A 10-yr Climatology of Amazonian Rainfall Derived from Passive Microwave Satellite Observations

Transcription

1 42 JOURNAL OF APPLIED METEOROLOGY A 10-yr Climatology of Amazonian Rainfall Derived from Passive Microwave Satellite Observations ANDREW J. NEGRI Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, Maryland EMMANOUIL N. ANAGNOSTOU* Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, Maryland, and Universities Space Research Association, Seabrook, Maryland ROBERT F. ADLER Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, Maryland (Manuscript received 1 October 1998, in final form 3 May 1999) ABSTRACT In this study, a satellite-derived precipitation climatology (climate description) over northern South America using a passive microwave technique, the Goddard Profiling algorithm, is presented. The results are statistically adjusted to have the same probability distribution as a rain gauge dataset. The climatologies take the form of the mean estimated rainfall for a 10-yr period, with subdivisions by month and meteorological season. For the 6-yr period , when two satellites were in operation, diurnal variability (to the extent it is discerned by four unequally spaced observations) is presented. In the mean, dramatic patterns of alternating morning and evening maxima are seen stretching from the northeast (Atlantic coast) across the continent to the Pacific. The effects of local circulations caused by topography, coastlines, and geography (river valleys) on the rainfall patterns are evident, particularly in the region around Manaus, Brazil, where the Negro and Solimoes Rivers merge. The interannual variability of the 10-yr rainfall estimate is examined by computing the deviations of yearly and warm-season (December February) rainfall from their respective long-term means. Rainfall anomalies associated with El Niño and La Niña events then become apparent. This gauge-adjusted satellite climatology enhances existing (gauge based) climatologies by increasing the spatial resolution and providing a common, spaceborne platform for assessing interannual variability. It maintains the same first- and second-order statistics as does the gauge dataset, and allows a first attempt at examining the diurnal cycles by utilizing passive microwave observations (up to) four times per day. 1. Introduction It has been more than a decade since the launch of the first Special Sensor Microwave Imager (SSM/I). This passive microwave sensor flies aboard the Defense Meteorological Satellite Program (DMSP), a series of polar-orbiting sun-synchronous meteorological satellites. The long record of data enables us to process and present a 10-yr climate description ( climatology ) of rainfall derived from the SSM/I observations via application of the Goddard Profiling (GPROF) algorithm. * Current affiliation: Department of Civil and Environmental Engineering, University of Connecticut, Storrs, Connecticut. Corresponding author address: Andrew J. Negri, NASA/GSFC, Code 912, Greenbelt, MD negri@agnes.gsfc.nasa.gov GPROF (Kummerow et al. 1996) is a physical retrieval technique that produces instantaneous rainfall estimates at 15-km resolution. These estimates are then aggregated into monthly composites and averaged to 0.5 resolution. The monthly estimates are adjusted by the technique of Anagnostou et al. (1999b), a power-law correction based on the evaluation of the distortion 1 between the probability density functions of the monthly rain gauge rainfall (the truth ) and the GPROF estimates. This combination of satellite observations and gauge measurements produces monthly rainfall fields with small bias, despite the inadequacies of the satellite algorithm and the satellite sampling. 1 According to Anagnostou et al. (1999b), when two instruments measure the same variable with varying degrees of accuracy, the less reliable measurement is assumed to be a distortion of the other measurable.

