Spatial and temporal rainfall variability near the Amazon-Tapajós confluence

Authors


Abstract

[1] Do the influences of river breezes or other mesoscale effects lead to a systematic river proximity bias in Amazon rainfall data? We analyzed rainfall for a network of 38 rain gauges located near the confluence of the Tapajós and Amazon rivers in the eastern Amazon Basin. Tipping bucket rain gauges worked adequately in the Amazon rainfall regime, but careful field calibration and comparison with collocated conventional rain gauges were essential to incorporate daily totals from our array into regional maps. Stations very near the large rivers miss the afternoon convective rain, as expected if a river breeze promotes subsidence over the river, but paradoxically, this deficiency is more than compensated by additional nocturnal rainfall at these locations. The NOAA Climate Prediction Center (CPC) Morphing technique (CMORPH) passive infrared inferred rainfall data do an adequate job of describing medium scale variability in this region, but some localized breeze effects are not resolved at 0.25° resolution. For areas inland from the rivers, nocturnal rainfall contributes less than half of total precipitation. A large-scale rainfall increase just to the west of Santarém manifests itself locally as a ‘tongue' of enhanced rain from along the wide area of open water at the Tapajós-Amazon confluence. The Amazon River breeze circulation affects rainfall more than does the Tapajós breeze, which moves contrary to the predominant wind. East of the riverbank, the effects of the Tapajós breeze extend only a few kilometers inland. Rainfall increases to the north of the Amazon, possibly the result of uplift over elevated terrain. Dry season rainfall increases by up to 30% going away from the Amazon River, as would be expected given breeze-induced subsidence over the river.

1. Introduction

1.1. Background

[2] Attempts to determine the hydrological balance observationally for the Amazon Basin as a whole, the world's largest rain forest ecosystem, have been hampered by uncertain estimates in each of the components of the hydrological balance: precipitation, interception, runoff, evaporation and advection. Moisture inflow from model reanalysis fails to account for as much as 50% of the estimated precipitation [Marengo, 2006]. Even in regions with a dense observational network as in the United States, the observational water balance fails to account for a large amount of water substance [e.g., Roads et al., 2002]. Precipitation is the most commonly measured component of the hydrological budget. Because of the patchy nature of convective rainfall, assembling regionally representative precipitation measurements remains a challenge. A significant systematic error in Amazon rainfall estimates could limit the utility of using the climate station record in assessing results of regional climate models, alter the forcing of ecosystem models [e.g., Botta et al., 2002], limit understanding of the role of drought in forcing ecosystem change, and impair assessments of the agricultural viability of the region.

[3] There is a clear systematic bias in the location of precipitation observing stations in the Amazon Basin. Since there are few roads in this region (Figure 1a), climate stations are near settlements along the rivers' banks (Figure 1b). Local mesoscale circulations near the 5–20 km wide rivers of the Amazon include the river breeze [Oliveira and Fitzjarrald, 1993, 1994; Silva Dias et al., 2004] and the ‘vegetation breeze', more frequently modeled than observed [D'Almeida et al., 2007]. The vegetation breeze, believed to result from circulations initiated by thermal contrasts at land cover change boundaries [Silva Dias et al., 2005; Ramos da Silva and Avissar, 2006] may promote cloudiness over cleared areas [Cutrim et al., 1995]. However, in areas where both effects are present, the much stronger water-land temperature contrasts mean that river breeze circulations should dominate any vegetation breeze.

Figure 1a.

Study area in South America. Santarém marked with the letter S.

Figure 1b.

Selected Brazilian Meteorological Service (INMET) climate stations in the eastern Amazon Basin for the rectangle shown in Figure 1a. T, stations along the Tapajós River; A, stations along the main channel of the Amazon River; other stations are shown with crosses. Manaus, Macapá, Belém, and Santarém stations are also shown. A solid rectangle shows the approximate location of Figures 1b, 1c, 3a, and 10. A dashed rectangle indicates the area detailed in Figures 3b, 11, and 12 (river locations from http://www.ngdc.noaa.gov/mgg coastline).

[4] Daytime subsidence-induced clearing over the river with cloudiness inland is commonly observed in satellite images (Figure 1c) [Molion, 1987]. Molion and Dallarosa [1990] examined data from four stations near Manaus for the period 1978–1988 and found that stations 100 km inland reported rainfall totals 20% higher than that observed at an island on the Negro River. They claimed similar rainfall depression observed at four stations near the Trombetas River near 56°W also resulted from breeze-induced subsidence, but in this case the explanation is not as convincing (see below). Other studies confirm that afternoon convective precipitation more typical of continental areas is observed inland from the river, with rainfall suppressed near the river [Ribeiro and Adis, 1984; Lloyd, 1990; Garstang and Fitzjarrald, 1999, p. 290; Cutrim et al., 2000]. The roughness difference between the river surfaces and the surrounding terrain causes the boundary layer wind to channel along the river course. We do not yet know whether proximity to the river influences the strength of nocturnal instability lines, a major source of precipitation in this region. Anecdotal evidence (e.g., Figure 1d) indicates that convection can even be enhanced along the Amazon River channel.

Figure 1c.

May 2001 average GOES visible low cloudiness near the Tapajós-Amazonas confluence. The horizontal dark line shows the relative clearing over the Amazon River channel; the dark area to the south is the Tapajós channel. Large dots indicate the locations of several LBA-ECO weather stations (see Figures 3a and 3c) (adapted from Moore et al. [2001] using a Google™ overlay).

Figure 1d.

Space Shuttle photo of clouds near the Tapajós-Amazonas confluence on 21 July 2003, 20:55 UT (16:55 local time). Locations of selected stations are shown. Original image ID ISS007-E-10748, available at http://ntrs.nasa.gov.

