SEARCH

SEARCH BY CITATION

Keywords:

  • Amazon;
  • TRMM;
  • rainfall;
  • large water bodies

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[1] Tropical Rainfall Measurement Mission (TRMM) data show lower rainfall over large water bodies in the Brazilian Amazon. Mean annual rainfall (P), number of wet days (rainfall > 2 mm) (W) and annual rainfall accumulated over 3-hour time intervals (P3hr) were computed from TRMM 3B42 data for 1998–2009. Reduced rainfall was marked over the Rio Solimões/Amazon, along most Amazon tributaries and over the Balbina reservoir. In a smaller test area, a heuristic argument showed that P and W were reduced by 5% and 6.5% respectively. Allowing for TRMM 3B42 spatial resolution, the reduction may be locally greater. Analyses of diurnal rainfall patterns showed that rainfall is lowest over large rivers during the afternoon, when most rainfall is convective, but at night and early morning the opposite occurs, with increased rainfall over rivers, although this pattern is less marked. Rainfall patterns reported from studies of smaller Amazonian regions therefore exist more widely.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[2] In the Amazon basin, uncertainty in rainfall causes difficulties in the study of hydrological processes, hydroclimatic variability [e.g., Espinoza et al., 2009], biogeochemical analysis and hydrological modeling [e.g., Coe et al., 2008; Collischonn et al., 2008], whilst the value of ground-level estimates of rainfall is limited by low raingauge density in a region where convective rainfall is spatially highly variable. Estimation of rainfall characteristics by remote sensing, using satellite-derived data from TRMM [Huffman et al., 2007] and CMORPH [Joyce et al., 2004] is an attractive alternative, giving better spatial cover of rainfall fields. Pereira Filho et al. [2010] analyzed hourly rainfall for a four-year record from CMORPH and concluded that convection over the Amazon region is often more organized than had previously been thought, and that satellite-derived rainfall estimates can be an important source of rainfall information where in situ observations are lacking. However, the use of remote-sensed rainfall estimates is not without problems: Tian and Peters-Lidard [2007], for example, reported systematic positive errors in TRMM 3B42 rainfall estimates for pixels associated within land water-bodies, and speculate that this inconsistency results from deficiencies in the TRMM assumptions about water-surface emissivity.

[3] In the Amazon, raingauges are preferentially sited along rivers where most settlements lie [Oliveira and Fitzjarrald, 1993; de Gonçalves et al., 2006; Fitzjarrald et al., 2008]. However, meso-scale circulations close to large rivers such as river breezes may also affect rainfall distribution locally [Silva Dias et al., 2004; Fitzjarrald et al., 2008] and lead to reduced rainfall [Garstang and Fitzjarrald, 1999].

[4] River breezes result from differences in sensible and latent heat fluxes over land and water, enhancing cloudiness over land during daytime, whilst skies over water remain clear; at nighttime, the opposite occurs. Garstang and Fitzjarrald [1999] stated that away from large Amazonian rivers, convergence zones lead to enhanced rainfall over forest and diminished rainfall near rivers, and that daytime enhancement/diminution has a greater net effect on rainfall near rivers than the reverse night-time situations.

[5] River breezes has been observed and modeled over limited areas by several authors [Ribeiro and Adis, 1984; Oliveira and Fitzjarrald, 1993, 1994; Silva Dias et al., 2004; Fitzjarrald et al., 2008]. Reduced annual rainfall in riverine areas near Manaus in central Amazonia has also been reported [Ribeiro and Adis, 1984; Garstang and Fitzjarrald, 1999; Cutrim et al., 2000]. Fitzjarrald et al. [2008], analyzing data from raingauges near the Amazon-Tapajos confluence, found that stations near large rivers missed the afternoon convective rain, but that this deficit was more than compensated by additional nocturnal rainfall. Silva Dias et al. [2004] used high-resolution numerical simulation to explore the atmospheric circulation near the Amazon-Tapajos confluence, concluding that, since most long-term climate stations lie near to Amazonian rivers, measurements taken very close to them must be interpreted with care.

