Introduction

Estimating groundwater recharge is a big challenge that still cannot be solved straightforwardly using any ground or satellite measurements (Healy 2010). Because recharge has a complex interaction with other water-budget components (Dripps et al. 2006), several methods are suggested for its estimation. In general, the methods used to estimate recharge differ from each other by the source of data input (surface water, unsaturated and saturated zone), the governing hypothesis, and the range of spatial and temporal applicability (Scanlon et al. 2002).

Use of multiple methods (three at least) is recommended to reduce uncertainty about recharge estimates (Delin et al. 2007; Misstear et al. 2009). However, in most developing countries, hydrological ground measurements (“truth”) data are scarce (Swenson and Wahr 2009), and rarely more than one method has been used to estimate recharge. In this context, remote sensing (RS) arises as a potential water-resource management tool to provide information on water-budget components (Oliveira et al. 2014; Armanios and Fisher 2014).

To date, there is no direct method to estimate recharge using RS data, which provides spatial, and temporal spectral data (Jackson 2002). Several studies have integrated satellite products with ground-based measurements to estimate recharge as the residual component of the water-budget equation (Brunner et al. 2007; Szilagyi et al. 2011; Khalaf and Donoghue 2012; Műnch et al. 2013).

Despite the intrinsic spatial resolution limitation of current RS products, it is desirable to use satellite data to assess hydrological temporal patterns and the validity of assumptions, and to compare some of them to other results using ground-measured data (Szilagyi et al. 2011; Wang et al. 2014). Without the comparison with ground-based measurements, RS estimates may provide unrealistic water-flux rates, even when they contain information on spatial patterns and relative spatial distributions (Szilagyi et al. 2011). However, hydrological monitoring networks may decrease in many developing countries and the improvement on validated satellite products becomes essential for water management (Anderson et al. 2012).

Although the Guarani Aquifer System (GAS) in an important transboundary groundwater reservoir, it is still poorly understood mainly because ground data are scarce or nonexistent. The GAS has an area of ~1.2 million km2 (Araújo et al. 1999) and is shared by Brazil (71 %), Argentina (19 %), Paraguay (6 %) and Uruguay (4 %). This aquifer is promising for economic growth because of its water volume of about 25,000–37,000 km3 and good water quality (OAS/GEF 2009).

More than 100 Brazilian cities in São Paulo State use water from the GAS, mainly for human supply (4,030 m3 h−1) and irrigation (4,574 m3 h−1) of perennial and semi-perennial plantations (IPT 2011). There is a potential water conflict in the Concordia(Argentina)–Salto(Uruguay) border (about 500 km2), where the GAS has been explored for hydrothermal tourism (temperature range from 44 to 48 °C). The total groundwater withdrawal of the GAS is estimated to be 1.04 km3 day−1 (Foster et al. 2009).

The outcropping sandstones cover approximately 10 % of the aquifer area; however, these areas are critically important because they are responsible for almost all the aquifer recharge (OAS/GEF 2009). As recharge rates are used to estimate groundwater resources and potential water withdrawal (Dages et al. 2009), these outcropping areas must be studied and protected against unsustainable land uses, and soil contamination and sealing. Data from hydrological monitoring networks are often unavailable or unpublished, and recharge in the GAS outcropping is one of the least known hydrological variables (Rabelo and Wendland 2009; Gómez et al. 2010; Lucas and Wendland 2012).

The goal of this study was to assess the suitability of using RS data in the water-budget equation for estimating and evaluating recharge inter-annual patterns in a representative GAS outcropping area. The remote sensing approach was compared to the water-table fluctuation (WTF) method (Healy and Cook 2002) and a water-budget method using ground-based measurements for the period from 2004 to 2012.

Study site description and data sources

The study site is an upland flat watershed (65 km2) called Onça Creek (Fig. 1), located in southeastern Brazil (22°10′ to 22°15′ south and 47°55′ to 48°00′ west) in the central region of the state of São Paulo. Because the Onça Creek watershed presents representative hydrogeological features and land uses of other GAS outcrop areas (Wendland et al. 2007), it has been chosen as an experimental watershed.