[5] For at least 45 years, it has been recognized that there is a ‘transverse dry zone' between Belém at the coast and Manaus [Reinke, 1962] (map reproduced as Figure 2 in the work of Haffer [1969]) [Pires-O'Brien, 1997] in both the dry (July–December) and rainy (January–June) seasons. Recent rainfall estimates based on observations obtained using the satellite-based CMORPH technique [Joyce et al., 2004] confirm a sharp precipitation increase just to the west of 55°W (Figures 2a and 2b). The transverse dry zone just to the east of this longitude is clearly shown, and is particularly prominent at night during the rainy season. The pronounced rainy and dry seasons in the eastern Amazon Basin [Sombroek, 2001; Mahli and Wright, 2004] make its ecosystems particularly sensitive to perturbation by prolonged drought and fire [Nepstad et al., 2004]. In the face of extensive drought, some natural forests in the Amazon may be converted to savanna [Sternberg, 2001; Oyama and Nobre, 2003], a process that could be accelerated by deforestation associated with increasing intensive agriculture in recent years [Brown et al., 2005]. The presence of the El Niño reduces precipitation in the eastern Basin. The correlation between the Southern Oscillation index [Trenberth, 1984] and rainfall anomaly is largest near 55°W, the longitude of Santarém [Zeng, 1999; Liebmann and Marengo, 2001], but the correlation coefficient (≈0.6) is so small that other factors are needed to explain the bulk of the interannual variance. South Atlantic sea surface temperatures may also play a role [e.g., Ronchail et al., 2002; Marengo et al., 2008].

Figure 2a.

November (minimum precipitation month) average CMORPH precipitation estimates (mm/d, scale at right) in the eastern Amazon Basin 2003–2006 averaged for (top) 12 UT–0 UT (daytime) and for (bottom) 0 UT–12 UT (nighttime). Locations of Manaus (MAO), Santarém (STM), Macapá (MAC), and Belém (BEL) are shown.

Figure 2b.

March (maximum precipitation month) average CMORPH precipitation estimates (mm/d, scale at right) in the eastern Amazon Basin 2003–2007 averaged for (top) 12 UT–0 UT (daytime) and for (bottom) 0 UT–12 UT (nighttime). Locations of Manaus (MAO), Santarém (STM), Macapá (MAC), and Belém (BEL) are shown.

[6] Amazon rainfall reflects contributions both from convective systems stimulated by local forcing and from organized instability lines (referred to here as ‘squall lines') that move inland from the coast [Molion, 1987; Garstang et al., 1994; Cohen et al., 1995]. The tendency toward nocturnal wet season rainfall at the longitude of Santarém, evident in Figure 2b, has often been noted [Cutrim et al., 2000; Angelis et al., 2004; Moraes et al., 2005]. The evening precipitation preference at Santarém led Nechet [1993] to describe the rainfall regime as ‘coastal', distinct from a typical inland pattern of afternoon rainfall [e.g., Lloyd, 1990].

[7] Squall lines arrive predominantly at night in Santarém, out of phase with afternoon heating. Molion [1987] postulated that this explains the transverse dry zone between Belém and Manaus. Recent analyses of satellite-based rainfall sensors confirm that the instability lines propagating inland from the coast reach the longitude of Santarém at 03–04 UT [Negri et al., 2000; Kousky et al., 2005]. We can probably associate much of the nocturnal precipitation near 55°W with the nocturnal arrival of the squall lines and afternoon precipitation with local convective forcing. The squall lines arrive in Manaus (60°W) in the afternoon, in time to be in phase with afternoon locally forced convection [Garstang et al., 1994; Lloyd, 1990; Cutrim et al., 2000].

[8] The need to assess model output and calibrate remotely sensed data has led to gridded precipitation data sets based on rain gauge records [Liebmann and Allured, 2005; Xie and Arkin, 1996]. Because of their accessibility, such processed precipitation data are widely used for comparison with model output and used to assess (“validate”) remotely sensed data products. The considerable range of annual total Amazon Basin precipitation estimates [e.g., Costa and Foley, 1998] reflects inputs from differing data sources and the use of alternate methods to grid data [e.g., Willmott and Johnson, 2005]. The gridding techniques proposed to date do not account for the singular nature of the station proximity to the great rivers of the region. In addition, many approaches use reanalysis data, which is known to underestimate both moisture inflow and precipitation in the region, in part owing to the limited sounding network in the region [Marengo, 2006].

[9] The aims of this work are (1) to introduce a new detailed precipitation data set for the region of the eastern Amazon Basin near the Tapajós-Amazon river confluence, an important location near a strong regional precipitation gradient; and (2) to identify interannual, seasonal, diurnal, and spatial patterns in regional precipitation. Documenting rainfall patterns sets the stage to determine the extent to which river breezes or other mesoscale circulations may introduce a bias in the regional rainfall climate record. If river proximity effects are systematic and properly documented, more physically reasonable explanations of precipitation patterns, spatial correlations, and interpolation procedures can be adopted. Both instrument and sampling issues must be addressed. It seems unlikely that any bias would change the conclusions of the large-scale correlation studies, but it might be critical in issues that require quantitative accuracy, such as observationally closing the hydrological balance [Marengo, 2006].

[10] Do the influences of river breezes or other mesoscale effects lead to a systematic river proximity bias in Amazon rainfall data? We document the temporal and spatial patterns of precipitation in one region likely to be affected by river proximity. We constructed a mesoscale weather station network near Santarém, Pará Brazil (2°25′48″S, 54°43′12″W) as part of the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) Ecological Component (LBA-ECO) [Keller et al., 2004]. The geometry of rivers' confluence and the fact that ‘synoptic' and ‘diurnal' precipitation contributions can be easily distinguished according to the hour of occurrence make this region singularly useful to assess mesoscale landscape influences on precipitation. We analyze rainfall observed in this network collected during the period 1998–2006, and supplement the network with data from regional operational rain gauge networks for the same period and with satellite-based microwave sensor rain estimates. Analysis of errors associated with rain gauge sensors and their siting is given in section 2. At two sites, a long-term comparison of tipping bucket with conventional wedge gauges is used to assess measurement accuracy. Analysis of precipitation results for interannual, seasonal, and diurnal scales for the period 1998–2006 is presented in section 3. Conclusions and suggestions for continuing work (section 4) complete the paper. A study of other aspects of this study region with emphasis on observing the forcing mechanisms of the river breeze and its effect on other aspects of mesoclimate will appear in a companion paper.