[6] Since raingauges are often sited near large rivers, it is possible that raingauge-derived estimates of Amazon rainfall may be biased. Such a bias would have important implications for hydrological modeling, water resource management, and the calibration and validation of remote-sensed rainfall estimates using raingauge data [e.g., Oliveira and Fitzjarrald, 1993; de Gonçalves et al., 2006; Hughes, 2006]. Using TRMM 3B42 records, this paper therefore explores rainfall spatial variability in the Brazilian Amazon and the evidence for lower rainfall near its large rivers. The paper presents results from TRMM 3B42 data analyses, and does not seek to elucidate physical mechanisms.

2. Data and Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[7] The area studied is that part of the Amazon River basin lying within Brazilian territory, shown in Figure 1. TRMM rainfall were provided by algorithm 3B42 [Huffman et al., 2007], with spatial resolution of 0.25° × 0.25° and temporal resolution of three hours, for the 12-year period 1998-2009. Data were selected within a grid defined by latitudes +6° to −17° and longitudes −74° to −47°.

image

Figure 1. Brazilian Amazon river basin (white), river network and test area (rectangle in central Amazon).

Download figure to PowerPoint

[8] This paper reports results of analyses of three variables. These are: 1) mean annual rainfall, P; 2) mean annual number of wet days, W, defined here as days with rainfall greater than a threshold of 2 mm (as given by Buarque et al. [2010]); and 3) mean annual rainfall accumulated in each 3-hour time interval of the TRMM 3B42 temporal resolution, P3hr. Analysis consisted of two steps. In the first, annual means of P and W were calculated for 1998–2009. In the second, the diurnal variation of rainfall was evaluated using P3hr. For both steps, rainfall reductions near large water bodies were tested for statistical significance. A proper significance test (in which a hypothesis is proposed, and additional data are collected to test it) was not possible given the context, but the following argument gives a reasonable substitute, despite being partly subjective and not fully rigorous. A test area in the central Amazon was selected which extended from longitude 68°W to 56°W and from the equator to 5°S (see Figure 1). Away from its river system, spatial distribution of rainfall within this area is relatively homogeneous and free from the large regional differences seen in Figure 2; any remaining spatial trend in P, W and P3hr was removed by linear regression X* = a1λ + a2ϕ + a3, where λ and ϕ are the latitude and longitude of points within the test area. The variable used in the test procedure was the regression residual X′ = XX*, where X is P, W or P3hr at a TRMM 3B42 grid-point. In what follows, all primed variables denote residuals derived from such regressions. The TRMM 3B42 grid-points were divided into two groups (“Water” and “No Water”) according to whether they were near to major rivers or other water bodies, using a 100 m resolution map of central Amazon wetlands [Hess et al., 2003]. The fraction of wetland within each TRMM 3B42 grid-square was computed: grid-squares with more than 20% wetland were classified as “Water”, the remainder as “No Water”. Grid-squares were identified where P, W or P3hr was significantly smaller than the “No Water” grid points, by means of a t-test [e.g., Wilks, 2006], where one mean was the grid-square value, and the other the mean of the “No Water” grid-squares. A second t-test for difference between the “Water” and “No Water” sample means was also performed [Wilks, 2006]. All tests used a 5% significance level. Finally, since the values obtained from neighboring TRMM 3B42 grid-points will be spatially correlated, an effective sample size was therefore computed for the means entering each t-test, by adapting, to a space of two dimensions, an expression for the equivalent number of independent observations in a serially-correlated time series [e.g., Cressie, 1993]. This expression is N= N(1−ρ)/(1+ρ) where N is the original sample size and ρ is the lag-one serial correlation; in the present context, ρ was averaged over the two grid directions.

image

Figure 2. (a) Mean annual rainfall and (b) mean number of wet days from TRMM 3B42 data (1998–2009).

Download figure to PowerPoint

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[9] Figure 2a, showing P for 1998–2009, gives a general picture of rainfall spatial variability shown by the TRMM 3B42 data-set. Some well-known regional characteristics are apparent, such as the higher rainfall in the Rio Negro headwaters and near the Amazon's mouth; and the E–W and N–S gradients in P. Figure 2a shows an apparent reduction near large rivers, an effect which stands out clearly along the Rios Solimões/Amazon, especially near the confluences with the Negro and Tapajós, and along most Amazon tributaries: Madeira, Purus, Juruá and Japurá, but particularly along the Tapajós and Negro. This reduction is consistent with the conclusions of other studies over more limited regions [e.g., Ribeiro and Adis, 1984; Garstang and Fitzjarrald, 1999; Cutrim et al., 2000], although it does not agree with the findings of Fitzjarrald et al. [2008], perhaps due to the scale of TRMM 3B42 data. The site of the Balbina reservoir (2,400 km2) stands out in Figure 2a, with rainfall lower than in surrounding areas. A reduction in W along larger rivers is also evident (Figure 2b) and more marked than in Figure 2a, especially downstream of the confluence of the Rios Negro and Solimões, over the Balbina reservoir, and near the Amazon- Tapajós confluence.

