potential of TRMM 3B42 V6 and CMORPH for flood-forecasting caused by extreme (storm) events, over smaller areas characterized by complex terrain remains a challenging proposition, that will undoubtedly prove useful to hydrologists and those involved in risk assessment and water management.

Acknowledgments

This work was supported by a NASA Precipitation Measurement Mission award (NNX07AE31G). We also acknowledge the E-OBS dataset from the EU-FP6 project ENSEMBLES (http://ensembles-eu.metoffice.com) and the data providers in the ECA&D project (http://eca.knmi.nl).

REFERENCES

  • Ali, A., Lebel T. , and Amani A. , 2005: Rainfall estimation in the Sahel. Part I: Error function. J. Appl. Meteor., 44, 16911706.

  • Anagnostou, E. N., Maggioni V. , Nikolopoulos E. I. , Meskele T. , Hossain F. , and Papadopoulos A. , 2009: Benchmarking high-resolution global satellite rainfall products to radar and rain-gauge rainfall estimates. IEEE Trans. Geosci. Remote Sens., 48, 16671683.

    • Search Google Scholar
    • Export Citation
  • Conner, M. D., and Petty G. W. , 1998: Validation and inter-comparison of SSM/I rain-rate retrieval methods over the continental United States. J. Appl. Meteor., 37, 679700.

    • Search Google Scholar
    • Export Citation
  • Dinku, T., Ceccato P. , Grover-Kopec E. , Lemma M. , Connor S. J. , and Ropelewski C. F. , 2007: Validation of satellite rainfall products over East Africa’s complex topography. Int. J. Remote Sens., 28, 15031526.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Gruber, A. V., and Levizzani V. , 2006: Assessment of global precipitation: A project of the Global Energy and Water Cycle Experiment (GEWEX) Radiation Panel. GEWEX, World Climate Research Program, WMO Rep., 61 pp.

  • Haylock, M. R., Hofstra N. , Klein Tank A. M. G. , Klok E. J. , Jones P. D. , and New M. , 2008: A European daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006. J. Geophys. Res., 113, D20119, doi:10.1029/2008JD010201.

    • Search Google Scholar
    • Export Citation
  • Hossain, F., and Huffman G. J. , 2008: Investigating error metrics for satellite rainfall data at hydrologically relevant scales. J. Hydrometeor., 9, 563575.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, 520.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 3855.

    • Search Google Scholar
    • Export Citation
  • Hughes, D. A., 2006: Comparison of satellite rainfall data with observations from gauging station networks. J. Hydrol., 327, 399410.

  • Joyce, J. R., Janowiak E. J. , Arkin P. A. , and Xie P. , 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487503.

    • Search Google Scholar
    • Export Citation
  • Kidd, C., Kniveton R. D. , Todd C. M. , and Bellerby J. T. , 2003: Satellite rainfall estimation using combined passive microwave and infrared algorithms. J. Hydrometeor., 4, 10881104.

    • Search Google Scholar
    • Export Citation
  • Krajewski, W. F., and Smith J. A. , 2002: Radar hydrology: Rainfall estimation. Adv. Water Resour., 25, 13871394.

  • Krajewski, W. F., Ciach G. J. , McCollum J. R. , and Bacotiu C. , 2000: Initial validation of the Global Precipitation Climatology Project monthly rainfall over the United States. J. Appl. Meteor., 39, 10711086.

    • Search Google Scholar
    • Export Citation
  • Kubota, T., and Coauthors, 2007: The Global Satellite Mapping of Precipitation (GSMaP) project. Proc. Seventh GPM Int. Planning Workshop, Bellesale Kudan, Tokyo, Japan, Ken’ichi Okamoto (Osaka Prefecture University), 14 pp. [Available online at http://www.eorc.jaxa.jp/GPM/ws7/pdf/7thDEC2007/7_am07.pdf.]

  • McCollum, J. R., Krajewski W. F. , Ferraro R. R. , and Ba M. B. , 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 10651080.

    • Search Google Scholar
    • Export Citation
  • McPhee, J., and Margulis S. A. , 2005: Validation and error characterization of the GPCP-1DD precipitation product over the contiguous United States. J. Hydrometeor, 6, 441459.

    • Search Google Scholar
    • Export Citation
  • Sapiano, M. R. P., and Arkin P. A. , 2009: An intercomparison and validation of high-resolution satellite precipitation estimates with 3-hourly gauge data. J. Hydrometeor, 10, 149166.

    • Search Google Scholar
    • Export Citation
  • Scheel, M. L. M., Rohrer M. , Huggel C. , Santos Villar D. , Silvestre E. , and Huffman G. J. , 2010: Evaluation of TRMM Multi-satellite Precipitation Analysis (TMPA) performance in the Central Andes region and its dependency on spatial and temporal resolution. Hydrol. Earth Syst. Sci. Discuss., 7, 85458586.

    • Search Google Scholar
    • Export Citation
  • Sorooshian, S., Hsu K.-L. , Gao X. , Gupta H. V. , Imam B. , and Braithwaite D. , 2000: Evaluation of PERSIANN system satellite–based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 20352046.

    • Search Google Scholar
    • Export Citation
  • Su, F., Hong Y. , and Lettenmaier D. P. , 2008: Evaluation of TRMM Multisatellite Precipitation Analysis (TMPA) and its utility in hydrologic prediction in the La Plata basin. J. Hydrometeor., 9, 622640.

    • Search Google Scholar
    • Export Citation
  • Tian, Y., Peters-Lidard C. D. , Choudhury B. J. , and Garcia M. , 2007: Multitemporal analysis of TRMM-based satellite precipitation products for land data assimilation applications. J. Hydrometeor., 8, 11651183.

    • Search Google Scholar
    • Export Citation
  • Turk, F. J., and Miller S. D. , 2005: Toward improved characterization of remotely sensed precipitation regimes with MODIS/AMSR-E blended data techniques. IEEE Trans. Geosci. Remote Sens., 43, 10591069.

    • Search Google Scholar
    • Export Citation
  • van Engelen, A., Klein Tank A. , van de Schrier G. , and Klok L. , 2008: Towards an operational system for assessing observed changes in climate extremes. European Climate Assessment & Dataset Rep., 68 pp.

  • Xie, P., and Arkin P. A. , 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 25392558.

    • Search Google Scholar
    • Export Citation
  • Yamamoto, J. K., 2000: An alternative measure of the reliability of ordinary kriging estimates. Math. Geol., 32, 489509, doi:10.1023/A:1007577916868.

    • Search Google Scholar
    • Export Citation
Save