Fig. 1
figure 1

Location of the Onça Creek watershed, showing the monitoring wells, climatological ground station (CREAH/USP) and sampling points of specific yield (S y )

Full size image

The topographic elevation of the Onça Creek watershed varies between 840 and 640 m above mean sea level (msl). This watershed is dominated by a low average slope steepness of 0.076 m m−1 (<1 %). Onça Creek is 16.0 km in length and the compactness coefficient (defined as the ratio of perimeter of the watershed to circumference of a circle, which equals the drainage area; Wisler and Brater 1959) of this watershed is 1.47. Based on water-level measurements in the monitoring wells, groundwater flow is topographically controlled and flows from recharge areas towards the river. One should note that there is no groundwater pumping in the study area.

Quaternary-age sediments (weathering sandstone; Wendland et al. 2007) cover the Onça Creek watershed. The hydraulic conductivity of the Quaternary-age soil varies from 1.0 × 10−5 to 7.1 × 10−6 m s−1. This soil has a fine-sand (66 %), course-sand (20 %) and silt-clay (14 %) texture, allowing minimal surface runoff. Onça Creek flows mainly over sandstone of the Botucatu Formation (eolian sandstones of the Jurassic period), while at the basin outlet it flows over the Botucatu-basalt complex (Rabelo and Wendland 2009).

Mean (±standard deviation, SD) annual rainfall (for the 2004–2012 period) was 1,531 mm(±216 mm). For the same period the seasonal rainfall obtained from monthly averages for the summer (December–February), fall (March–May), winter (June–August) and spring (September–November) was, respectively, 256 mm(±117 mm), 97 mm(±72 mm), 44 mm(±51 mm), and 106 mm(±50 mm). According to the Köppen climate classification system the climate in the region is humid subtropical (Cwa; Wendland et al. 2007). Mean monthly temperature varies from approximately 24 °C in the summer to 18 °C in the winter.

The native vegetation in the Onça Creek watershed is woody savannah called Cerrado, which is present in several regions of South America. Following the replacement of natural vegetation by agriculture, this watershed presents various land uses such as eucalyptus, sugarcane, citrus, and grassland. Eleven monitoring wells were drilled to depths between 10 and 50 m. Well screen depth varies depending on location, but most are nearly 25 m below the terrain. Groundwater levels usually are deeper than 5 m.

Remote sensing for estimation of rainfall and evapotranspiration

Two types of remote sensing datasets were used for the study: The first one is the Tropical Rainfall Measuring Mission (TRMM; Huffman et al. 2007) and the second is the Moderate Resolution Imaging Spectroradiometer (MODIS) product MOD16 (Mu et al. 2011). The study was conducted from the 2004–2005 to the 2011–2012 water years (October–September).

The rainfall product from the TRMM (Version 7) Multisatellite Precipitation Analysis (TMPA) algorithm was used. It was developed by the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC) and provides rainfall estimates at spatial and temporal scales of 0.25° × 0.25° and 3 h between 50° north and 50° south respectively (Huffman et al. 2007). TRMM has a rainfall radar, passive microwave imager, and nine-channel microwave radiometer system to get rainfall data (Kummerow et al. 1998; Prakash et al. 2013). The TMPA research products, V7 (3B42), at daily time scales were obtained from NASA (2014) and they were accumulated on a monthly and an annual basis.

Evapotranspiration (ET) product MOD16 is estimated from MODIS satellite observations and daily meteorological inputs using an algorithm to solve the Penman-Monteith equation (Mu et al. 2011). ET product MOD16 data are available at 1-km2 spatial resolution with a temporal resolution of 8-days and accumulated on a monthly and annual basis. The MOD16 product was obtained from the Numerical Terradynamic Simulation Group website for the study period, available at NTSG (2014).

Ground data

Climatological ground-measured data were provided by the Center for Water Resources and Applied Ecology of the University of São Paulo (CRHEA/USP). A conventional climatological station located approximately 1.5 km outside the study area (Fig. 1) was regularly monitored. Ground rainfall data were collected using a Ville de Paris rain gauge. The solar radiation at land surface, wind speed, sunshine duration and air temperature were recorded using, respectively, an actinometer, a hemispherical cup anemometer, a Campbell-Stokes recorder, and glass thermometers filled with mercury and alcohol.

The water-level depth in 11 monitoring wells was measured manually every 15 days, providing data on water-table fluctuation. The specific yield (S y) at the Onça Creek watershed (Table 1) was determined during campaigns to collect undisturbed soil samples at five locations and at 10 different depths that correspond to the water-table fluctuation (Wendland et al. 2015). The undisturbed samples were analyzed using the Haines funnel technique (Haines 1930) as described by Wendland et al. (2015).