[10] Figure 3a shows the test area with the largest rivers and the Balbina reservoir, together with the TRMM 3B42 grid points classified as “Water” in black. Based on the values in Table 1, the relative difference between P and W from areas within the influence of large water bodies (“Water”), and from areas without (“No Water”), are −5% ((−82.5 − 41.2)/2486) and −6.5% ((−7.3 − 3.7)/169.8), respectively. Trend removal was necessary for computing these differences because “Water” pixels are not uniformly distributed (see Figure 3a), being more concentrated to the west of the test area, and because there is also a strong E–W gradient in P (Figure 2). The pooled lag-one spatial correlation for both variables was ρ = 0.84. In a first statistical test, it was found that P and W were significantly smaller at “Water” than at “No Water” grid-points (p < 0.05), confirming the lower rainfall near large water bodies. In the second test, Figure 3b shows TRMM 3B42 grid-points in the test area where P is lower than the mean for “No Water” grid-points. These are concentrated along the Solimões/Amazon and Negro rivers, but reductions also appear near the Balbina reservoir and at a grid-point on the Tapajós. Similar results were also found for W, given in Figure 3c, but with more grid-points with smaller values concentrated along Rio Solimões/Amazon. A sensitivity analysis of the threshold value for grid-squares to be classified as “Water” confirmed that results from the statistical tests and the consequent conclusions were unchanged when the threshold ranged from 5% to 50%.

image

Figure 3. (a) TRMM 3B42 grid-squares constituting the “Water” group; (b) grid-squares with lower P than grid-squares in the “No Water” group, (p < 0.05); and (c) grid-squares with smaller W then for the “No Water” group (p < 0.05).

Download figure to PowerPoint

Table 1. Sample Size, Effective Sample Size, and Mean and Standard Deviation of Residuals After Trend Removal in P and Wa
RegionNN′P [mm]W [days]
MeanSDMeanSD
  • a

    N, number of TRMM 3B42 grid-points; N′, effective sample size; SD, standard deviation. Values from grid-squares denoted by “Water”, “No Water”.

  • b

    Statistics computed using original P and W values.

Water32028−82.5147.2−7.310.9
No Water6405741.2129.13.77.7
All grid pointsb960842486211.7169.818.4

[11] Figures 4a and 4b show mean annual rainfall occurring during afternoon-night (A-N) period (15 to 06 UTC, or approximately 11 to 02 h in local time) and during night-morning (N-M) period (06 to 15 UTC or approximately 02 to 11 h in local time). The lower rainfall near large rivers is very marked in the A-N period and is even more pronounced than in Figure 2a, mainly over the Rio Amazon. During the N-M period the opposite is found. A large increase in P is observed along the Rio Amazon between its confluences with Rios Madeira and Tapajós, with a less pronounced increase in P along the Rios Negro and Solimões and Balbina reservoir. Figure 4c shows the diurnal variation of averaged P3hr in the “Water” (P3hr,W) and “No Water” (P3hr,N) grid-points of the test area. Detailed information concerning diurnal variation of precipitation is presented in Table 2.

image

Figure 4. P occurring between (a) 06 and 15 UTC; and (b) between 15 and 06 UTC. (c) Diurnal variation of P3hr averaged over grid-points denoted by “Water” (line with stars) and by “No Water” (line with circles) and its relative difference (bold line).

Download figure to PowerPoint

Table 2. Summary of Diurnal Variation in Rainfalla
Time Interval (UTC)Total P3hr (%)P3hrpbRel. Dif.
MeanSD
WaterNo WaterWaterNo Water
  • a

    Fraction of P (1998–2009); mean and standard deviation (SD) of P3hr from grid-points denoted by “Water” and “No Water”; probability (p) of observed difference between P3hr,N and P3hr,W, on the null hypothesis of no difference in rainfall between “Water” and “No Water”; and relative difference between P3hr,W and P3hr,N.