2 JANUARY 2000 NEGRI ET AL. 43 The area of interest in this study is northern South America, specifically the rain forests of Amazonia and the drier coastal regions of Northeast Brazil. The current study extends the work of Negri et al. (1994) who examined warm-season Amazonian precipitation using a scattering-based technique, the Goddard Scattering (GSCAT) algorithm. Over land, GSCAT is nearly identical to full GPROF retrieval. The climatologies presented here take several forms. R The mean estimated (adjusted) rainfall for July 1987 February 1998 is presented with subdivisions by month and 3-month season. R Unadjusted estimates are stratified by time of satellite overpass. At least two observations (0600 and 1800 LST) are available for the entire period. For the 6-yr period January 1992 November 1997, two satellites were in operation, providing up to four observations per day. Diurnal variability (to the extent it is discerned by four unequally spaced observations) is presented. R Interannual variability is presented in the form of deviations from the 10-yr GPROF mean rainfall, both yearly and for December February. The strong El Niño event of is identified. The microwave estimates can provide what previous studies lack: monthly rainfall at high spatial resolution (for small-scale effects on rainfall), observations up to four times per day (for a first attempt at studying the diurnal cycle), and a consistent platform from which to study interannual variation. The climatologies will show that the circulations associated with the topography and geography (coastlines, river valleys) of the region have a major impact on the rainfall distribution. 2. Previous studies The most comprehensive study of Amazonian rainfall using gauges was done by Figueroa and Nobre (1990), who presented both seasonal and yearly analyses using a database of 226 stations with records greater than 7 yr. They related the precipitation patterns to average values of the seasonal circulation and looked at the pronounced effect of topography on the rainfall. They report a maximum in yearly precipitation (3635 mm) in the Amazon Basin proper at a station near the equator and 68 W. Other maxima of note were found along the west coast of Colombia (6047 mm) near 4 N, 78 W and in the mountains of Peru (7443 mm) near 14 S, 72 W. Kousky (1980) investigated the period in Northeast Brazil using surface observations and is one of the few researchers to examine the diurnal cycle of rainfall using hourly data. That study revealed that coastal areas experience a nocturnal maximum in rainfall due to convergence of onshore flow and offshore land breeze. It also showed that inland ( km) experienced a daytime maximum, associated with the development of an inland advance of the sea breeze. Kousky also found diurnal variability at inland locations to be due to mountain valley breezes. Rao et al. (1993) using station data from determined the rainy and dry seasons of eastern Northeast Brazil and examined the link between rainfall and sea surface temperatures. Using three years of 3-hourly infrared geosynchronous satellite data, Meisner and Arkin (1987) examined the diurnal cycle of tropical convective cloudiness over the Americas. Among their findings relevant to this study were 1) there is far more diurnal variance over the continents than over oceans; 2) local maxima of diurnal variance were found in areas of strong relief, or where convergent land sea breezes are common; and 3) there is a pronounced maximum in cloudiness at 1800 LST over the interior of South America during the austral summer. Horel et al. (1989) used a 15-yr record of outgoing longwave radiation to describe the annual cycle of convection that resides over the Amazon Basin during the austral summer. They found that the onset of the Amazon Basin wet season typically occurs within a single month (sometimes a 5-day period) and that the onset may be blurred by monthly averaged climatologies. Also relevant to this paper is the work of Garreaud and Wallace (1997) who used nine years of Geostationary Operational Environmental Satellite infrared data to define the diurnal march of convective cloudiness over the tropical Americas and surrounding waters. In the process, they favorably compared their infrared results with the microwave-estimated rainfall of Negri et al. (1994). Regional climatologies using passive microwave data are just now coming to the fore. For examples see Negri et al. (1993) for a high-resolution climatology of warm-season rainfall over Mexico and the southwest United States, and Negri et al. (1994) who examined four regions: the southern United States and Mexico, northern South America, equatorial Africa, and the western Pacific (Tropical Ocean Global Atmosphere Coupled Ocean Atmosphere Research Experiment) region. Much research has also been done on the relationship between large-scale precipitation patterns and the El Niño Southern Oscillation (ENSO). Ropelewski and Halpert (1987) found consistent departures of area-averaged rainfall during ENSO events. Relevant to this study, they found that the coastal region of northern South America, from Venezuela east to Brazil, had one of the most consistent ENSO precipitation relationships of any area studied. Sixteen dry episodes (of 17 ENSOs) were found and the period of deficiency was determined to be from July of the ENSO onset year to March of the following year. This relationship was further quantified in Ropelewski and Halpert (1996). Janowiak and Arkin (1991) observed large-scale shifts in tropical rainfall during the transition from warm ( ) ENSO conditions to cold ( ). An area near the mouth of the Amazon River at Belem, Brazil, stood

