An Improved Model for the Evaluation of Groundwater Recharge Based on the Concept of Conservative Use Potential: A Study in the River Pandeiros Watershed, Minas Gerais, Brazil

: Water resources have been increasingly impacted due to the growth of water demand associated with environmental degradation. In this context, the mapping of groundwater recharge potential has become attractive to water managers as it can be used to direct public policies and conserve this natural asset. The present study modiﬁes (improves) a spatially explicit model to determine groundwater recharge potential at the catchment scale, testing it in the Pandeiros River basin located in the state of Minas Gerais, Brazil. The model is generally based on the water balance approach and the input variables were compiled from institutional sources and processed in a Geographic Information System. The novelty brought by the aforementioned modiﬁcation relates to the coupling of physical variables (conventional way) and land management practices (introduced here) in the estimation of a percolation factor. The role of land management practices for percolation was assessed by the so-called Conservative Use Potential (PUC) method, which classiﬁes the areas of a river basin in terms of their potential for sustainable use. The results were validated by an independent method, namely the recession curve method based on the interpretation of hydrographs. In general, the groundwater recharge potential is favored in ﬂat to gently undulating areas and forested regions, as well as where the landscape is characterized by well-structured soils, good drainage conditions and large hydraulic conductivity. The map of groundwater recharge potential produced in this study can be used by planners and decision makers in the Pandeiros River basin as a tool to achieve sustainable use of groundwater resources and the protection of recharge areas.


Introduction
Clean water resources are becoming scarcer due to the increasing demand for its use and to environmental degradation [1]. It has been estimated that 1/3 of all countries will have to adapt  According to the Brazilian Institute for Geography and Statistics (2018), the municipalities comprising the PRB have a total population of 86,311 inhabitants, most of them living in Januária. The largest part of this population live in rural areas, especially in the municipalities of Bonito de Minas and Cônego Marinho (Table 1) [48]. The PRB is in an area with transitional vegetation, presenting phytophysionomies of Cerrado and Caatinga biomes [49]. This ecotone is characterized by swamp regions that contain the springs of the São Francisco River. These springs are responsible for the reproduction of most fishes that live between the Três Marias (MG) and Sobradinho (BA) dams [50].
The relief is predominantly flat. The plain was formed by the filling of the São Francisco Depression with sediments that were sourced from the erosion of rocks from the São Francisco Plateau [51]. This process is also responsible for a small proportion of steep slope areas in the basin, as shown in Figure 2 and quantified in Table 2 [52].
Water 2020, 12, x FOR PEER REVIEW 5 of 24 According to the Brazilian Institute for Geography and Statistics (2018), the municipalities comprising the PRB have a total population of 86,311 inhabitants, most of them living in Januária. The largest part of this population live in rural areas, especially in the municipalities of Bonito de Minas and Cônego Marinho (Table 1) [48]. The PRB is in an area with transitional vegetation, presenting phytophysionomies of Cerrado and Caatinga biomes [49]. This ecotone is characterized by swamp regions that contain the springs of the São Francisco River. These springs are responsible for the reproduction of most fishes that live between the Três Marias (MG) and Sobradinho (BA) dams [50].
The relief is predominantly flat. The plain was formed by the filling of the São Francisco Depression with sediments that were sourced from the erosion of rocks from the São Francisco Plateau [51]. This process is also responsible for a small proportion of steep slope areas in the basin, as shown in Figure 2 and quantified in Table 2 [52].    As regards the local geology, there is the presence of alluvial deposits over the main drains in the area and in wetland regions (Veredas), which are the results of natural and anthropogenic erosion Water 2020, 12, 1001 6 of 24 processes with consequent transport and deposition of sediments. Figure 3 and Table 3 show the spatial distribution and proportion of lithotypes in the studied basin, respectively.
Water 2020, 12, x FOR PEER REVIEW 6 of 24 As regards the local geology, there is the presence of alluvial deposits over the main drains in the area and in wetland regions (Veredas), which are the results of natural and anthropogenic erosion processes with consequent transport and deposition of sediments. Figure 3 and Table 3 show the spatial distribution and proportion of lithotypes in the studied basin, respectively.  The Soil Map of Minas Gerais [53] presents an area predominantly composed of red-yellow latosols, which cover more than 87% of the hydrographic basin, followed by fluvic neosols (5.48%). The occurrence of quartzarenic neosols, litolic neosols, cambisols and melanic gleysols are related to much smaller areas ( Figure 4 and Table 4).
The spatial distribution of the land use and cover classes in the basin ( Figure 5) shows the predominance of a typical Cerrado vegetation (savanna) of low to medium size [54], which is usually associated with the occurrence of latosols and sandstone regions. These are found mainly in the central part, towards the northern and northeast of the Pandeiros River basin. This phytophysiognomy covers 183,719.88 ha in the basin, representing 46.3% of its area ( Table 5). The second most significant land use and cover class in the study area is the dense Cerrado (21.5%), followed by the sparse Cerrado (10.42%).  The Soil Map of Minas Gerais [53] presents an area predominantly composed of red-yellow latosols, which cover more than 87% of the hydrographic basin, followed by fluvic neosols (5.48%). The occurrence of quartzarenic neosols, litolic neosols, cambisols and melanic gleysols are related to much smaller areas ( Figure 4 and Table 4).     The spatial distribution of the land use and cover classes in the basin ( Figure 5) shows the predominance of a typical Cerrado vegetation (savanna) of low to medium size [54], which is usually associated with the occurrence of latosols and sandstone regions. These are found mainly in the central part, towards the northern and northeast of the Pandeiros River basin. This phytophysiognomy covers 183,719.88 ha in the basin, representing 46.3% of its area ( Table 5). The second most significant land use and cover class in the study area is the dense Cerrado (21.5%), followed by the sparse Cerrado (10.42%).

