1. Introduction
Precipitation is a key parameter of the earth’s weather and climate. The release of latent heat by condensation is one of the most important drivers of the global atmospheric circulation. Latent heat release is particularly important in the tropics, where the majority of the earth’s precipitation falls. Accurate measurement of precipitation is important for a number of scientific and societal applications, but it is a technically challenging problem, in part because of the statistical properties of precipitation. Rain is highly variable in space and time and has very short spatial and temporal correlation length scales. This means, for example, that a rain gauge may provide good data at a point but will give only limited information about rain rates at nearby locations. For some purposes these difficulties can be overcome by appropriately averaging rain measurements in space and/or time.
Over land, precipitation can be measured by using conventional rain gauges and radars, but measurement of precipitation over the world’s oceans remains a difficult problem. In situ rain observations are provided by ships of opportunity, which tend to follow a small number of major shipping lanes, and by a limited number of rain gauges on ocean buoy networks. Rain gauges on islands may be affected by the presence of the island and may not provide data that are representative of the surrounding ocean. However, satellite microwave remote sensing instruments now provide a wealth of precipitation information with potentially global coverage. The Defense Meteorological Satellite Program (DMSP) operational meteorological satellites have carried a series of Special Sensor Microwave Imagers (SSM/I). These are conically scanning multichannel passive microwave radiometers. For the past 11 years the Tropical Rainfall Measuring Mission (TRMM) satellite has carried a microwave radiometer that is based on the SSM/I and is known as the TRMM Microwave Imager (TMI). These instruments have found widespread use in precipitation remote sensing, and several groups use the observed microwave radiances to estimate surface rain rates.
This study compares several different satellite rain-rate estimation (retrieval) algorithms with each other and with data from rain gauges on ocean buoys. The purpose is to evaluate the accuracy of the satellite retrievals and to document the biases between the various observing systems.
2. Data
a. TAO/TRITON/PIRATA buoy rain gauge data
Surface rain-rate observations are available from a subset of the Tropical Atmosphere–Ocean (TAO) buoy network, Triangle Trans-Ocean Buoy Network (TRITON), and Pilot Research Moored Array in the Atlantic (PIRATA) buoy network in the tropical Pacific and Atlantic Oceans (Hayes et al. 1991; McPhaden et al. 1998; Serra et al. 2001; Serra and McPhaden 2004). The locations of buoys with rain gauge data are shown in Fig. 1. In this study we use data from 43 buoys, omitting 6 buoys that have fewer than 6 months of data. Data were downloaded from the Tropical Atmosphere Ocean Project (http://www.pmel.noaa.gov/tao/data_deliv/deliv.html) at the National Oceanic and Atmospheric Administration (NOAA) Pacific Marine Environmental Laboratory (PMEL). In this paper, the gauges are numbered for reference; numbers increase from west to east and north to south.
The buoys measure rain accumulations using R. M. Young Company self-siphoning capacitance-type rain gauges. Rain rates are calculated at 1-min intervals by differencing the instantaneous accumulations. The time series of 1-min values are filtered and converted to 10-min rain rates by PMEL. This study uses data observed from December 1997 to November 2006. The availability of rain gauge data and the data quality flags are shown in Fig. 2. The color of each horizontal bar corresponds to the quality flags in the data files. As can be seen, the gauge data are available intermittently, with virtually all gauges having relatively long gaps. There are some small gaps that are not visible at the scale of the figure, but they are infrequent. The great majority of the observations are classified as “default” quality (green), with some intervals of “adjusted” or “low” quality (yellow or red). There are no precipitation observations in this set of buoys flagged as “high” quality (blue). Eleven buoy records have data flagged as low quality. Almost all buoys have at least some data flagged as adjusted, but the amounts are typically small (<0.3% of the observations, except for four buoys with amounts of 0.8%, 1.4%, 2.2%, and 5.3%, respectively).