3 44 JOURNAL OF APPLIED METEOROLOGY TABLE 1. Data used in this study. Combination no. DMSP satellites Data period Length (month) Morning equatorial crossing (LST) F8 Jul 1987 Feb 1990* F11 Jan 1992 Dec (0545 nominal) F13 Jan 1996 Feb F10 Jan 1991 Nov (0900 nominal) 1 F8/11/13 Jul 1987 Feb 1990* and 95 (0600 nominal) Jan 1992 Feb F8/11/13/10 Jul 1987 Feb 1990* and and 0900 (all data) 3 F8/11/13/10 (common period) * Except Dec Jan 1991 Feb 1995 (nominal) Jan 1992 Nov and 0900 (nominal) out as having a marked difference in mean precipitation between these two periods. 3. Data and methodology The microwave data used in this study come from the SSM/I instrument on board the DMSP satellites designated F8, F10, F11, and F13. The SSM/I is a sevenchannel passive microwave radiometer with horizontally (H) and vertically (V) polarized frequencies at (H and V), (V), 37.0 (H and V), and 85.5 (H and V) GHz. Multifrequency, false-color displays of microwave imagery from the SSM/I may be found in Negri et al. (1989). The satellites are in sun-synchronous, circular (except F10) nearly polar orbits. A complete description of the SSM/I is given in Hollinger et al. (1987). With a swath width of 1400 km, coverage of the entire globe is not possible in a 24-h period. At the equator, monthly sampling of a given point is 30 observations, subdivided into 15 morning and 15 evening overpasses. The sampling has the potential for introducing both bias and random error into the rain estimates. Table 1 lists some of the orbit characteristics of each satellite, as well as the period and length of the usable data. Three combinations of the data are also listed in Table 1. Data from the F8/11/13 series constitute a 95- month (noncontinuous) period of up-to-twice-a-day observations. Including observations from F10 extends the period of usable data but introduces observations approximately 3 h later than the other series. Finally, a 71-month set of common data is available between January 1992 and November 1997, allowing a comparison of mean precipitation at four distinct times. The estimates cover the northern half of South America from 15 N to20 Sand from 80 W to30 W. The rainfall estimates derived herein are produced by the GPROF algorithm (Kummerow et al. 1996), a computationally simple technique for retrieving both precipitation and vertical hydrometeor profiles. GPROF is an extension of the original multichannel retrieval of Kummerow and Giglio (1994). Over land surfaces, GPROF has a greater reliance upon the 85-GHz channel than does the retrieval of Kummerow and Giglio, resulting in estimates largely based on scattering by precipitating ice hydrometeors. Designed with the operational goals of the Tropical Rain Measuring Mission in mind, it attempts to meet two essential requirements. The first is computational efficiency: one month of global SSM/I data can be processed on a Silicon Graphics workstation in 24 h. Second, individual processes within the retrieval are clearly separated, so that their effect upon the result is both predictable and quantifiable. One shortcoming of GPROF is its apparent overestimate of the intensity of light and moderate stratiform rain (Kummerow et al. 1996). GPROF produces instantaneous rain-rate estimates at 15-km resolution. These are then aggregated to monthly totals and averaged to 0.5 grid boxes. The monthly GPROF estimates are then adjusted by the statistical technique of Anagnostou et al. (1999b). This correction takes the form of a power law: GPROF adj a(gprof unadj ) b, where (a, b) (7.2, 0.67) for the 0600 and 1800 LST observations and (a, b) (13.8, 0.58) for the 0900 and 2100 LST observations. This correction is important for two reasons. First, it minimizes the uncertainty (error) in the satellite rainfall estimates. Anagnostou et al. define uncertainty as the distortion in the probability distribution of the true surface rainfall caused by errors introduced by the satellite estimation. This correction at a local time forces the distribution of GPROF rainfall estimates to match that of the monthly gauge rainfall. Second, the adjustment provides a consistent way of merging rain estimates from different DMSP satellites, compensating for biases introduced by differences in the time of satellite overpass. One note of caution: the correction was applied universally to the area, but since the adjustment was derived only over land, results over ocean must be viewed qualitatively. Based on the parameter values noted above, Anagnostou et al. conclude