Material
The materials used in this study (Table 6) consisted of (i) a digital elevation model (ALOS PALSAR), with a spatial resolution of 12.5 meters; (ii) a land use and cover map (scale of 1: 25.000); (iii) the soil map of Minas Gerais state (scale of 1: 600.000); (iv) values of groundwater recharge calculated by the PUC method and the hydraulic conductivity of each soil class in the basin; (v) rainfall and evapotranspiration data from meteorological stations located in municipalities near the basin; and vi) flow data records from the station code 45,250,000 located near the mouth of the Pandeiros River.

Material
The materials used in this study (Table 6) consisted of (i) a digital elevation model (ALOS PALSAR), with a spatial resolution of 12.5 meters; (ii) a land use and cover map (scale of 1: 25.000); (iii) the soil map of Minas Gerais state (scale of 1: 600.000); (iv) values of groundwater recharge calculated by the PUC method and the hydraulic conductivity of each soil class in the basin; (v) rainfall and evapotranspiration data from meteorological stations located in municipalities near the basin; and vi) flow data records from the station code 45,250,000 located near the mouth of the Pandeiros River.

Methodology
The spatially explicit groundwater recharge potential in the Pandeiros River basin was evaluated using the methodology proposed by Costa et al. (2019) [10], with the adjustments described below. The workflow was divided into 5 main steps: (i) acquisition of a land use and cover map, digital elevation model, soil type map, and climate data (rainfall and evapotranspiration); (ii) calculation of surface runoff using the slope length and steepness factor, and runoff coefficients for the land use and cover types; (iii) calculation of water percolation based on the PUC method [43] (replacing the effective porosity of Costa's approach and representing the proposed methodological improvement) and adapted hydraulic conductivity values with fuzzy logic [55,56]; (iv) calculation of groundwater recharge in different points of the basin and a mean value for the whole basin, using a geographical information system; (v) validation of results by comparing the previously calculated mean groundwater recharge with the value estimated by the hydrograph recession curve analysis ( Figure 6).

Methodology
The spatially explicit groundwater recharge potential in the Pandeiros River basin was evaluated using the methodology proposed by Costa et al. (2019) [10], with the adjustments described below. The workflow was divided into 5 main steps: (i) acquisition of a land use and cover map, digital elevation model, soil type map, and climate data (rainfall and evapotranspiration); (ii) calculation of surface runoff using the slope length and steepness factor, and runoff coefficients for the land use and cover types; (iii) calculation of water percolation based on the PUC method [43] (replacing the effective porosity of Costa's approach and representing the proposed methodological improvement) and adapted hydraulic conductivity values with fuzzy logic [55,56]; (iv) calculation of groundwater recharge in different points of the basin and a mean value for the whole basin, using a geographical information system; (v) validation of results by comparing the previously calculated mean groundwater recharge with the value estimated by the hydrograph recession curve analysis ( Figure  6).  [10]. Abbreviations: LS-slope length and steepness factor [57]; Crunoff coefficient (Table 7); RF-runoff factor (Equation (1)); PUC-Conservative Use Potential [43]; Ks-soils' hydraulic conductivity (Table 8); PF-water percolation factor (Equation (2)).  [10]. Abbreviations: LS-slope length and steepness factor [57]; C-runoff coefficient (Table 7); RF-runoff factor (Equation (1)); PUC-Conservative Use Potential [43]; Ks-soils' hydraulic conductivity (Table 8); PF-water percolation factor (Equation (2)).
In the first step, the mean annual rainfall and evapotranspiration of municipalities near the study area were estimated using data from the Brazilian Institute for Meteorology (INMET) relative to the 2009-2018 period, which were obtained from meteorological stations in Arinos (MG), Januária (MG), Montes Claros (MG), Salinas (MG), Cariranha (BA), Espinosa (MG), Formoso (MG), Posses (GO), and Brasília (DF). Longer temporal series would be more adequate for estimating groundwater recharge. However, these records were not available. The flaws in the temporal series were resolved by using the regional weighting method, and the information was interpolated through inverse distance weighting (IDW) raised to the power of two [58]. This method was used because it raises the importance of closer stations in the interpolation. The data were spatialized and trimmed up to the study area limits.
In the second step, the land use and cover map and the topographical information from the digital elevation model (slope length and steepness factor-LS factor) were used to calculate the surface runoff factor, based on the method proposed by Böhner and Selige (2002) [59]. The runoff factor was calculated according to Equation (1), reproduced from Costa et al. 2019 [10].
where RF is the runoff factor (dimensionless); C is the runoff coefficient (dimensionless values adopted from Table 7); LS FUZZY is the slope length and steepness factor estimated using the method of Desmet and Govers (1996) [57], but recast to the 0-1 range using a fuzzy logic algorithm (the steeper the slope the closer to 1).