The rain-rate data used here have not been corrected for undercatch due to wind. Serra et al. (2001) and Serra and McPhaden (2003), using earlier studies by Koschmieder (1934), WMO (1962), and Yang et al. (1998), estimated the wind-induced undercatch for the buoy rain gauges to be in the range of 10%–50%. Systematic errors in the rain gauge data are potentially very significant, but this problem has received surprisingly little attention, and the uncertainty is very large.
b. TRMM data
The TRMM satellite is a joint U.S.–Japanese mission that was launched in November of 1997. Designed to observe precipitation in the tropics, it operates in a low-inclination (35°), low-altitude orbit. During approximately the first 3.5 yr of operation, TRMM flew at an altitude of ∼350 km. In August of 2001 the orbit was boosted to ∼400 km to reduce drag, and thus fuel use, and to increase the lifetime of the mission. TRMM has significantly outlasted its nominal 3-yr design lifetime. Details of the TRMM and instruments can be found in Simpson et al. (1988) and Kummerow et al. (1998, 2000).
The primary TRMM precipitation instruments are the precipitation radar (PR) and the TMI. This study uses data from the TMI, which is a multichannel, dual-polarization, passive microwave radiometer based on the SSM/I instruments used on DMSP polar orbiters. The TMI uses an offset antenna to scan the earth’s surface in a conical pattern across the satellite’s flight track. This scanning pattern is designed to maintain a constant viewing angle at the earth’s surface. This design eliminates variations in emitted radiation due to viewing-angle dependence of the surface microwave emissivity. The size of the instrument’s field of view depends on the frequency, ranging from 63 km × 37 km at 10.65 GHz to 7 km × 5 km at 85.5 GHz. The TMI swath width is ∼780 km, which means that the mean sampling interval at the equator is about 1 day. Precipitation retrieval algorithms over the ocean rely primarily on emission of microwave radiation from hydrometeors observed against the relatively cold background provided by the low-emissivity ocean surface.
As noted above, the TMI views a given location at intervals of about 1 day. Because of the large variability of precipitation in space and time, this relatively infrequent sampling leads to significant constraints on estimates of space- and time-averaged rain rates. Unlike the operational polar-orbiting meteorological satellites, the TRMM orbit is not sun synchronous. It precesses with respect to the sun (i.e., through the diurnal cycle) with a period of about 6 weeks. If one counts both ascending and descending segments of the orbit, at the equator TRMM observes a complete diurnal cycle in about 3 weeks. At higher latitudes, sequential observations are closer together in local time. At the highest latitudes viewed by the satellite, 6 weeks are required to sample a complete diurnal cycle, but swath overlap means more samples are collected at higher latitudes than near the equator. Negri et al. (2002) evaluated the geographic inhomogeneity of the diurnal sampling of the TRMM orbit and its dependence on orbital characteristics such as altitude. In this study with ∼9 yr of TRMM data, the sampling is smoother than in the 3-yr examples in Negri et al. (2002). Area averaging can be used to reduce further the sampling inhomogeneity.
This study uses version 6 of the TRMM 3G68 data product, which contains instantaneous precipitation retrievals from the TMI 2A12 [Goddard profiling (GPROF)] algorithm that have been area averaged onto a 0.5° × 0.5° grid. There are very few gaps in the version-6 TRMM data record. For comparison, we also present results from version-4 TMI retrievals by Remote Sensing Systems, Inc. (RSS). The RSS retrievals use a different retrieval algorithm (Wentz 1997; Wentz and Spencer 1998; Hilburn and Wentz 2008). The RSS retrievals for TRMM are available on 0.25° × 0.25° grids. The RSS data for each day are split into two grids, one containing the ascending segments of the orbits for that day and one containing the descending orbits. Because the orbits overlap significantly at higher latitudes, some data in the subtropics are not included in the RSS files. For comparison with the TRMM data, the RSS data are area averaged onto the 0.5° × 0.5° 3G68 grid. The GPROF algorithm retrieves rainfall over land, but the RSS algorithm does not; therefore, we restrict our analysis to oceanic rainfall.
Over oceans we also compare the monthly means of the GPROF retrievals, a product known as 3A12, with an independent physical–statistical retrieval algorithm (3A11). The 3A11 retrievals are carried out, over ocean only, for 5° × 5° longitude–latitude boxes. The distribution of instantaneous rain rates is modeled by a mixed distribution consisting of a discrete probability of no rain and a lognormal distribution of nonzero rain rates (Wilheit et al. 1991; Chiu et al. 1993). The rain-rate distribution is matched to a histogram of observed brightness temperatures by using radiative transfer calculations, and a beam-filling correction is applied. For this comparison, the 3A12 data are averaged onto the same 5° × 5° grid as the 3A11 data.