4 JANUARY 2000 NEGRI ET AL. 45 FIG. 1. Scatterplot of GPROF gauge-adjusted estimates vs monthly raingauge observations over the Amazon Basin (dark circles) and the northeast Brazilian state of Ceara (open circles). that the diurnal cycle of precipitation is the dominant error factor in the satellite microwave monthly rainfall estimates. The diurnal cycle is examined more fully using a combination infrared microwave technique described by Anagnostou et al. (1999a). A scatterplot of GPROF gauge-adjusted estimates versus monthly raingauge observations (both area averaged over 5 ) is shown in Fig. 1. Plotted data include gauges from both the Amazon and eastern Brazil (state of Ceara) during The Ceara dataset is independent of the adjustment procedure; the Amazon data are quasi-independent. The plot shows that the overall bias of the adjusted GPROF estimates is close to zero ( 1.3% of the mean), and the estimates are highly correlated (0.84) with the independent gauges. Considerable scatter still exists at rain rates above 200 mm month 1. The root-mean-square error is 28.2% of the mean, significantly lower than the rms of 41% from the unadjusted estimates. More detailed statistics are available in Anagnostou et al. (1999b). 4. Results a. Mean rainfall Figure 2 presents the mean GPROF rainfall (mm month 1 ) during the 107-month period of usable data between July 1987 and February Monthly estimates from both series of satellites were gauge-adjusted separately, then combined, to form this product. The figure also displays contours of 500-m elevation. Several major rainfall features can be discerned. The mean position of ITCZ oceanic rainfall, with a maximum exceeding 3myr 1, is found along 5 N from 50 W to 30 W, with a sharp gradient along its northern boundary. A pronounced maximum extends inland from the Gulf of Panama, near 78 W from 7 N to the equator. This region is quite possibly one of the rainiest places on the globe, the result of both the concave shape of the coast (increasing offshore convergence) and the proximity of the northern extent of the Andes. This serves as both a barrier to the low-level flow as well as being an elevated heat source in its own right. In the Amazon Basin itself, at least the portion to the west of Manaus (3 S, 60 W) there is a broad rainfall maximum in excess of 200 mm month 1, which decreases eastward. This is in general agreement with the gauge analyses of Figueroa and Nobre (1990) and Salati (1987). Molion (1987) attributes this maximum to the result of a thermally driven circulation over this heated region. An interesting and distinct maximum appears at (2 N, 67 W), between the Orinoco River (at the base of the Sierra Pacaraima) and the Negro River. With a high plateau just to the northeast, this is most likely the result of a mountain valley circulation of the type suggested by Kousky (1980). The analysis of Salati (1987) shows a maximum of 300 mm month 1 to the southwest of this satellite-estimated maximum but is, unfortunately, at the extreme western end of that analysis. Near the highest peaks of the Sierra Pacaraima (5 N, 63 W) is a topographically induced maximum ( 250 mm month 1 ), which possibly has not been identified in previous climatologies. Farther to the northeast, we find a coastal maximum at the mouth of the Amazon River, near Belem (2 S, 48 W). This is displaced slightly eastward with respect to the gauge-analyzed maximum reported by Molion (1987). An elongated rainfall maximum is found along the eastern slope of the Andes from 10 S, 75 Wto20 S, 65 W just westward of the 500-m height contour. Note that in the higher elevations of the Andes, we find areas where GPROF cannot distinguish between cold surface and rainfall, and these areas are flagged (whitened). The driest ( 50 mm month 1 ) regions are found in extreme northeastern Brazil and along the southern Pacific coast. In Fig. 3, we present the gauge-adjusted GPROF estimates aggregated by month. Each panel represents the average of 9 10 yr of data for that particular month and includes data from all available satellites. We find that the oceanic ITCZ rainfall progresses southward from its northernmost position in October toward the northeast coast of Brazil in March April and retreats northward during May. The strong rainfall maximum in the Gulf of Panama (5 N, 78 W) persists virtually year-round but weakens during January to March. The local maximum at the mouth of the Amazon (0, 50 W) is a persistent feature from January to May. During May through August, much less rain is received in the portion of the Amazon Basin south of the Amazon River. A maximum in the Sierra Pacaraima of Venezuela (5 N, 63 W) forms in April and persists through December, reaching its maximum intensity in June. The changes in the areal