Land and Cover Class Runoff Coefficient
Anthropized areas 0.50 Urban areas 0.85 Forest formation 0.10 Exposed soil and ground vegetation 0.60 Wetland regions ("Veredas") 0.09 In the third step, the water percolation was calculated considering the soil classes in the basin [53]. A systematic literature review was performed to set up hydraulic conductivity values for each soil class, at the second category level of Brazilian soil classification system (SiBCS) [61].   [43]. The PUC is a method that allows for mapping areas of a basin based on their limitations and potentialities for conservationist land use, through the combined assessment and weighting of several environmental variables (soils, geology, and geomorphology) [43].
The PUC assigns values from 1 to 5 to the different classes of lithology, slope and soil in the watersheds from the state of Minas Gerais, Brazil. The analyses are focused on groundwater recharge, agricultural use potential and soil resistance to erosion in the catchments [43]. For the identification of lithologies, slopes and soil classes existing in Minas Gerais, the available official databases are used.
The attribution of grades for the different types of lithologies took into consideration their potential to provide nutrients (greater weight for rocks with higher absolute content of essential macro elements to plants) and their susceptibility to weathering processes (considering the main mineral constituents and scored according to their resistance to weathering, based on Goldish's stability [63]). Regarding the slope parameter, the same weights were attributed for groundwater recharge potential for agricultural use and resistance to erosion. For groundwater recharge, it was considered that the slope has a direct relationship with the flow velocity and the opportunity time for water infiltration. The higher the slope, the higher the water velocity and the shorter the time for water infiltration. In this context, the mountainous relief received a weight 1 and the flat relief a weight 5.
For the attribution of grades to different soils, the variables texture, drainage, effective depth and fertility were considered. For groundwater recharge, the attribute fertility was disregarded and the classes of soils characterized as favorable to water infiltration and percolation received greater weight.
The recharging potential for each soil type was obtained by the simple average of the values of texture, drainage and effective depth, normalized so that the final scale was in the range of 1 to 5.
In this work, only the soil parameter of groundwater recharge was used, which assigns the basin´s soil classes a score from 1 to 5. This parameter takes into account the effective depth, texture and drainage of each soil class at the first category level of SiBCS [61]. Thus, to the soil that presents considerable effective depth, satisfactory drainage and texture favorable to infiltration, a higher score regarding groundwater recharge is attributed.
However, in order to adapt the PUC method to the recharge model, a rescaling process was implemented, whereby the ratings from 1 to 5 were recast to the range 0 to 1. The values used in the model are represented in Table 8. The percolation factor was evaluated using Equation (2), reproduced from Costa et al. 2019 [10]: where PF is the water percolation factor (dimensionless); Ks FUZZY (dimensionless) is the soil's hydraulic conductivity fitted to the 0 to 1 range by the fuzzy logic algorithm (the higher the soil's hydraulic conductivity in the class the closer to 1); PUC are the scores of groundwater recharge set up by the PUC method but fitted to the 0 to 1 range.
In the fourth step, the groundwater recharge potential was calculated for each point in the basin, using Equation (3), according to Costa et al. 2019 [10]: where RPot is the groundwater recharge potential (m 3 ha −1 year −1 ); P is the mean annual rainfall depth (mm year −1 ); ET R is the mean evapotranspiration (mm year −1 ); RF is the surface runoff factor; and PF is the water percolation factor. The results were validated in the fifth step through comparison of the calculated mean groundwater recharge with a homologous median value estimated by the hydrograph recession method based on the Maillet equation [64,65]. The hydrograph for the Pandeiros River was drawn from daily average stream flow data measured at the hydrometric station nº 4425000 located in the Pandeiros River mouth (Pandeiros River Dam), compiled from the Hidroweb portal [66]. Streamflow data from 2013 to 2018 were the most recent and more continuous in the historical series, showing no flaws or absent data. Thus, these data were used to calculate the Maillet equation (Equation (4)). To separate and analyze the recession curve and the recession days, the methodology proposed by Barnes (1939), Dewandel et al. (2003) and Kovacs et al. (2005) was used [64,65,[67][68][69]: where Q T is the flow at time t (m 3 s −1 ); Q 0 is the flow at the beginning of a recession (m 3 s −1 ); α is the coefficient of recession; t is the time (days); and e is the basis of Neperian logarithm (2.71828). Thus, the coefficient of recession can be determined numerically, based on the logarithmic form of Equation (4), represented and rearranged in Equation (5): Subsequently, the groundwater recharge volume was calculated using Equation (6): where V is the recharge volume (m 3 ); Q 0 is the flow at the beginning of recession (m 3 s −1 ); t' is the converter of the t unit (days into seconds; 86,400); α is the coefficient of recession (dimensionless).
The constant of recession (α; Equation (5)) is dependent on the aquifer characteristics and therefore should not vary significantly from year to year. On a Q versus t plot (hydrograph), where the Q values are represented in logarithmic scale and the values are in linear scale, the baseflow within a hydrologic year (from the recharge period to the end of recession) should define as a straight line, the slope to which is related in terms of α. If t cycle is the time of a log cycle for discharge, meaning the time for discharge to change from 1 to 10 m 3 /s, from 10 to 100 m 3 /s, and so forth, then LogQ 0 − LogQ t = 1 and α = 1/(0.4343t cycle ). This simplified representation of Equation (5) is frequently used in the calculation of α and will be adopted in the present study. Conversely, the values of Q 0 can vary in response to the annual variations of precipitation. In this case, a value of Q 0 should be calculated for each hydrologic year, while mean ± standard deviation values are derived therefrom.