c. SSM/I data
SSM/I instruments have operated on a number of DMSP satellites. In this study, we use SSM/I rain estimates from the F13, F14, and F15 satellites, which operate in sun-synchronous polar orbits at an altitude of ∼850 km and inclination of 98.8°. Equator crossing times are 0500–0600 local time. Because the orbits do not sample throughout the diurnal cycle, some systematic biases are possible when comparing the satellite datasets. In addition, the equator crossing times of some DMSP satellites have drifted by as much as several hours. By sampling on both the ascending and descending parts of the orbit, however, the diurnal harmonic, which normally dominates the diurnal cycle, tends to average to zero. The higher components of the diurnal cycle over the ocean are generally small, and therefore potential diurnal cycle effects are ignored in this study (Bowman et al. 2005). For the satellite–gauge comparisons, the gauge data are matched in time to the satellite overpasses, which largely removes any diurnal biases in the differences. As with the TRMM TMI data, we compare SSM/I rain-rate retrievals made by using both the National Aeronautics and Space Administration (NASA) version-6 GPROF algorithm and the RSS version-6 SSM/I algorithm. Both datasets are area averages on a 0.25° × 0.25° grid.
3. Results
a. Rain gauge noise and data quality
The rain gauge accumulation measurements contain noise that, when differenced, leads to small, fluctuating, positive and negative rain-rate values. A representative time series for one gauge is shown in Fig. 3. This is a dry location in the eastern tropical Pacific, with a mean rain rate of only ∼0.1 mm day−1. Rain events are infrequent, and larger events are easily visible during periods of low noise as large positive spikes that stand out above the noise. The noise itself is also easily visible in the plot as both positive and negative values. Using only data flagged as adjusted quality or higher, for most buoys the values are approximately evenly split between negative and nonnegative (zero or greater). The fraction of negative observations ranges from 44.6% to 63.4%, with a mean of 51.6%. Because positive rain rates due to noise cannot be distinguished from real rain at low rain rates, the noise cannot be removed in any simple fashion. Although they are not physically realistic, negative rain rates cannot simply be set to zero without introducing a bias into the data. Averaged across the 46 buoys, setting negative rain rates to zero would increase the mean rain rate by ∼0.5 mm day−1. These values vary by about a factor of 2 among the buoys, from 0.27 to 1.07 mm day−1. In relative terms, these biases are small in rainy locations (∼5%), whereas in dry locations they can be >100%. In this study we include all negative rain rates when calculating time averages.
A separate issue is whether to include data that are flagged as low quality. For the gauge shown in Fig. 3, the data quality varies substantially over the 6-yr period shown. Two intervals with large noise levels, from late 2000 to early 2001 and from late 2003 to early 2004, are flagged as low quality, indicated in the graph by the red bars. Note, however, that the intervals at the beginning of 1999 and in March of 2003, which also appear to have larger-than-normal noise amounts, are flagged as default quality. To qualitatively identify noisy periods, 5-day running means are computed using only the negative rain-rate amounts, ignoring zero and positive values. The resulting smoothed values and their negatives are plotted to show an approximate envelope of the noise (the red curves that are symmetric around zero).
Table 1 shows the impact on the long-term time mean of omitting data flagged as low quality from the records for the 11 buoys that have such data. The buoys are sorted by mean rain rate from rainiest to driest. The second and third columns of the table show the long-term time-mean rain rate with and without low-quality data included. The absolute changes caused by deleting low-quality data are small, generally <0.1 mm day−1, except for two cases (0.23 and 0.44 mm day−1, respectively). These results indicate that, even during periods of high instrumental noise, the mean of the noise is very close to zero. Note that the observed differences might not be due solely to the noise, because the real rain rate also varies with time. For the buoys in wetter locations the relative changes due to excluding the low-quality data are small (<7%). In dry areas the relative differences can be large.