5 46 JOURNAL OF APPLIED METEOROLOGY FIG. 2. Mean rain rate (mm month 1 ) using all available data from Jul 1987 to Feb Amounts are derived from GPROF and have been adjusted by the method of Anagnostou et al. (1999b). Topography (500-m contour) and major rivers are overlain. Whitened areas along the Andes Mountains indicate grid boxes where a persistent cold surface makes the precipitation retrieval unreliable. extent of cold surface in the southern Andes Mountains may be seen as the increased whitened area representing ambiguous rain retrieval. This area reaches maximum extent during the austral winter months of June through August. Figure 4 presents the seasonal (3 month) climatology for the combined 0600 and 1800 and 0900 and 1200 LST series using the meteorological definition of seasons. The December February period brings the greatest rainfall to the central Amazon, as well as to an elongated area along the eastern slopes of the Andes Mountains. The rainfall maximum at Belem is most pronounced in March May, as is a maximum between the Orinoco River and Negro River near 2 N, 65 W. The topographically induced maximum in the Sierra Pacaraima of Venezuela (5 N, 63 W) is prominent during June August. Beginning in September November, and continuing through December February, a broad maximum appears in the highlands of eastern Brazil (Mato Grosso) near 10 S, 50 W. Extreme northeast Brazil is consistently the driest part of that country. This seasonal climatology lends itself to comparison with the gaugederived results of Figueroa and Nobre (1990). In location, magnitude, and orientation of the major features the agreement is very good. One place where results differ is along the eastern slopes of the Andes (just south of the equator along 78 W); Figueroa and Nobre found a more pronounced maximum in each season than does GPROF. Upslope flow, converging upon nighttime katabatic flow, might produce warm rain that would

6 JANUARY 2000 NEGRI ET AL. 47 FIG. 3. As in Fig. 2 except the data have been aggregated by month.

7 48 JOURNAL OF APPLIED METEOROLOGY FIG. 4. As in Fig. 2 except the data have been aggregated by meteorological season. be underestimated by a scattering-based technique. In the next section, we find that this eastern slope maximum is indeed most pronounced at 0600 LST. b. Diurnal variation The diurnal variation of rainfall is best studied with hourly rain gauge data or even half-hour interval (but less accurate) infrared rain estimates (Anagnostou et al. 1999a). A first attempt at identifying the diurnal signal may be obtained by examining the unadjusted GPROF estimates from two satellites. Figure 5 (upper four panels) presents the rainfall climatology by time of satellite overpass for the 71-month period of two-satellite operation, January 1992 to November 1997 (Table 1, combination 3). It was felt that the effects of internannual variability could be reduced by using the same period of record for each overpass time. Results do not substantially change if all the available data are used (lower two panels). Note the unequally spaced observation times of 0600, 0900, 1800, and 2100 LST. The following diurnal features are noted: R a morning maximum in the Gulf of Panama (5 N, 78 W); R a preference for increased morning oceanic rainfall in the ITCZ (5 N, 35 W), more pronounced at 0600 than at 0900 LST; R a morning (both 0600 and 0900 LST) maximum offshore along the northeast Brazilian coast, including a

8 JANUARY 2000 NEGRI ET AL. 49 FIG. 5. Mean rain rate (mm month 1 ) derived from GPROF, without gauge adjustment, and stratified by the nominal time of satellite overpass. Areas of ambiguous rain retrieval are whitened. Data from the 71-month period of two-satellite operation, Jan 1992 Nov 1997, are shown in the upper four panels. For comparison, the 0600 and 1800 LST composites for the entire period of data (Jul 1987 Feb 1998) are displayed in the lower two panels. Outlined area over the Amazon River is shown in more detail in the next figure.