1.2. Qualitative Predictions

[11] If river breezes generated by river-land temperature contrasts exert a dominant influence, stations near the river should be in a subsident region and lack an afternoon convective rainfall peak. One expects breezes to be more important during the dry season. The Amazon River is roughly parallel to the predominant easterlies, while the Tapajós is approximately normal. Mean low cloudiness is indeed suppressed over the rivers near the Tapajós-Amazon river confluence (Figure 1b). Naïve expectation is that the Tapajós breeze might produce the largest effect on precipitation since it lies roughly normal to the prevailing easterly winds, promoting enhanced convergence east of the river. This expectation was supported by model studies [Silva Dias et al., 2004; Lu et al., 2005], which correctly simulated enhanced cloudiness on the east bank of the Tapajós River for selected cases.

[12] Large instability lines propagate inland from the coast, arriving at Santarém (about 55° W) predominantly at night [Cohen et al., 1995; Kousky et al., 2005]. The nocturnal arrival time facilitates objective partition of rainfall into events of ‘basin scale' associated with these lines and those of local convective origin, a task previously done subjectively [e.g., Garstang et al., 1994]. One expects rainfall at all stations subject to predominantly squall rainfall should exhibit a nocturnal maximum. This has been taken to be the norm at the longitude of Santarém [e.g., Nechet, 1993; Angelis et al., 2004]. Since near-river stations would lack the afternoon convective component, one might expect a negative precipitation bias overall at stations near the rivers. Topographic effects would be expected to lead to increased rainfall along slopes oriented normal to the predominantly easterly regional flow. No simple qualitative prediction can be made from consideration of the channeling of the airflow over the rivers. Clear evidence of channeling exists for the mouth of the Amazon [e.g., Cohen et al., 2006]. We hypothesize that there is enhanced rainfall in convergent regions where the channel narrows after attaining a long trajectory over a wider water area, such as just to the west of the Amazon-Tapajós confluence.

2. Instrumentation and Observations

2.1. LBA-ECO Network

[13] In 1998, we began deployment of what grew to be a nine-station mesoscale weather station network (Figure 3a and Table 1). This network began with two stations acquired from the U.S. Forest Service. Later, three stations from the Brazilian agricultural research corporation EMBRAPA and four obtained using LBA-ECO resources were added. To promote data uniformity, we upgraded each station to include a GPS receiver (to ensure compatible timing), a barometer, and soil temperature and moisture sensors. Data were recorded and processed in real time using similar Campbell Scientific data loggers running identical programs. Data were retrieved manually at weekly intervals.

Figure 3a.

LBA-ECO rain gauge locations and regional topography. Station 15 is the Santarém airport. Stations 1 and 18 are adjacent in Belterra. Scale shown is terrain altitude above mean sea level (m).

Table 1. Precipitation Stations
StationNameFrequencyLatitudeLongitudeID
1Belterra30 min−2.64631−54.94361c
2Cacoal Grande30 min−2.38944−54.32861e
3Vila Franca30 min−2.34861−55.02889f
4Guarana30 min−2.67694−54.32472g
5Jamaraqua30 min−2.80639−55.03639j
6km11730 min−3.3502−54.924k
7Mojui30 min−2.76667−54.57917m
8Sudam (Curua Una)30 min−2.54417−54.09083s
9km671 hour−2.88528−54.92047km67
10km7730 min−3.01983−54.89425117
11km83 above canopy1 hour−3.0502−54.92800km83a
12km83 below canopy1 hour−3.0502−54.92800km83b
13CPTEC1 hourNANAcp
14Liebmann1 dayNANAlb
15LBAoffice Santarém30 min−2.423279−54.74451lba
16Terra Rica1 day−2.84257−54.97461TR
17Casa da Onça1 day−2.89924−54.95478CO
18Belterra INMET1 day−2.64631−54.94361IN
19Arapari Maicuru1 hour−1.75−54.41732210
20Boca do Inferno1 hour−1.567−54.81732187
21Nova Esperanca1 hour−2.439−54.70532186
22Obidos1 hour−1.91−55.50832185
23Oriximinã1 hour−1.733−55.8532184
24Santarém1 hour−2.408−54.73632207
25Vila Conceicão1 hour−1.759−55.8632179
26Prainha1 day−1.7989−53.453300153000
27Arapari1 day−1.7597−54.369400154000
28Inferno1 day−1.5−54.861900154001
29Alenquer1 day−1.9239−54.727500154003
30Santarem1 day−2.4236−54.692500254000
31Barragem1 day−2.785−54.268900254005
32Curuauna1 day−2.8−54.2700254009
33Curuai1 day−2.265−55.452800255000
34Arua1 day−2.6492−55.712800255001
35São Jose1 day−2.5356−55.359200255002
36Juruti1 day−2.1478−56.079200256001
37Nhamunda1 day−2.1769−56.689400256002
38São Pedro1 day−3.8778−54.28500354000
39Carvalho1 day−3.5833−55.36600355000
40Mutum1 day−3.35−55.116700355001

[14] Additional LBA-ECO tipping bucket data were available for limited periods from the project office in Santarém (station 15, Table 1). Rainfall data was also collected at three flux towers at sites known by their approximate location along the BR-163 highway. These were an agricultural site (km77, station 10 [Sakai et al., 2004]), an old-growth forest site (km67, station 9 [Saleska et al., 2003; Hutyra et al., 2005]), and a forested site where selective logging was done (km83, station 11 [Miller et al., 2004; da Rocha et al., 2004]). The recording rain gauges used were Texas Electronics TE-525 tipping buckets that report 0.1 mm per tip except at the Cacoal Grande, Sudam, and Vila Franca sites where English unit rain gauges reported 0.254 mm tips. The tipping bucket mechanism is identical in all gauges, but the ones denominated in English units have an appropriately larger bucket diameter. Data were sampled at 5-s intervals and accumulated to half-hourly totals. We recorded data at 5-s intervals at the km77 site. The sensors were installed approximately 1 m above the surface, except at the forest sites where gauges were placed on towers just above or near the 45 m average canopy top. Sensors were inspected and cleared of obstructions each time data were retrieved. Windshields were not installed at stations in this network. Hanna [1995] notes that their utility for rain measurement is not certain. Daily totals from wedge rain gauges in forest clearings were available at two sites (Table 1 and Figure 3a), one approximately 500 m (Casa da Onça, CO) and the other 6.4 km (Terra Rica, TR) away from the km67 study site [Nepstad et al., 2004]. We acquired daily total precipitation data from these sites (except for weekends and holidays) from the beginning of 1999 to the present. Limited in situ calibrations were performed at each tipping bucket station annually.