We also examined each of the buoy records subjectively to identify periods such as those in Fig. 3 that have high noise but are flagged as default quality. Omitting these segments from the data record also had only a small impact on the long-term time means, indicating that the mean of the noise is close to zero, even during high-noise periods. Although the impact of the low-quality data is generally small, and most gauges do not contain data flagged as low quality, in this study we restrict our analysis to buoy rain gauge data flagged by PMEL as adjusted quality or higher. As explained earlier in this section, we include negative rain rates to avoid biasing the data.
b. Comparison of satellite and gauge data
The GPROF and RSS algorithms use different methods to estimate surface rain rates. In this section, the GPROF and RSS satellite retrievals are compared with the buoy rain gauge data by computing long-term time means of the precipitation rates from the satellite retrievals and the gauge data. The gauge data are time averaged in a 6-h window centered on each satellite overpass. The results are not very sensitive to the length of the time-averaging window, and 6 h is close to optimum (Bowman 2005). For the TMI, the satellite data for each overpass are area averaged in a 1° × 1° longitude–latitude box centered on each buoy position. For the SSM/I data, which are available on a higher-resolution grid, the satellite data are averaged in a 0.5° × 0.5° box centered on each buoy. Because the buoys are moored at the corners of the grid boxes for both of the original satellite analysis grids, these are the smallest grid boxes for each grid that have the gauges located at the centers of the boxes, rather than the edges. The resulting matched space and time averages from all overpasses are then averaged to estimate the time-mean area-mean rain rates at each gauge location. For the satellite data, all area and time averages are weighted by the number of pixels in each grid box. The differences in grid resolution should not affect the comparisons of mean rain rates. When comparing the satellite data with each other (sections 3c,d), identical grids are used. The gauges cover a range of climate regimes from dry to wet in the tropical Atlantic and Pacific Oceans. Note that, because of the inhomogeneities in the buoy records, the time averages for each gauge are for different time periods (see Fig. 2). All gauges used in this comparison have at least 6 months of data. The results are summarized in Fig. 4 and Table 2.
Figure 4 shows the relationship between the gauge data and the GPROF and RSS estimates for the TMI and the F14 SSM/I. For each satellite, the intercept a and slope b of the linear least squares fit to the data and their 95% confidence intervals are given in Table 2. The confidence intervals assume that the residuals follow a Gaussian distribution. Because of the large amount of averaging involved, this is a reasonable assumption in this case (Bowman et al. 2005). The table also gives the number of months of satellite data available for each comparison. This is an upper bound on the length of the data record available at each buoy. The RSS values for a are slightly larger than the GPROF values, but in all cases the intercepts are zero within the 95% confidence limits. For all four satellites the slopes of the regression lines are lower for the GPROF retrievals than for the RSS retrievals. The slopes of the fits for the RSS retrievals are equal to 1 within the uncertainty, while the slopes of the GPROF retrievals are negatively biased with respect to the gauges. For the TMI the bias is small, but for the GPROF SSM/I retrievals the biases range from 13% to 24%. In all cases the confidence limits on the RSS retrievals include 1, and therefore we cannot reject the null hypothesis that b = 1. For the GPROF retrievals, the confidence limits do not include 1, from which we can conclude that there is a statistically significant difference between the gauges and the GPROF rain-rate estimates at the 95% confidence level.
c. Comparison of GPROF and RSS retrievals
For each satellite it is possible to compare the GPROF and RSS retrievals directly with each other. This section compares the results of the two algorithms for the TMI instrument. Here we briefly summarize some significant differences between the two datasets. Because of the TRMM orbit inclination and TMI swath width, it is possible to have several observations per day in the subtropics, where successive orbital swaths overlap repeatedly. The RSS data files, however, store at most two observations per day, whereas the GPROF data files store all of the overpasses. Because of these differences in the gridding schemes between the two datasets, it is not possible to match every GPROF and RSS retrieval. This problem is most prominent near the northern and southern extremes of the TRMM satellite orbit at ±35°. Data are omitted from the analysis at these latitudes and along coastlines, where the number of samples is small.
Figure 5 is a scatterplot of the climatological time-mean precipitation for each box on the 0.5° × 0.5° grid. The averaging period is from December 1997 to October 2006. The dashed line indicates the one-to-one relationship. Black curves are contours of the density of the points, binned into 1 mm day−1 × 1 mm day−1 bins. The contour interval is logarithmic. The relationship between the two algorithms is relatively compact and is close to linear but does exhibit some curvature at higher rain rates. Fitting the data with a linear function of the form rR = a + brG (gray line), where rG is from the GPROF data and rR is the RSS value, yields a = −0.05 and b = 1.06. The slope b is consistent with the differences between the slopes from the regression of each retrieval separately with the rain gauge data (Fig. 4). The RSS retrievals are positively biased relative to the GPROF retrievals for rain rates up to about 10 mm day−1. Above about 5 mm day−1 there appears to be a separate population for which the RSS is negatively biased relative to GPROF. This is most noticeable at the highest rain rates (≳8 mm day−1). The results vary somewhat from month to month but exhibit the same general features seen in Fig. 5 for all months (not shown). Scatterplots for individual months show similar functional fits but have larger scatter, as would be expected from the smaller sample size used to compute each time-averaged value.