9 50 JOURNAL OF APPLIED METEOROLOGY FIG. 6. Same as Fig. 5 except a blowup of a portion of the Amazon Basin, with superposition of detailed surface hydrological features. strong signal near the mouth of the Amazon River (0, 50 W), possibly due to increased convergence in that uniquely shaped region of the coast; R an 1800 LST maximum just onshore along virtually the entire northern coast of South America; this band moves inland by 2100 LST, the likely result of synoptic flow in combination with a pronounced seabreeze circulation; R an 0600 LST maximum between the Negro and Orinoco Rivers (2 N, 67 W), possibly related to morning convergence associated with the mountains to the southeast of the maximum; R an 1800 LST maximum in eastern Brazil along the rivers of the Mato Grosso plateau (10 S, 50 W); R a morning maximum (both at 0600 and 0900 LST) along the eastern slopes of the Andes Mountains; and R an 1800 LST maximum between the Negro and Solimöes Rivers near 2 S, 63 W, with a weaker maximum at 0900 LST just downriver from their confluence. The 1800 LST maximum is part of a broad region of late afternoon rainfall in the Amazon Basin, but the effects of coastal squall lines and local circulations along the river cannot be discounted. This last area is examined in detail in Fig. 6, a blowup of a portion of the Amazon River Basin superimposed with detailed surface hydrological features. At 0900 LST there is a broad area of 100 mm month 1 rainfall along the main river valley of the Amazon. One finds a minimum ( 50 mm month 1 ) between the aforementioned Negro and Solimöes Rivers (3 S, 63 W) and a maximum of 252 mm month 1 just south of the river at 3.5 S, 58 W. At 1800 LST the reverse is true, with a large area of 300 mm month 1 rainfall between the rivers, and an elongated minimum of less than 50 mm month 1 downstream. This dry region is also evident at 2100 LST. Interestingly, at 1800 LST, relative minima are found along the Negro and Solimöes Rivers. This could be the result of a local circulation, whereby the cooler river inhibits the convection. Future iterations of GPROF done at the full 15-km resolution of the data may aid in our understanding of the complicated interactions in this area. An alternate way of presenting the diurnal cycle is shown in Fig. 7, by creating the difference between the unadjusted GPROF estimates at 1800 and 0600 LST (top panel), during the period January 1992 November The ( LST) difference is shown in the bottom panel of Fig. 7. Cross section AB is defined for use in Fig. 8. The result is an alternating pattern of morning (blue shading) and evening (red shading) maxima stretching from the northeast (Atlantic coast) clear across the continent to the Pacific. Beginning at the northeast coast, the offshore morning (0600 and 0900 LST) maximum has its evening

10 JANUARY 2000 NEGRI ET AL. 51 FIG. 7. (top) Difference between the unadjusted GPROF estimates at 1800 and 0600 LST, each averaged over the period Jan 1992 Nov (bottom) Same as in top panel except difference between 2100 and 0900 LST. Blue shades indicate morning maxima, red shades indicate afternoon maxima. Cross section AB is defined for use in Fig. 8.

11 52 JOURNAL OF APPLIED METEOROLOGY the analysis represents the mean rainfall over many years, the rainbands along the coast are certainly due to the formation of squall lines as described by Garstang et al. (1994). In Fig. 8 we plot unadjusted GPROF estimates along the cross section AB in Fig. 7 (top panel), for the 0600 and 1800 LST overpass times. The values plotted represent the average rain rate over three grid boxes (1.5 ) on either side of the line AB for the years A cross section of topography is shown in the bottom panel of Fig. 8. Beginning at the west coast ( A ), we find a pronounced morning maximum of 330 mm month 1 just east of the Andes Mountains. (The afternoon maximum in the mountains is masked by the cold surface problem.) Proceeding across the Amazon Basin, we find broad maxima in both the 0600 and 0900 LST estimates, with the afternoon estimates consistently higher. As we approach the Amazon River, we find that the morning and afternoon estimates are out of phase and remain that way as we reach the coast. There we find a pronounced onshore afternoon maximum of 200 mm month 1 (with virtually no morning rainfall). Finally, the morning maximum offshore is noted. A slight increase in elevation may also contribute to this diurnal circulation. FIG. 8. Unadjusted GPROF estimates along the cross section AB in Fig. 7, at 0600 (triangles) and 1800 LST (squares), for A cross section of topography is shown (bottom panel), with the location of the Amazon River identified. counterpart inland, with apparent propagation of the previous day s rainfall even farther inland (the secondary morning maxima roughly paralleling the coast). This alternation is less distinct in central Amazonia, where features related to the distribution of rivers seem to dominate: note the 1800 LST maximum at 2 S, 63 W (between the rivers) and the isolated 0900 LST maximum just downriver at 2 S, 58 W. On the Pacific coast meanwhile, we find a pronounced evening morning evening sequence as one moves northeastward from the coast. This may be related to the mountain valley (or range plain) circulation. The strong oscillation in the Gulf of Panama (5 N, 80 W), with its broad oceanic morning maxima and inland evening maxima (along the mountains) stands out, with secondary morning maxima just east of the mountains, southward from 10 N, along 75 W. It is worth noting that a simple characterization of Amazon Basin rainfall as having a late-afternoon maximum is only true in that portion south of the Amazon River. The pattern in Fig. 7 suggests an analogy to that of waves in a string fixed at both ends. In this case, the ends are the Andes to the southwest and the land ocean boundary to the northeast. Both ends generate convection that propagates into the Amazon Basin, initiating new convection, and interacting with convection initiated within the basin itself, all apparently complicated by the effects of the rivers themselves. Although c. Interannual variability In the next two figures, we attempt to assess the interannual variability of the GPROF estimates. Figure 9 is the adjusted GPROF estimates for individual years minus the mean estimate for the 8-yr period and No estimates were available in Though somewhat noisy, the strongest signal is the decrease in Amazonian and northern Brazilian (coastal) rainfall during the 1992 and 1997 El Niño events (large regions of blue shading). Increases in precipitation across most of northern South America appeared during the La Niña years 1988 and 1989 (large regions of red shading). The interannual variability of precipitation on smaller timescales is presented in Fig. 10 for the months of December through February (DJF), the beginning of the rainy season in Amazonia. No data are available for DJF or DJF In any 3-month estimate near the equator, there are only about 90 samples of data, so some caution should be taken in interpreting the results. Despite this, we find evidence of both the increased DJF precipitation during the La Niña event of and the decrease during the strong El Niño event of Increased rainfall in northern Amazonia was evident in DJF (not an ENSO period). 5. Summary and conclusions In this study, we presented a satellite-derived, gaugeadjusted precipitation climatology over northern South America using a microwave technique, the Goddard