  • b

    Significant values are marked in bold.

00–0310−18.99.550.171.10.15−11%
03–0610−9.84.964.184.40.35−6%
06–09116.0−3.057.164.30.643%
09–121013.9−6.944.649.40.899%
12–15918.7−9.444.245.20.9413%
15–1817−7.63.883.394.20.32−3%
18–2120−53.526.899.2103.50.01−16%
21–0013−30.715.454.163.10.01−14%

[12] The reduction of P3hr,W and the increase of P′3hr,N occurs during afternoon and early evening, with a relative peak difference of −16% between 18 and 21 UTC. In fact, this effect occurs in all 3-hour intervals between 15-06 UTC, which are the wettest, accounting for 70% of total annual rainfall. The increase of P3hr,W, and the decrease of P3hr,N, are observed during the night and morning.

[13] The relative differences between P3hr,W and P3hr,N are larger for rainfall occurring in most 3-hour intervals than those found for P′. However, these differences are only statistically significant for the 18-21 UTC and the 21-00 UTC time intervals, where the lower rainfall near rivers is observed. The increased P near large rivers during the N-M period was not statistically significant although evident in Figure 4b.

[14] The overall reduction of rainfall over water bodies reported here runs counter to the TRMM 3B42 systematically-positive errors over inland water bodies reported by Tian and Peters-Lidard [2007] in southeastern USA. However, the lower rainfall over large rivers is significant during afternoon, when most rainfall is convective, in agreement with Yang and Smith [2008]. The observed inversion and timing of the rainfall pattern over large rivers is supported by other observations [Cutrim et al., 2000; Fitzjarrald et al., 2008; Negri et al., 2000] and by the descriptions of river breeze effects on precipitation [Garstang and Fitzjarrald, 1999].

[15] Although quantitative estimates of reduction in rainfall over large Amazonian water bodies have been given here, they cannot be regarded as fully definitive measures, since (among other limitations) the test area was selected subjectively. Furthermore, the spatial resolution of TRMM 3B42 (∼25 km) may be larger than the width of most Amazon rivers: when TRMM 3B42 mean annual rainfall is computed at a grid-point, the corresponding grid-square may include areas remote from the water body itself, so that the reduction in rainfall close to large water bodies may well be locally higher than reported in this study.

4. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[16] The reportedly-lower rainfall over large water bodies in the Brazilian Amazon was explored using TRMM 3B42 rainfall data. Three descriptors of rainfall (mean annual total, P; mean number of days with more than 2 mm of rain, W; and mean annual rainfall accumulated in 3-hour intervals, P3hr) were analyzed. It is concluded that:

[17] 1. Visually, the TRMM 3B42 data show a very clear reduction in P and W near large water bodies. This effect is particularly marked along the course of the Solimões/Amazon rivers, along the major tributaries (particularly along the Rios Tapajós and Negro), and near the extensive Balbina reservoir;

[18] 2. This visual evidence was complemented by a quantitative analysis within a selected test area, using an analytical procedure similar to a statistical significance test. This showed that both P and W were lower near large water bodies, by 5% and 6.5% respectively;

[19] 3. The reduction in rainfall near large rivers is greatest during the afternoon (15 to 06 UTC), when most rainfall is convective. An opposite pattern occurring during night-morning (06- 15 UTC) is clearly discernible, although not statistically significant using the approximate test-procedure used.

[20] 4. The observed reduction of precipitation over large water bodies and its diurnal variation is not consistent with errors over inland water bodies in TRMM 3B42 data reported by Tian and Peters-Lidard [2007] but is in accordance with other observational studies and description of river breeze effects on precipitation.

[21] Factors that include the uncertainty in TRMM 3B42 rainfall estimates and the scale of the TRMM 3B42 grid suggest that the quantitative estimates of rainfall reduction close to large rivers reported here should not be regarded as definitive; and reductions in P, W and P3hr may well be locally greater than those given.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

[22] The authors are grateful for support from the Brazilian agencies FINEP and ANA; the TRMM data supplied by NASA and associated agencies; and for constructive comments of David Fitzjarrald, Jhan Carlo Espinoza Villar, and anonymous reviewers.

[23] Paolo D'Odorico thanks two anonymous reviewers.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Conclusion
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
grl27553-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
grl27553-sup-0002-t02.txtplain text document1KTab-delimited Table 2.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.