Figure 6 shows the geographical variations of differences in the retrievals, which have an interesting geographical pattern [see also Fig. 7 in Hilburn and Wentz (2008)]. As Fig. 5 shows, the RSS retrievals are greater than the GPROF values (shown in red) at most locations. The RSS tends to be larger than the GPROF values in the Atlantic ITCZ, the eastern Pacific north of the ITCZ, the western Pacific warm pool and SPCZ, and the tropical Indian Ocean. The RSS is less than the GPROF in the eastern Pacific ITCZ (shown in blue), which has some of the highest rain rates observed over the tropical oceans, and also in the dry regions in the eastern subtropical ocean basins. Thus, the bias between the retrievals is not random, but is systematic and is highly dependent on location. Berg et al. (2002) looked at regional variations in TRMM retrievals. Using the TRMM PR they found east–west variations in the vertical structure of rain systems across the Pacific Ocean. It is likely that these physical differences propagate through the RSS and GRPOF algorithms in different ways, leading to the observed biases. Comparing the TRMM TMI and PR retrievals, Berg et al. (2006) found that the ratio of TMI to PR rain rates is correlated with the column water vapor.
d. Comparison of TRMM 3A11 and 3A12 data
Last, we compare the monthly means of the TMI GPROF retrievals (3A12 data product) with the independent monthly-mean TMI retrievals done using the histogram-matching technique (3A11 data product). Figures 7 and 8 compare the long-term climatological means of the TRMM 3A11 and 3A12 retrievals on a 5° × 5° longitude–latitude grid. The averaging period is 1998–2007.
The scatterplot of values for each grid box in Fig. 7 indicates that there are two distinct populations. In one branch, the 3A11 values are higher than the 3A12 (labeled “Extratropical”); in the other, the 3A11 values are lower (labeled “Tropical”). The differences between the two retrievals increase with rain rate in an approximately linear fashion in both cases and amount to ∼10%–15%.
The geographical variations of the bias can be seen in the difference map in Fig. 8, which shows that negative differences occur in high-rain areas in the tropics, including the Atlantic and Pacific ITCZs, the Pacific warm pool, the SPCZ, and the Indian Ocean. In other areas, primarily poleward of about 25° latitude, the 3A11 data are larger than the 3A12 values. The differences are larger in the western portions of the ocean basins, with the largest differences in the western Pacific. The pattern of the biases between these two retrievals (Fig. 8) is very different from that between the RSS and GPROF shown in Fig. 6.
4. Summary and discussion
This study compares several microwave precipitation datasets for the tropical and subtropical oceans with each other and with rain gauge data from ocean buoys. In this study, long-time averages are used to reduce the random errors. Care should be taken when averaging over shorter periods, because the sampling errors can be substantial.
For the TMI instrument on the TRMM satellite, the biases between the different satellite rain algorithms, and between the satellite estimates and the rain gauge data, are generally small. The RSS retrievals have no statistically significant bias with respect to the rain gauges, but the GPROF retrievals are biased slightly negatively. The rain gauges might be biased slightly negatively with respect to the true rain rate because of undercatch during windy conditions. Insofar as the gauges can be considered to be the best measurements of the true rain rate, the GPROF retrievals are somewhat low (∼7%). This is consistent with the earlier comparison of TMI and gauge data in Bowman (2005), which found the TMI to be about 5% low with respect to the gauges.
The bias between the GPROF and RSS retrievals is confirmed by comparing them directly with each other. The RSS is high with respect to the GPROF except in the eastern Pacific ITCZ and in the eastern subtropics of all ocean basins (Fig. 6). This bias has an interesting geographical distribution and is not simply rain-rate dependent. That is, the bias is not simply proportional to the rain rate but also depends strongly on location. Rainy areas that might be expected to have similar biases, such as the ITCZ in the eastern tropical Pacific, and the corresponding ITCZ in the tropical Atlantic, have biases of opposite sign. Because the retrieval algorithms require assumptions about physical parameters, such as the beam-filling error and the freezing-level altitude, or must estimate those parameters as part of the retrieval, it is not surprising that different algorithms respond differently to regional variations in those quantities. Berg et al. (2002, 2006) found regional biases in the TMI retrievals related to differences in storm structure and column water vapor. Analysis of the sensitivity of the retrieval algorithms to errors in these parameters will help to identify and reduce systematic errors in the retrievals.