2.2. Operational Network Data

[15] Precipitation data from the additional stations in the operational network (Figure 3b) complement our analysis. Stations of the Brazilian Hidro network [Angelis et al., 2004] (hidroweb.ana.gov.br) are equipped with Vaisala 55 ES 13S tipping buckets that report hourly rainfall totals. A larger number of stations equipped with conventional gauges were read manually for daily rainfall total reports (Table 1). The data period considered in this paper for these and all other stations is shown in Figure 3c.

Figure 3b.

Hidro network rain gauge locations. Also shown are LBA-ECO stations from Figure 3a (stars), the Belterra INMET station (B), and the Taperinha station (T). Stations in red have hourly reports.

Figure 3c.

Data availability during the period 1998–2006 for the stations listed in Table 1.

2.3. Gridded Data Products

[16] Three regional data products were obtained: (1) Rainfall data from several of the operational stations has been gridded to 1° resolution [Liebmann and Allured, 2005]; (2) output from the Brazilian Numerical Weather Prediction Center (CPTEC) 40-km resolution Eta model reanalysis data [Chou et al., 2005]; and (3) the NOAA Climate Prediction Center Morphing (CMORPH) 0.25° resolution data product [Joyce et al., 2004] for the period December 2002 to May 2007. The CMORPH data (Figure 2), used here as a guide to regional mean rainfall gradients, are based on the twice-daily polar orbiting satellite-based passive microwave precipitation estimates, with the time resolution improved by tracking identified cloud elements between microwave estimates following geostationary satellite infrared cloud images. This data is particularly valuable for understanding the mean horizontal precipitation gradients for our region of interest. Recently Tian and Peters-Lidard [2007] argued that the CMORPH precipitation estimates are too high near large water bodies. However, they identified these positive anomalies occurring with light rain rates (<2 mm/d). We do not believe that this potential difficulty affects the conclusions of this study.

3. Analysis

3.1. In Situ Rainfall Measurement Issues

[17] Tipping bucket (TB) rain gauge estimates depend on rain rate, wind speed, and sensor exposure, among other factors. All of the LBA-ECO automatic rain gauges are tipping bucket sensors from the same manufacturer. At Belterra (station 1, Figure 3a), the tipping bucket sensor was approximately 100 m from the long-term climate station operated for many years by the Brazilian Meteorological Service (INMET). We compare daily rainfall totals between the LBA tipping bucket and the Belterra INMET station to facilitate meshing our data with operational station reports in the region.

[18] Sevruk [1996] found that daily total tipping bucket rainfall at one site in Switzerland was on average 14% lower than that measured with a weighing gauge. For rain rates above 2 mm/h, the underestimate varied approximately between 8% and 15% for wind speeds between 1.3 m/s and 5.3 m/s. The Humphrey et al. [1996] dynamic calibration the LBA-ECO rain gauge model yielded a 12% underestimate for rain rates between 0 and 50 mm h−1 that are characteristic of the study region. Rollenbeck et al. [2007] found that their tipping bucket data was 11% lower than that registered with a totaling device over an annual cycle in Ecuador. Even when identical rain gauges are deployed in regions with convective rainfall, differences as large as 9% are found over short distances [Tokay et al., 2003]. Our own laboratory tests of the tipping bucket mechanisms in the LBA-ECO sensors verified the manufacturer's calibration.

[19] Though we conducted occasional checks to verify that the LBA-ECO rain gauges were unobstructed and the mechanisms were working properly, it was not feasible to perform regular high quality calibrations. Comparison between the Belterra tipping bucket and the nearby INMET manually sampled rain gauge links our network values with those in the historical record. Each gauge can be expected to experience very similar wind conditions during rain. Data were screened to include only days with rain. To facilitate comparison with the daily precipitation records from conventional rain gauges recorded manually; daily totals reported in this paper are the original half-hourly data summed over a 24-h period ending at 12 UT (close to sunrise) on the nominal day of record. Data were excluded when the tipping bucket data record was incomplete or when there was no INMET data. Daily (Figure 4a) and monthly comparisons (Figure 4b) between stations confirm that the tipping bucket observations in this Amazon rainfall regime exhibits an approximately 15% underestimate relative to the manual observations for both totaling intervals. Though dynamic corrections [e.g., Calder and Kidd, 1978] may be needed to estimate rain rates from the tipping bucket instrument, undercounting during high rain rates can be ‘made up' because water accumulated in the rain gauge eventually passes through the bucket. Noting this fact, and that the 1-min rain rates recorded at the km77 site did not reach the values for the dynamic correction to be significant, we elected not to include the quadratic dynamic correction on 1-min rain rates done by the U.S. National Weather Service for its ASOS automatic weather stations [Nadalski, 1998]. The half-hourly average wind speed and maximum gust during rainfall events were 2.3 m/s and 7.3 m/s, respectively. The similar behavior between tipping bucket and conventional sensor for daily and monthly totals at Belterra gives us confidence in the field intercomparison. Given uncertainty about the turbulence near the rain gauges, we elected to apply only a fixed 15% correction at all stations in the LBA-ECO network.

Figure 4a.

Comparison of daily averaged LBA-ECO tipping bucket (station 1) and conventional wedge gauge daily totals for the Belterra INMET station (station 18) 1998–2006. The 1:1 line is shown along with the least squares regression line passing through zero (slope equal to 0.87). The circled point was not included in the regression.

Figure 4b.

Comparison of monthly averaged LBA-ECO tipping bucket and conventional wedge gauge monthly totals for the Belterra INMET station 1998–2006. The slope of the least squares fit through the origin is approximately 0.87.