Table 1 Descriptive statistics of specific yield in the Onça Creek watershed
Full size table

The mean(±SD) S y value used in the WTF method was 12 %(±2.9 %) and it is considered to be spatially representative throughout the watershed. The values determined in the laboratory for undisturbed soil samples were consistent with those in the literature (Healy and Cook 2002; Johnson 1967), which range between 10 and 28 % for the same textural class. Fetter (1994) showed S y values between 15 and 32 % for medium sandy soils. Tizro et al. (2012) reported an average S y estimate of 15 % using geoelectrical measurements (vertical electrical soundings) for sandy clay in the aquifer of Mahidashat plain, west Iran.

Methods

Two methods and three different data sources were used to determine groundwater recharge in an unconfined aquifer in the Onça Creek watershed (Table 2). Recharge estimates using RS products and climatological ground data, R sat and R gr, respectively, were based on a simple water-budget equation (Szilagyi et al. 2011; Khalaf and Donoghue 2012). The control volume extends from land surface to the water table and the groundwater storage changes were neglected. Over a long period of typically several years, aquifer storage tends to remain constant in the absence of significant climate change (Healy 2010; Szilagyi et al. 2013). Since surface runoff (R off) can be negligible in sandy soils and/or flat topography (Brunner et al. 2007), R sat and R gr were calculated as the difference between annual rainfall (P) and evapotranspiration (ET; Szilagyi et al. 2013):

Table 2 Summary description of the recharge methods used in the Onça Creek watershed (Adapted from Scanlon et al. 2002)
Full size table
$$ {R}_{\mathrm{sat}}\approx P-\mathrm{E}\mathrm{T} $$
(1)

In Eq. (1), groundwater ET was ignored because the water table is deeper than 5.0 m below land surface, and the capillarity effect is insufficient to raise that height in a sandy soil (Wendland et al. 2007). The R sat represents the potential recharge estimates or drainage (Healy 2010) and their negative values reflect a water deficit in soil and was accounted for in Eq. (1) for steady-state conditions. As mentioned earlier, there is no groundwater pumping in the watershed.

Recharge rates were also estimated by Eq. (1) using daily climatological ground data. Potential evapotranspiration (PETo) was calculated using the complete Food and Agriculture Organization of the United Nations (FAO) modification of the Penman-Monteith equation (FAO56-PM; Allen et al. 1998). The daily ET values were estimated as

$$ \mathrm{E}\mathrm{T}={K}_{\mathrm{c}}{\cdot \mathrm{PET}}_{\mathrm{o}} $$
(2)

The ET values were accumulated on a monthly and an annual basis. K c is a coefficient for each crop, as indicated in Table 3. Average K c values were used for eucalyptus, sugarcane, citrus and grassland for the ET calculations for the 2004–2005 to the 2011–2012 water years. The mean ET for the entire watershed was obtained from the weighted averages for the different crop, considering their respective land use areas.

Table 3 Single crop coefficients for each land use in the Onça Creek watershed, for use with FAO Penman-Monteith equation (Allen et al. 1998)
Full size table

The WTF method was employed to estimate recharge from 2004 to 2012 using biweekly measurements of the water-table elevation. The WTF method evaluates the change in water-table position (if any) following a rain event and, thus, provides an estimate of total recharge (Callahan et al. 2012). The WTF recharge (R wtf) is calculated as follows (Healy and Cook 2002):

$$ {R}_{\mathrm{wtf}}\approx {S}_{\mathrm{y}}\cdot \frac{\Delta H}{\Delta t} $$
(3)

where ΔH is the difference between the peak of the rise and low point of the extrapolated antecedent recession curve (EARC) at the time of the peak. The power law function (Wendland et al. 2007) was used to extrapolate the water-table recession curve, since there is no groundwater pumping in or near the monitoring wells. Evapotranspiration from the water table is negligible due to the thickness of the unsaturated zone (>5.0 m). Changes in the atmospheric pressure were assumed to be minimal during the study. Errors associated with the S y and EARC contributes to the overall uncertainty of the WTF estimates.

Uncertainty on the water-budget components

Ten eddy covariance flux tower sites were used to evaluate the ET from MOD16 in different land uses and land covers (tropical rainforest, tropical dry forest, selective logged forest, seasonal flooded forest, pasture, cropland and Cerrado; Loarie et al. 2011). An uncertainty of 13 % in MOD16 ET (U ET) was found in pasture/agriculture areas (Loarie et al. 2011). Since the uncertainty in FAO56-PM ET could not be evaluated the R gr, uncertainty is not accounted for here.