12 JANUARY 2000 NEGRI ET AL. 53 FIG. 9. Difference of the gauge-adjusted GPROF estimates for individual years from the mean estimate of the 8-yr period and Only data from the F8/11/13 series are used, with nominal overpass times of 0600 and 1800 LST. Because of incomplete data, no estimates were possible for the years

13 54 JOURNAL OF APPLIED METEOROLOGY FIG. 10. As in Fig. 9 except estimates are for the 3-month period beginning each Dec.

14 JANUARY 2000 NEGRI ET AL. 55 Profiling algorithm. A period of data slightly longer than 10 years was examined. The results were statistically adjusted to have the same probability distribution as the gauges. Validation of the monthly GPROF estimates was performed using quasi-independent rain gauges. The gauge-adjusted estimates had bias close to zero and were highly correlated (0.84) with the gauges. Considerable scatter existed at rain rates above 200 mm month 1 and the rmse was 28.2% of the mean. The climatologies took the form of the mean estimated (adjusted) rainfall for a 10-yr period, with subdivisions by month and meteorological season. Through higher spatial resolution and the benefit of a common observing system, these satellite-based climatologies complement and expand upon existing gauge climatologies. For the 6-yr period , when two satellites were in operation, diurnal variability was discerned by four unequally spaced observations of the unadjusted estimates. Interannual variability (i.e., deviations from the 10-yr GPROF mean rainfall) was presented as both yearly and seasonal (DJF) deviations, and included the strong El Niño event of The analysis showed a preference for morning oceanic rainfall (ITCZ), notably at 0600 LST, off the northern Brazilian coast. A rainfall maximum was observed offshore in the Gulf of Panama, exclusively a morning phenomenon, and was most intense at 0900 LST. With estimated rainfall in excess of 3 m, and evident for nine months of the year, this may be one of the wettest spots on the planet. We also found a persistent local maximum at the mouth of the Amazon during January to May, possibly due to increased convergence in that uniquely shaped region of the coast. A rainfall maximum at 1800 LST can be found along virtually the entire onshore side of the northern coast of South America. This moves inland by 2100 LST, the apparent result of a pronounced sea-breeze circulation. Attendant to this is a morning (0600 and 0900 LST) maximum offshore along the northeast Brazilian coast. In the Amazon Basin itself, we find a broad maximum in excess of 200 mm month 1 in west-central Amazonia and decreasing eastward. Several rainfall maxima are the apparent results of local circulations as suggested by Kousky (1980). These include an elongated (morning) maximum along the eastern slopes of the Andes Mountains and a maximum between the Negro and Orinoco Rivers at 2 N, 67 W, the possible result of a mountain valley circulation. A topographically induced afternoon maximum in the mountains of Venezuela (Sierra Pacaraima) near 5 N, 63 W was observed during May through September. A prominent feature was the 1800 LST rainfall maximum between the Negro and Solimoes Rivers, where they come together to form the Amazon River. There is a suggestion of a weak 0900 LST maximum just downriver from this confluence. There was an 1800 LST maximum in eastern Brazil along the rivers of the elevated Mato Grosso plateau. The interannual variability of the GPROF precipitation estimates was