[20] We also examined tipping bucket and nearby manually collected daily total rainfall observations for sites near the km67 tower site (station 9, Table 1). Wedge rain gauge data from the Seca Floresta rainfall exclusion project (TR, station 16; CO, station 17) show much poorer agreement (see monthly comparison, Figure 5) with the tipping bucket rain gauge for either the daily or monthly total comparisons than was seen for the Belterra wedge-TB comparison. The two manual stations showed extraordinary fidelity; the linear least squares fit with zero intercept of TR versus CO monthly totals had slope = 1.01. For monthly totals, the least squares fit through zero (Figure 5) between TR and km67 had slope = 0.83 after the 15% upward adjustment discussed in section 3a. Note that if only months with totals < = 250 mm are included, the slope drops to 0.74. Perhaps because water remains in the manual rain gauges until data are eventually recorded and the gauge emptied, monthly totals show much better agreement than do the daily totals that show a much smaller correlation. No better agreement resulted from excluding days of the week or holiday periods during which no manual readings were made. We have no simple explanation for this discrepancy.

Figure 5.

Comparison of monthly total precipitation (mm) between the conventional wedge gauge at Terra Rica station (station 16, NepstadTR) and. the LBA-ECO tipping bucket at km67 (station 9, km67). Fitted curves include the 1:1 line (heavy solid line), least squares fit through zero (thin solid line, slope 0.83), and the least squares fit to all data < = 250 mm monthly (dashed line, slope 0.74) are shown. The tipping bucket rain gauge estimate was corrected 15% upward before the regressions were done (details in section 3.1).

[21] Spatial rainfall patterns are not affected by this adjustment, since the network rain gauges are identical. Robinson et al. [2004] found that gauges installed at or just above canopy height agree well with those in nearby gauges outside the forest, but they cited one study for which rain totals abruptly drop to 80% of nominal values when the gauge was greater than 4 m above the canopy. We have insufficient detailed information about the sensor type and exposure to understand the large difference between the km67 and Casa da Onça rainfall totals.

3.2. Precipitation Climatology

3.2.1. Historical Records

[22] Differences in annual total precipitation as large as 20% in long-term averages occur (Figure 6) near the confluence. Molion and Dallarosa [1990] (hereinafter MD) argued that difference in annual average rainfall (1978–1988) between the two inland sites (Figure 6, letters c and d) and two just next to the Trombetas River (Figure 6, letters a and b) was the result of rainfall inhibition caused by the river breeze. MD did not report wind data but it seems equally plausible that the 0.32 m annual rainfall difference might have resulted from topographic uplift, or other mesoscale circulation apart from a breeze, hypotheses that warrant subsequent model testing. Very near the rivers' confluence, average precipitation is somewhat larger near to the Amazon channel (Figure 6, letter e, 1914–1981 Taperinha; letter g, 1965–1984 at Santarém airport [Nechet, 1993]) than the 1988–2006 average measured at the Belterra climate station (Figure 6, letter f).

Figure 6.

(left) Regional topography (meters MSL, scale at right) and river courses. Stations cited by Molion and Dallarosa [1990] (a, b, c, d) and by Nechet [1993] (Santarém, g) are shown; Taperinha, e; Belterra INMET station, f; with Manaus (MAO). (right) Historical annual average total rainfall (m).

[23] Convective rainfall can be quite patchy, making individual case study days difficult to analyze. On 21 July 2003, a striking image captured by a Space Shuttle astronaut shows a striking multicellular convective complex (Figure 1d) that produced unevenly distributed rainfall throughout the network (Figure 7).

Figure 7.

(top) Rainfall (mm/half-hour) for 20–21 July 2003 (DOY 202). (bottom) Incident global solar radiation (W/m2) at the same stations.

[24] Infrequent but extremely large rainfall events are observed at many tropical stations. Voss [1922] reported maximum rainfall over 24 h in Pará state Brazil of 178 mm. Liebmann and Allured [2005] were obliged to omit such ‘outliers' in their data, noting that though such events actually occur they are not amenable to the task of making grids meant to be representative of a wide area. Our data series are neither sufficiently long nor continuous to warrant investigating extreme value distributions [e.g., Katz et al., 2002], but some examples illustrate the importance of a few extreme events to monthly and annual totals. The most frequent site of extreme events was the Cacoal Grande station (station 2). In 2005, for example, on days 48, 124, 125, 137, and 138 there were periods during which precipitation exceeded 50 mm in a 30–min period.

[25] The most striking case is for day 137 (17 May 2005) when 402 mm of rain fell between 10:30 and 16:30 UT at the Cacoal Grande station. During four 30-min periods, an identical value 72 mm was registered. Such a total represents 25% of the average annual total. Laboratory tests confirmed that this rate is what this rain gauge reports if it is totally filled with water and allowed to empty at its maximum tipping rate. In the laboratory, we could not duplicate any data logger malfunction that would lead to such observations. Data from other stations as well as the CMORPH data product show that there was indeed a storm on this day, but no other station experienced such an extreme downpour. We believe that this event was real, not the result of sensor malfunction or vandalism. Not only did the instrument continue operating normally after this event but also there were several other high-rainfall intensity events later. However, we have only one such event, and we reluctantly discarded the rainfall total from this event in subsequent analyses. Clearly longer data records are needed to describe extreme rainfall event statistics in this region.

[26] The top three daily precipitation totals in the current record (Figure 8) tend to occur at stations very close to the rivers. The three stations along the Amazon River and rainfall observed at the two Tapajós National Forest flux towers (km67, station 9; km83, station 11) have the highest triplets of extremes in the region. Extreme rainfall at Jamaraqua station (station 5), along the Tapajós riverbank is suppressed. Enhanced rainfall in the National Forest on the plateau just a few kilometers to the east is evidence of a precipitation bias possibly caused by the Tapajós breeze. Reports from the wedge gauges in the National Forest (stations 16 and 17) may have included more than one day in the daily total and were not plotted. They were slightly larger than the results from the tipping buckets for km67 and km83.

Figure 8.