To validate the TRMM V7 data, rainfall ground measurements were used. Monthly statistical BIAS from the TRMM data was determined by subtraction of the rain gauge precipitation (see Eq. 5). The mean BIAS (2004–2012 period) of each month was used for the monthly TRMM correction—see Table S1 of the electronic supplementary material (ESM). The uncertainty of TRMM P (U P) is given by the standard deviation of the annual BIAS (TRMM – rain gauge):

$$ {U}_{\mathrm{P}}=\frac{\sqrt{{\left(\mathrm{BIAS}\right)}^2}}{N-1} $$
(4)

where, N is the number of rainfall events per month. For each water year, statistical metrics such as the root-mean-square error (RMSE) function, Pearson’s coefficient of correlation (r) and BIAS were calculated for TRMM P. The rain gauge reference was the CRHEA climatological station. The statistics of TRMM P were computed as (Moazami et al. 2014)

$$ \mathrm{BIAS}={\displaystyle {\sum}_{i=1}^N\frac{\left({P}_{{\mathrm{RS}}_i}-{P}_{{\mathrm{o}}_i}\right)}{N}} $$
(5)
$$ \mathrm{RMSE}=\sqrt{\frac{{\displaystyle {\sum}_{i=1}^N{\left({P}_{{\mathrm{RS}}_i}-{P}_{{\mathrm{o}}_i}\right)}^2}}{N}} $$
(6)

where, \( {P}_{{\mathrm{RS}}_i} \) and \( {P}_{{\mathrm{o}}_i} \) are the monthly values of TRMM rainfall and rain gauge observations, respectively and the index i is the number of months. The R sat uncertainty (U Rsat) was calculated for each water year by applying the error propagation equation (Taylor 2012) as the quadratic sum of the MOD16 ET and TRMM P uncertainties:

$$ {U}_{{\mathrm{R}}_{\mathrm{sat}}}=\sqrt{{\left({U}_{\mathrm{ET}}\right)}^2+{\left({U}_{\mathrm{P}}\right)}^2} $$
(7)

A single value of U Rwtf (%) was applied as the lower and upper limits of the R wtf uncertainty for each water year.

Uncertainty of the recharge estimates

Uncertainty associated with the WTF method (U Rwtf) is linked to the difficulty in determining a representative specific yield (Coes et al. 2007). The standard deviation of all S y values in the watershed (Table 1) was considered as a measure of uncertainty (\( {U}_{{\mathrm{S}}_{\mathrm{y}}} \)) to compute the U Rwtf. The standard deviation of S y is subjected to limitations because it does not account for the uncertainty that is inherent in the individual measurements of S y. Since the soil physical properties vary with depth below land surface and position within the watershed, specific yield tends to vary.

The fractional uncertainty (Taylor 2012) of S y was used to calculate U Rwtf (%) as:

$$ {U}_{\mathrm{Rwtf}}=\frac{U_{S_{\mathrm{y}}}}{{\overline{S}}_{\mathrm{y}}}\cdot 100 $$
(8)

A single value of U Rwtf (%) was applied as the lower and upper limits of R wtf uncertainty for each water year during the study period.

Results and discussion

Rainfall and evapotranspiration based on remote sensing and ground data

Figure 2 shows the scatter plots of monthly TRMM P against ground data, which exhibited good agreement (r = 0.93 and RMSE = 42 mm month−1). The result of TRMM P after seasonal BIAS correction was r = 0.93 and RMSE = 41 mm month−1. Oliveira et al. (2014) found a RMSE value of 53.58 mm month−1 using TRMM P data in the Brazilian Cerrado biome area. The BIAS-corrected TRMM improves P, for example about 295 mm in 2007–2008, in comparison with the original TRMM data in the water year basis (Table 4). TRMM P tends to have a systematic overestimation BIAS (December–March) of 60 mm over the 2004–2012 study period. The uncertainty of TRMM P for the entire period of study was 183 mm year−1.