The TMI 3A11 and 3A12 (monthly-mean GPROF) retrievals, which use the same basic input data but different conceptual approaches, have a different geographic structure to their bias. In the tropics, the 3A11 values are low with respect to those of 3A12, and the differences are approximately linearly proportional to the rain rate. Outside the deep tropics, the 3A11 data are high in comparison with those of 3A12, and again the biases are roughly proportional to the rain rate.
Comparisons with in situ rain gauge data from ocean buoys show that satellite rain estimates over the ocean have relatively small biases with respect to the gauges. The exception to this is the GPROF retrievals that use SSM/I data from the DMSP satellite. These data have a substantial low bias with respect to the gauges (as much as ∼24%).
Acknowledgments
We acknowledge the TAO Project Office of NOAA/PMEL for providing the buoy rain gauge data. The NASA retrievals of TRMM and DMSP satellite data were acquired as part of the Tropical Rainfall Measuring Mission. The algorithms were developed by the TRMM Science Team. The data were processed by the TRMM Science Data and Information System (TSDIS) and the TRMM Office; they are archived and distributed by the Goddard Earth Sciences Data and Information Services Center (http://daac.gsfc.nasa.gov/). TRMM is an international project jointly sponsored by the Japan National Space Development Agency (NASDA) and the U.S. National Aeronautics and Space Administration (NASA) Office of Earth Sciences. SSM/I and TMI data produced by Remote Sensing Systems are sponsored by the NASA Earth Science REASoN DISCOVER Project. Data are available at www.remss.com.
REFERENCES
Berg, W., C. Kummerow, and C. A. Morales, 2002: Differences between east and west Pacific rainfall systems. J. Climate, 15 , 3659–3672.
Berg, W., T. L’Ecuyer, and C. Kummerow, 2006: Rainfall climate regimes: The relationship of regional TRMM rainfall biases to the environment. J. Appl. Meteor. Climatol., 45 , 434–454.
Bowman, K. P., 2005: Comparison of TRMM precipitation retrievals with rain gauge data from ocean buoys. J. Climate, 18 , 178–190.
Bowman, K. P., J. C. Collier, G. R. North, Q. Wu, E. Ha, and J. Hardin, 2005: Diurnal cycle of tropical precipitation in Tropical Rainfall Measuring Mission (TRMM) satellite and ocean buoy rain gauge data. J. Geophys. Res., 110 , D21104. doi:10.1029/2005JD005763.
Chiu, L. S., A. T. C. Chang, and J. Janowiak, 1993: Comparison of monthly rain rates derived from GPI and SSM/I using probability distribution functions. J. Appl. Meteor., 32 , 323–334.
Hayes, S. P., L. J. Mangum, J. Picaut, A. Sumi, and K. Takeuchi, 1991: TOGA-TAO: A moored array for real-time measurements in the tropical Pacific Ocean. Bull. Amer. Meteor. Soc., 72 , 339–347.
Hilburn, K. A., and F. Wentz, 2008: Intercalibrated passive microwave rain products from the unified microwave ocean retrieval algorithm (UMORA). J. Appl. Meteor. Climatol., 47 , 778–794.
Koschmieder, D. H., 1934: Methods and results of definite rain measurement. Mon. Wea. Rev., 62 , 5–7.
Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15 , 809–817.
Kummerow, C., and Coauthors, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39 , 1965–1982.
McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean–Global Atmosphere observing system: A decade of progress. J. Geophys. Res., 103 , 14169–14240.
Negri, A. J., T. L. Bell, and L. Xu, 2002: Sampling of the diurnal cycle of precipitation using TRMM. J. Atmos. Oceanic Technol., 19 , 1333–1344.
Serra, Y. L., and M. J. McPhaden, 2003: Multiple time- and space-scale comparisons of ATLAS buoy rain gauge measurements with TRMM satellite precipitation measurements. J. Appl. Meteor., 42 , 1045–1059.
Serra, Y. L., and M. J. McPhaden, 2004: In situ observations of diurnal variability in rainfall over the tropical Pacific and Atlantic Oceans. J. Climate, 17 , 3496–3509.