The three highest daily rainfall totals (mm) at selected stations in the area during the observation period 1998–2006.

3.2.2. Annual Mean Rainfall

[27] Complementary methods were used to estimate the spatial pattern of regional rainfall from the new data. First, annual mean rainfall totals were calculated for the stations for which daily total data were available. Since records are not continuous (Figure 3c), monthly averages of daily total were calculated for each station and then summed to get a representative annual average. The second approach was to calculate monthly mean precipitation from the CMORPH 0.25° data for the period December 2002 to May 2007.

[28] Interannual differences in total rainfall can be greater than 380 mm (Figure 9). Annual total rainfall at upland stations more than 10 km south of the Amazon channel is 200–400 mm below that recorded at the three stations very near the river, verifying the gradient previously identified in the long term records. The 1° precipitation grid point that includes Santarém from the Liebmann and Allured [2005] (hereinafter LR) data is 2108 mm, a value that more closely resembles precipitation at the riverbank stations than that seen inland. (Note that the LBA-ECO stations are not included in the LR synthesis.) Rainfall suppression during the 1998 El Niño event is apparent in Figure 9, but there is little evidence of the 2005 Amazon drought, whose influence was mostly felt further west in the interior Amazon Basin.

Figure 9.

Monthly total rainfall (mm) for (top) upland stations and (bottom) stations near the Amazon River. Heavy solid line is the Liebmann and Allured [2005] (LR) gridded data that includes the city of Santarém. Annual total precipitation (mm) (top) from the Belterra station and (bottom) for the LR data are given across the top of each panel.

3.2.3. Spatial Variability of Total Precipitation

[29] A precipitation map (Figure 10) was made by finding the average daily rainfall during each month and then estimating monthly and annual sums. Rainfall totals are highest to the north and west of the Tapajós-Amazon confluence. East of the Tapajós River and south of the Amazon (the “LBA-ECO sector”) total rainfall diminishes to the south, a pattern common to the individual station totals as well as in the CMORPH composite, the opposite of the usual expectation of the influence of a river breeze. The ‘tongue' of higher precipitation along the Amazon River west of the confluence identified by the CMORPH precipitation contours is consistent with the high values of precipitation seen at Vila Franca (station 3, Figure 3a), approximately 33% larger than the rainfall expected in the LBA-ECO sector. Precipitation totals are lower than the CMORPH estimates at sites north of the Amazon River (e.g., Arapari, station 19) except at the station Boca do Inferno (station 20), which is at somewhat higher elevation than are surrounding stations. These differences may reflect topographic enhancement in the first case and the possibility of ‘rain shadows' along the Maicuru tributary to the north in the latter. Average rainfall at Jamaraqua (station 5, at the Tapajós riverbank) is 0.3 m less than in the Tapajós National Forest sites (km67, station 9; km83, station 11). Results from the two Seca Floresta wedge gauges (Terra Rica, station 16; Casa da Onça, station 17) are slightly higher, as discussed earlier, but still indicate a modest rainfall enhancement in the National Forest, possibly a consequence of the breeze-induced subsidence suppressing rainfall near the river.

Figure 10.

(left) Annual total precipitation (m) based on the 0.25° resolution CMORPH averages for the period 2003–2006. (middle) Annual totals based all direct observations for the periods described in Figure 3c. (right) LBA-ECO area shown on an expanded scale. Additional stations Terra Rica (station 16) and Casa da Onça (station 17) and the Belterra INMET station (station 18) are shown in parentheses.

[30] The distinctive rainfall spatial maximum at the western end of the river confluence (e.g., station Vila Franca, station 3, Figure 10) reflects large wet season precipitation relative to that at other stations (Figure 11b). Bourgoin et al. [2007] observed similar large annual rainfall totals for the Obidos (station 22) region and confirm that there were strong north-south precipitation gradients.

[31] Since data were not available continuously at all sites for the complete data period, we compare precipitation ratios for hours during which groups of stations were all functioning. At the longest running stations for 2178 days of observation, the ratio of precipitation between the station furthest south of Santarém, km117 (station 6) to that at Belterra (station 1) was 0.83 in the annual average but 1.35 taking only the dry season (days 196–348). It is plausible that the diminished dry season rainfall is of local convective origin, and rainfall suppression near the Amazon River owing to river breeze subsidence there plays a stronger role during that season. We found that rainfall totals for 2000–2004 in the CPTEC reanalysis data [Chou et al., 2005] follow the observed seasonal patterns well but are approximately 40% lower than LBA-ECO measurements over the entire study area.

3.2.4. Diurnal Rainfall Characteristics

[32] At most sites in the LBA-ECO network the average diurnal precipitation pattern shows one peak at about 06 UT (03 LT) and another at 18 UT (15 LT) in both the dry (Figure 11a) and wet seasons (Figure 11b). The patterns for stations south from the Amazon River have remarkably similar shapes. The same pattern of twin diurnal peaks occurs for the Hidro stations both to the west and to the north of the confluence (Figures 11a and 11b). The early morning precipitation peak in hours is consistent with previous studies that propagating squall lines arrive from the coast at about that time discussed earlier [Garstang et al., 1994; Ferreira et al., 2003; Cohen, 1989].

Figure 11a.

Rain dials for LBA rain gauges during the dry season (days 196–348). A steady rain amounting to a daily rate of 5 mm/d is shown in the box at lower left. Numbers are the hour (UT).

Figure 11b.

Rain dials for LBA rain gauges during the wet season (days complementary to the dry season). A steady rain amounting to a daily rate of 10 mm/d is shown in the box at lower left. Numbers are the hour (UT).

[33] We had some difficulty in analyzing the Hidro hourly data. Data were available as the accumulated rainfall from the gauge register. At certain stations, this register was reset quite frequently. We chose a discontinuity threshold of 50 mm/hour to distinguish real rain events from spurious ones perhaps caused by a jumping register. Consistent regional gradients were then obtained for daily totals, which we obtained as a separate data set. However, when we used the hourly data for Santarém airport, the rain dial did not show the single nocturnal peak that the three other nearby riverbank stations exhibited. The daily totals made using daily and hourly data sets did not match for Santarém airport, and we excluded this rain dial.