Fig. 2
figure 2

Monthly scatter plots between: a TRMM-3B42 V7 and rain gauge data. b TRMM-unbiased (after BIAS correction) and rain gauge data. The red solid line shows the ideal correlation (1:1)

Full size image
Table 4 Comparing annual rainfall (P) TRMM V7 with rain gauge in the Onça Creek watershed. Annual MODIS ET (product MOD16) and FAO56-PM actual evapotranspiration data are presented
Full size table

The TRMM P revealed strong inter-annual variability, with a maximum value during the rainy season (December–March) ranging from 822 mm in 2006–2007 to 1,167 mm in 2010–2011 (Fig. 3). There is some expected inter-annual discrepancy between the ground P data and the TRMM P data. For example, the rain gauge had increasing P values from 2008 to 2009 to 2010–2011, whereas the TRMM data had decreasing P values in this same period. As shown in Fig. 3, the higher positive R sat occurred in the summer (December = 156 mm, January = 231 mm and February = 81 mm), while little R sat occurred in the winter (June = −4 mm, July = −8 mm and August = −12 mm) from 2004 to 2012.

Fig. 3
figure 3

Monthly MOD16 evapotranspiration (ET), TRMM rainfall (P) and groundwater recharge (R sat ) over the 2004–2012 period. Negative values of R sat indicated soil water deficit

Full size image

By comparison, the MOD16 ET presented less inter-annual variation, with a minimum rainy season of 428 mm in 2005–2006 and a maximum value of 476 mm in 2010–2011 (Table 4). The MOD16 ET showed a maximum inter-annual difference of 156 mm between 2006 and 2007 and 2007–2008. Following Anderson et al. (2012), this result suggests that R sat inter-annual variability is more linked to the TRMM P than to MOD16 ET. However, because MOD16 ET could not be validated with ground data, it is not possible to confirm that ET may be under or overestimated. Furthermore, the annual uncertainty in the MOD16 ET was lower (130 mm year−1) than the TRMM P (183 mm year−1).

ET estimated using the FAO56-PM and K c coefficients ranged from 947 mm year−1 in 2007–2008 to 1,125 mm year−1 in 2011–2012 (Table 4). The mean annual ET estimated by using FAO56-PM was 1,046 mm year−1 and using MOD16 ET was 1,001 mm year−1 over the study period, which is a difference of less than 5 %. Comparing the differences between the ET estimated using the FAO56-PM (and K c coefficient) and the MOD16 ET, the farthest and closest result were, respectively, 172 mm year−1 in 2006–2007 and 5 mm year−1 in 2009–2010 (Table 4).

Results indicate that grassland annual FAO56-PM ET ranged from 672 mm (in 2008–2009) to 806 mm (in 2011–2012) and exhibited a mean(±SD) of 721 mm (±42 mm) during the study period. Sugarcane annual ET varies from 895 mm (in 2008–2009) to 1,074 mm (in 2011–2012) and showed a mean(±SD) of 961 mm(±56) mm. Eucalyptus annual ET ranged from 970 mm (in 2007–2008) to 1,343 mm (in 2011–2012) and presented a mean(±SD) of 1,174 mm(±108) mm. These results were similar to those reported in previous studies. For example, a summary compilation of several ET studies reported a value of 756 mm for tropical grassland (Schlesinger and Jasechko 2014). A value of about 950 mm year−1 (a daily ET value of 2.6 mm) was found using an Eddy Covariance System and meteorological sensors in a grassland (Brachiaria brizantha) area of Brazil (Meirelles et al. 2014). The ET of 1,124 and 1,235 mm year−1 was reported, respectively, for 1- and 2-year eucalyptus (grandis and urophylla; Cabral et al. 2010).

Analysis of recharge temporal series

The minimum and maximum R wtf estimates for the study period were 0 and 846 mm year−1, respectively (Table 5). The minimum and maximum annual mean of R wtf were 116 mm (10 % of rain gauge P) in 2006–2007 and 562 mm (31 % of rain gauge P) in 2010–2011.

Table 5 Annual WTF recharge (R wtf) estimates in the Onça Creek watershed over the study period
Full size table

There is a poor multi-annual agreement between R sat and R wtf (Fig. 4). The mean(±U Rwtf) of R wtf was close at 311 mm (±75 mm) and the coefficient of variation was 0.50 over the 8-year study period. The uncertainty of R wtf corresponds to a value of 24 % of the annual mean R wtf, which is consistent with other results (Maréchal et al. 2006). Maréchal et al. (2006) computed the error for the water-budget components at the watershed scale, and reported a relative error of 22–24 % in the recharge estimates.