Serra, Y. L., P. A’Hearn, H. P. Freitag, and M. J. McPhaden, 2001: ATLAS self-siphoning rain gauge error estimates. J. Atmos. Oceanic Technol., 18 , 1989–2002.
Simpson, J., R. F. Adler, and G. R. North, 1988: A proposed Tropical Rainfall Measuring Mission (TRMM) satellite. Bull. Amer. Meteor. Soc., 69 , 278–295.
Wentz, F. J., 1997: A well-calibrated ocean algorithm for special sensor microwave/imager. J. Geophys. Res., 102 , 8703–8718.
Wentz, F. J., and R. W. Spencer, 1998: SSM/I rain retrievals within a unified all-weather ocean algorithm. J. Atmos. Sci., 55 , 1613–1627.
Wilheit, T. T., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol., 8 , 118–136.
WMO, 1962: Precipitation measurements at sea. Tech. Rep. 47, World Meteorological Organization No. 124.TP.55, 18 pp.
Yang, D., B. E. Goodison, J. R. Metcalfe, V. S. Golubev, R. Bates, T. Pangburn, and C. L. Hanson, 1998: Accuracy of NWS 8″ standard nonrecording precipitation gauge: Results and application of WMO intercomparison. J. Atmos. Oceanic Technol., 15 , 54–68.
Locations of selected buoys with rain gauge data. For reference, the buoys are numbered from west to east and north to south.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Data quality flags as a function of time for the selected buoys used in this study. The buoys are sorted from west to east and from north to south. On the lhs are the buoy reference numbers and the total number of days with data in the complete record for each buoy. On the rhs are the buoy longitudes and latitudes.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Time series of 10-min rain-rate values from buoy 20 at 0°, 110°W for 6 yr. Blue represents 10-min rain rate; the red lines represent 5-day running mean of negative rain rates only, plotted both positively and negatively to give a qualitative estimate of the “envelope” of the noise; colored bars represent quality flags colored as in Fig. 2. To be able to see noise during low-noise periods, the scale of the ordinate is expanded to show only the range between −2 and +2 mm h−1.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Scatterplot of time-mean rain rates at each gauge location as measured by the buoy rain gauge and two different satellite instruments. The gray diagonal line is the one-to-one line. (top) TRMM TMI and (bottom) F14 SSM/I containing (left) GPROF values and (right) RSS values. Because the samples for the two satellite retrievals are not identical, there are small differences in the buoy rain-rate estimates for a given satellite. Periods of comparison are December 1997–October 2006 for TRMM TMI and November 2001–October 2006 for F14.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Scatterplot of climatological RSS vs GPROF TMI rain rates. Each point represents one 0.5° × 0.5° grid box. The dashed diagonal line is the one-to-one line. The gray line is a least squares linear fit to the data. The black contours are the density of points binned into 0.25 mm day−1 × 0.25 mm day−1 bins. The contour interval is logarithmic. Most of the points are at relatively low rain rates (<6 mm day−1).
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Map of climatological differences between the RSS and GPROF rainfall retrievals (RSS − GPROF) for the period December 1997–October 2006. Values are not plotted for latitudes between ∼30° and 35° in each hemisphere because of differences in the gridding schemes.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Scatterplot of the long-term climatological TRMM 3A11 and 3A12 rain-rate estimates for 5° × 5° longitude–latitude boxes.
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Map of climatological differences between the TRMM 3A11 and 3A12 rainfall retrievals (3A11 − 3A12).
Citation: Journal of Applied Meteorology and Climatology 48, 7; 10.1175/2009JAMC2149.1
Impacts of low-quality (LQ) data on long-term time averages. Values are in millimeters per day. This table includes all gauges with at least 6-month records that reported LQ data. “All data” is the time mean using all available data; “Omit LQ” is the time mean omitting data flagged as low quality; “Diff” and “% diff” are the absolute and relative differences, relative to “All data”; “Tot obs” is the total number of 10-min observations for each buoy; “LQ obs” is the number of observations flagged as low quality; and “% LQ obs” is the percentage of observations flagged as low quality.
Comparisons of satellite retrievals with buoy gauge data. The values given are the intercept a (mm day−1) and slope b (dimensionless) of the least squares fit to a straight line and their 95% confidence intervals. “Months” is the number of months of satellite data available for each satellite. This is the upper bound on the number of months of data in the time means.