[34] The observed inland afternoon peak related convective rainfall provoked by surface heat fluxes is similar to that found elsewhere in interior continental situations, but it contrasts with the common understanding that the Santarém region has its primary rainfall at night [Nechet, 1993; Angelis et al., 2004; Kousky et al., 2005]. Three stations along the main channel of the Amazon, Vila Franca (station 3), LBA-Office (station 15) and Cacoal Grande (station 2) show only the nocturnal peak. The large total value for the LBA Office is a consequence of the importance of a small number of extreme events that occurred during the relatively short data record. The diurnal rainfall pattern at Jamaraqua (station 5), our only station at the Tapajós River bank well south of the confluence, is similar to the pattern at the Tapajós National Forest sites (stations 9, 10, and 11). However, there is less total precipitation at Jamaraqua, due to a somewhat smaller afternoon peak during the wet season.

[35] Regional rainfall data splits naturally into 0–12 UT and 12–24 UT categories. At the three Amazon River stations near Santarém, rainfall is predominantly nocturnal (Figure 12). The diurnal frequency pattern changes to the north and west of the confluence in the CMORPH averages (Figure 13), but the 0.25° data set may lack sufficient resolution. In ongoing work we will assess this using the 8-km CMORPH data. There appears to be a narrow band of predominantly nocturnal precipitation very near the Amazon River, and this leads to the annual total rainfall to be higher there than inland. This occurs despite the fact that the Amazon and Tapajós river breezes effectively suppress afternoon convection near the riverbanks in both the dry and wet seasons.

Figure 12.

(left) Annual total rainfall (m) for stations that report hourly. (right) Nocturnal rainfall fraction from stations (%). Hidro station values are given in red.

Figure 13.

Percentage difference between daytime rainfall (12 UT–24 UT) and nocturnal annual total rainfall (0 UT–12 UT) for the 0.25°CMORPH data in the study area.

3.2.5. Rainfall Intervals

[36] Understanding the interval between rainfall events has important implications for soil moisture state and evapotranspiration rates, especially as the rainfall total is often used to describe the precipitation climate. Shallow-rooted crops, for example, suffer water stress before forests as soil moisture drops during prolonged dry spells. Dry season rainfall is so infrequent that residents of Belterra area supplement their rain barrels with municipal water distributed by truck. In nearby forests, Parrota et al. [1995] and Nepstad et al. [2004] found that trees developed deep root systems to help survive the drought months.

[37] We take a rain event to be the number of continuous hours during which at least a single tip was recorded. The rain recurrence interval (tr) was taken as the time between the start of consecutive events. Details of the intervals between rainfall events and their frequencies (Tables 2a2e) indicate that precipitation is most strongly affected by subsidence associated with the river breezes during the dry season. The dry season intensity, identified by the mean duration between rains and the number of stretches of dry days, decreases as one goes inland from the rivers. In a companion study, we will focus how soil moisture in the cleared region drops off when subject to continuing evapotranspiration during the interval between rains. Duration of rain-free periods during the dry season (days 196–348) is also important to determine the fire susceptibility of the rain forest [Cochrane and Schultz, 1999; Ray et al., 2005].

Table 2a. Rainfall Recurrence Interval and Frequency by Location, Year, and Season: Amazon River Bank North Sidea
Cacoal Grande (Station 2)200320042005
Wet
   tr1.020.850.87
   (trtr > 1 day)3.032.352.31
   trtr > 1 day241546
   trtr > 5 days302
   trtr > 10 days100
Dry
   tr3.285.922.91
   (trtr > 1 day)6.788.5510.87
   trtr > 1 day162015
   trtr > 5 days7107
   trtr > 10 days274
Table 2b. Rainfall Recurrence Interval and Frequency by Location, Year, and Season: River Bank West of the Confluence of the Tapajós and Amazon Rivers
Vila Franca (Station 3)200320042005
Wet
   tr0.870.790.51
   (trtr > 1 day)2.572.452.2
   trtr > 1 day453038
   trtr > 5 days401
   trtr > 10 days100
Dry
   tr4.63.883.24
   (trtr > 1 day)6.598.816.85
   trtr > 1 day251923
   trtr > 5 days1379
   trtr > 10 days745
Table 2c. Rainfall Recurrence Interval and Frequency by Location, Year, and Season: 4.5 km East of the Tapajós River
Belterra (Station 1)1999200020012002200320042005
Wet
   tr0.53750.410.460.480.620.470.58
   (trtr > 1 day)1.81.51.51.92.71.92.1
   trtr > 1 day40353939163834
   trtr > 5 days0001201
   trtr > 10 days0000100
Dry
   tr1.541.463.092.882.102.492.46
   (trtr > 1 day)3.493.355.825.424.846.225.21
   trtr > 1 day26362630282819
   trtr > 5 days561010785
   trtr > 10 days0154344
Table 2d. Rainfall Recurrence Interval and Frequency by Location, Year, and Season: Inland South of the Amazon, East of the Tapajós
km77 (Station 10)20012002200320042005
Wet
   tr0.390.440.500.490.51
   (trtr > 1 day)1.451.941.642.032.05
   tr > 1 day4840504339
   tr > 5 day01032
   tr > 10 day00000
Dry
   tr2.191.881.230.841.86
   (tr > 1 day) (day)4.265.023.194.393.91
   tr > 1 day2927413027
   tr > 5 day107488
   tr > 10 day65221
Table 2e. Rainfall Recurrence Interval and Frequency by Location, Year, and Season: Further Inland From Table 2d
km117 (Station 6)199920002001200220032004a2005
  • a

    Partial year data record.