Fig. 4
figure 4

Recharge estimates using TRMM and MOD16 data (R sat ), ground data (R gr ), and WTF method (R wtf ) in the Onça Creek watershed for the 2004–2012 period. Red shading indicates uncertainty in the WTF-estimates (U Rwtf ), while green shading shows uncertainty in the recharge using TRMM and MOD16 ET data (U Rsat )

Full size image

On an annual basis, the uncertainty of the R wtf estimates overlapped the uncertainty of R sat in three water years (2006–2007, 2007–2008 and 2010–2011). The closest results between the annual R sat and R wtf estimates were 113 mm in 2010–2011 and 149 mm in 2007–2008 (Table 6). The profile of the R wtf temporal series was similar that of the R sat series between 2004 and 2005 and 2007–2008 and exhibits different general trends from 2008 to 2009 to 2011–2012.

Table 6 Recharge statistics over 2004–12 period
Full size table

The estimates of R sat based on remote sensing data in the GAS outcrop area show a good multi-annual (8-years) agreement with R gr using ground-based data (Fig. 4). The mean(±SD) of R sat and R gr were similar at 537 mm(±224 mm) and 469 mm, respectively over the study period. On an annual basis, however, there is some discrepancy between the R sat and R gr estimates. The R sat estimates range from 160 mm year−1 in 2007–2008 to a maximum of 681 mm year−1 in 2004–2005, while the R gr estimates increase from 161 mm year−1 in 2005–2006 to a maximum of 692 mm year−1 in 2010–2011 (Fig. 4). Other than water years 2010–2011 and 2011–2012, the profile of the R sat and R gr temporal series exhibited the same general trends (Fig. 4). The greater agreement between annual R sat and R gr values were 63 mm in 2010–2011 and 86 mm in 2006–2007 (Table 6).

If groundwater recharge, R sg, had been estimated solely as the difference between rain gauge rain and MOD16 ET, with an uncertainty P value of 10 %, the mean(±SD) of the R sg estimate would be 513 mm(±200 mm) over the 2004–2012 period. The R sg estimates overlapped both the R sat and R wtf estimates at annual scale and showed closer results than the R sat in comparison with the R wtf (Fig. 5). Moreover, the R wtf was positively correlated with both R sg (r = 0.51) and R gr (r = 0.54) estimates. The latter result demonstrated the potential use of MOD16 ET instead of FAO56-PM, which requires an extensive in situ climatological network.

Fig. 5
figure 5

Recharge estimates using TRMM and MOD16 data (R sat ), MOD16 and rain gauge data (R sg ), and the WTF method (R wtf ) in the Onça Creek watershed for the 2004–2012 water years. Red shading indicates uncertainty in the WTF-estimates (U Rwtf ), green shading shows the recharge uncertainty using TRMM and MOD16 data (U Rsat ), and blue shading indicates the uncertainty using rain gauge and MOD16 ET data (U Rsg )

Full size image

Conclusions

The suitability of using remote sensing data on recharge estimation was evaluated in a representative outcrop area of the GAS. Recharge methods show that most significant recharge occurs in the rainy season (from December to March). The estimates of R sat based on remote sensing data in the GAS outcrop area shows a good multi-annual (8-year) agreement with R gr based on ground-based data. Over the entire study period, the mean(±SD) R sat, R gr and R wtf were similar, respectively at 537 mm(±224 mm), 469 mm and 311 mm(±75 mm). The mean recharge between R sat, R gr and R wtf was 439 mm year−1 (about 29 % of the mean of rainfall over the 2004–2012 period) for the entire watershed over the study period.

Results indicated a good agreement between ET calculated from FAO56-PM with a K c coefficient and MOD16 ET. This result indicates that MOD16 ET has a great potential of applicability in this GAS outcrop area. At present, TRMM is not precise enough to use for estimating groundwater recharge with a simple water budget in the Onça Creek watershed, because it tends to have a systematic upward BIAS on an annual basis. This large TRMM BIAS explains the main discrepancy among inter-annual R sat and other recharge estimates employed here. On the other hand, TRMM is useful for monthly hydrological applications in the study area.

Results demonstrate the need of applying multiple methods (three at least) for estimating recharge. Since accurate and precise recharge estimation still is uncertain, the recharge satellite-based results presented here are considered acceptable in the Onça Creek watershed. Future studies should take into account the addition of more water-budget components (for example, surface runoff and water storage) to obtain more realistic R sat and R gr estimates.

As remotely sensed data have improved in spatial, temporal and spectral resolution, they have been identified as a useful tool for evaluating hydrologic systems. Results provide the first insight about an intercomparison of water budgets generated from remote sensing and measured data used to estimate recharge in the GAS. These results should be interesting for future studies on assessing recharge in the GAS outcrop zones where data are scarce or nonexistent.