examined. Increased precipitation over Amazonia and northern (coastal) Brazil during DJF was evident during the La Niña event of and decreases appeared during the strong El Niño event of Increased rainfall in northern Amazonia was evident in DJF as well, a year with little ENSO signal. A subsequent effort (Anagnostou et al. 1999a) describes an infrared technique tuned to the instantaneous microwave results that allows the full diurnal cycle to be resolved. Acknowledgments. The authors thank Dr. George Huffman for his critical review of the manuscript, as well as the anonymous reviewers for their helpful suggestions. REFERENCES Anagnostou, E. N., A. J. Negri, and R. F. Adler, 1999a: A satellite infrared technique for diurnal rainfall variability studies. J. Geophys. Res., in press.,, and, 1999b: Statistical adjustment of satellite microwave monthly estimates over Amazonia. J. Appl. Meteor., 38, Figueroa, S. N., and C. A. Nobre, 1990: Precipitation distribution over central and western tropical South America. Climanálise, 5, Garreaud, R. D., and J. M. Wallace, 1997: The diurnal march of convective cloudiness over the Americas. Mon. Wea. Rev., 125, Garstang, M., H. L. Massie Jr., J. Halverson, S. Greco, and J. Scala, 1994: Amazon coastal squall lines. Part I: Structure and kinematics. Mon. Wea. Rev., 122, Hollinger, J., R. Lo, G. Poe, R. Savage, and J. Pierce, 1987: Special Sensor Microwave/Imager user s guide. Naval Research Laboratory, Washington, DC, 177 pp. Horel, J. D., A. N. Hahmann, and J. E. Geisler, 1989: An investigation of the annual cycle of convective activity over the tropical Americas. J. Climate, 2, Janowiak, J. E., and P. A. Arkin, 1991: Rainfall variations in the tropics during as estimated from observations of cloud-top temperature. J. Geophys. Res., 96, Kousky, V. E., 1980: Diurnal rainfall variation in northeast Brazil. Mon. Wea. Rev., 108, Kummerow, C., and L. Giglio, 1994: A passive microwave technique for estimating rainfall and vertical structure information from space. Part I: algorithm description. J. Appl. Meteor., 33, 3 18., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, Meisner, B. N., and P. A. Arkin, 1987: Spatial and annual variations in the diurnal cycle of large-scale tropical convective cloudiness and precipitation. Mon. Wea. Rev., 115, Molion, L. C. B., 1987: On the dynamic climatology of the Amazon Basin and associated rain-producing mechanisms. Geophysiology of Amazonia, R. E. Dickinson, Ed., John Wiley and Sons, Negri, A. J., R. F. Adler, and C. D. Kummerow, 1989: False-color display of Special Sensor Microwave/Imager (SSM/I) data. Bull. Amer. Meteor. Soc., 70, ,, R. A. Maddox, K. W. Howard, and P. R. Keehn, 1993: A regional rainfall climatology over Mexico and the southwest

15 56 JOURNAL OF APPLIED METEOROLOGY United States derived from passive microwave and geosynchronous infrared data. J. Climate, 6, ,, E. J. Nelkin, and G. J. Huffman, 1994: Regional rainfall climatologies derived from Special Sensor Microwave Imager (SSM/I) data. Bull. Amer. Meteor. Soc., 75, Rao, V. B., M. C. delima, and S. H. Franchito, 1993: Seasonal and interannual variations of rainfall over eastern northeast Brazil. J. Climate, 6, Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation. Mon. Wea. Rev., 115, , and, 1996: Quantifying Southern Oscillation precipitation relationships. J. Climate, 9, Salati, E., 1987: The forest and the hydrological cycle. Geophysiology of Amazonia, R. E. Dickinson, Ed., John Wiley and Sons,