Wet
tr (days)0.410.350.380.500.531.120.52
   (trtr > 1 day)1.621.391.542.202.094.112.22
   trtr > 1 day3835232836439
   trtr > 5 days0002213
   trtr > 10 days0000011
Dry
   tr (days)1.221.191.181.961.251.861.26
   (trtr > 1 day)3.273.654.745.163.464.124.58
   trtr > 1 day28342330342731
   trtr > 5 days4768569
   trtr > 10 days1434323

[38] Over the whole area, the mean dry season rain recurrence interval for most stations is about double that of the wet season. Wet season rainfall intervals cluster about a half-day. These averages complement the earlier rain dial interpretation, that there is a combination of afternoon convection and nocturnal squall rainfall (see Figure 11). At stations near the large rivers (Tables 2a2c) the mean interval between rainfall events, about one day in the wet season grows to nearly three days in the dry season. The frequency of dry sequences of days is exaggerated for stations along or near the riverbanks. In the dry season, there were 16–20 cases of 1- day dry periods. In 2004, the Cacoal Grande site experienced seven 10-day droughts. At more inland stations the presence of more convective precipitation means that such long stretches of dry days do not occur. At Belterra, for example, the dry season is more intense, there is a longer interval between rains, than that experienced further inland at km117, especially during the 2001–2004 dry seasons. At inland stations km77 (station 10) and km117 (station 6), there were between 35 and 40 cases of one-day rain-free periods in the wet season. The mean rainfall interval is 12 h, again reflecting the presence of the bimodal diurnal pattern of rainfall.

4. Conclusions

[39] We found that the near-river stations do indeed miss the afternoon convective rain as would be expected if the river breeze influence dominates, but paradoxically this deficiency is more than compensated by additional nocturnal rainfall. This effect is local; for areas only a few kilometers inland from the rivers, nocturnal squall lines contribute less than half of total precipitation.

[40] Describing the proper mixture of types of precipitation should be a concern for those assessing model sensitivity, especially since the reanalysis rainfall data are currently flawed [Marengo, 2006], and for those applying isotope studies to infer moisture recycling in this region [Henderson-Sellers et al., 2002]. It is clear that models and remote sensing products should be evaluated not only comparing rainfall daily totals, but also by the ability to reproduce the diurnal precipitation pattern. At a larger scale, the rainfall totals increase just to the west of the Tapajós-Amazon confluence.

[41] In general, the breeze circulations associated with the Amazon River (with a wind component approximately normal to the mean flow) affect rainfall more than does the Tapajós breeze (which approximately opposes the prevailing wind). Evidence for the importance of the Tapajós breeze on precipitation is sketchy, but it appears that the breeze influence extends only a few kilometers inland. There is a distinct increase in the top three extreme rainfall events comparing Jamaraqua, (station 5 in Figure 3a; 96, 80, 76 mm, Figure 8) as compared to that measured at the km67 site in the nearby forest on the plateau just a few km inland (station 16; 150, 150, 150 mm). Increased rainfall north of the Amazon is possibly the result of orographic uplift, but further studies are needed to confirm this.

[42] We found that tipping bucket rain gauges work adequately in the Amazon rainfall regime, but careful field calibration and comparison with collocated conventional rain gauges is essential to incorporate daily totals from operational array in area-wide maps. The 0.25°CMORPH data product does an adequate job of describing medium scale variability in this region, but very localized breeze effects are not resolved.

[43] Nechet [1993] first hypothesized that local mesoscale circulations, perhaps related to the large lake-like expanse of water at the confluence, are responsible for the nocturnal precipitation preference. As squall lines approach this region, enhanced moist inflow is possible, perhaps augmented by southerly channeling up the Tapajós channel as the storm approaches, a feature that has been observed elsewhere (e.g., LaPenta et al. [2005] in the Hudson Valley of New York). No simple isolated breeze or topographic enhancement hypothesis is adequate to explain these data. The relative importance of the nocturnal squall lines to overall precipitation is exaggerated if one relies solely on data from stations along the Amazon River channel near Santarém. Larger rainfall totals further west could plausibly result from convergence promoted by air following the narrowing channel. We plan to address this issue using mesoscale modeling case studies of well-documented individual squall line passages. Complementary case studies of fair weather days will be conducted to understand the important of orographic effects on the apparently enhanced rainfall just to the north of the Amazon River in this region.

[44] To improve our analysis, we plan to analyze higher resolution 8-km CMORPH rainfall data. We will acquire longer data sets and investigate the statistical distribution of extreme rainfall events. With colleagues, we will conduct mesoscale model studies to estimate bias at other points in the Amazon Basin, with the express aim to understand how the river orientation relative to prevailing easterly winds affects the nature of the rain biases. River proximity bias in other climate variables will be discussed in a companion paper that emphasizes extracting wind, radiative flux, and diurnal pressure gradients and exploring how these vary seasonally.

Acknowledgments

[45] This work is part of the LBA-ECO project, supported by the NASA Terrestrial Ecology Branch under grants NCC5–283 and NNG-06GE09A to the University at Albany. The UFSM authors also acknowledge support by CNPq, the Brazilian Science Agency. Michael Keller got the network started by facilitating the transfer of the original two weather stations to the region, whose installation at Belterra and km117 was guided by expert technician Jorge de Melo of CPTEC before the rest of the LBA-ECO project began in earnest. Teams from Harvard University (S. C. Wofsy and L. Hutyra) and University of California, Irvine (M. Goulden and S. Miller), generously allowed us access to data from the Tapajós National Forest flux towers. Daniel Nepstad and his team also kindly made available their wedge gauge data from the Tapajós National Forest. Carlos F. Angelis of CPTEC provided data from the Hidro network. We acknowledge help in the field by Kathleen Moore, Dwayne Spiess, Ralf Staebler, Alex Tsoyref, Eleazar Brait, and Valdelírio Miranda. Staff at the LBA-ECO office in Santarém for logistics, especially Bethany Reed, provided additional support. Rainfall data from the LBA office, part of research efforts led by Paulo Artaxo, were provided by Rodrigo da Silva (UFPa, Santarém). Crucial daily total rainfall data for the Belterra INMET station were provided through the kind offices of Alaor Moacyr Dall'Antonia Junior and Divino Moura of INMET. We acknowledge the work of the CPC CMORPH team and especially R. Joyce at NOAA who have made those data available to the community. The manuscript was greatly improved by suggestions from J. W. Snow, MIT Lincoln Laboratories, the Editor, and three anonymous reviewers.

Ancillary

Advertisement