Results
The interpolation of precipitation data from 2009 to 2018, obtained from climatologic stations close to the Pandeiros River basin, showed mean rainfall depths ranging from 904.7 to 1056.3 mm year −1 (Figure 7). = 0.4343 Subsequently, the groundwater recharge volume was calculated using Equation (6): where V is the recharge volume (m 3 ); Q0 is the flow at the beginning of recession (m 3 s -1 ); t' is the converter of the t unit (days into seconds; 86,400); α is the coefficient of recession (dimensionless). The constant of recession (α; Equation (5)) is dependent on the aquifer characteristics and therefore should not vary significantly from year to year. On a Q versus t plot (hydrograph), where the Q values are represented in logarithmic scale and the values are in linear scale, the baseflow within a hydrologic year (from the recharge period to the end of recession) should define as a straight line, the slope to which is related in terms of α. If tcycle is the time of a log cycle for discharge, meaning the time for discharge to change from 1 to 10 m 3 /s, from 10 to 100 m 3 /s, and so forth, then − = 1 and α = 1/(0.4343tcycle). This simplified representation of Equation (5) is frequently used in the calculation of α and will be adopted in the present study. Conversely, the values of Q0 can vary in response to the annual variations of precipitation. In this case, a value of Q0 should be calculated for each hydrologic year, while mean ± standard deviation values are derived therefrom.

Results
The interpolation of precipitation data from 2009 to 2018, obtained from climatologic stations close to the Pandeiros River basin, showed mean rainfall depths ranging from 904.7 to 1,056.3 mm year -1 (Figure 7).  The evapotranspiration ranged from 729.6 to 856.5 mm year −1 (Figure 8). The spatial distribution of this variable was similar to rainfall; the lowest values were found in the southeast, east, and northeast regions, whereas the highest values were found in the northwest region of Pandeiros River basins.
Water 2020, 12, x FOR PEER REVIEW 13 of 24 The evapotranspiration ranged from 729.6 to 856.5 mm year -1 (Figure 8). The spatial distribution of this variable was similar to rainfall; the lowest values were found in the southeast, east, and northeast regions, whereas the highest values were found in the northwest region of Pandeiros River basins. The evaluation of the runoff factor exposes a strong effect of runoff coefficients on the RF, as observed when the land use and cover map ( Figure 5) is compared with the runoff map ( Figure 9). The map of Figure 9 showed a lower groundwater recharge in urban areas, which have a higher runoff. In contrast, higher potential for groundwater recharge takes place in dense vegetation areas, indicating that the presence of vegetation decreases surface runoff. The evaluation of the runoff factor exposes a strong effect of runoff coefficients on the RF, as observed when the land use and cover map ( Figure 5) is compared with the runoff map ( Figure 9). The map of Figure 9 showed a lower groundwater recharge in urban areas, which have a higher runoff. In contrast, higher potential for groundwater recharge takes place in dense vegetation areas, indicating that the presence of vegetation decreases surface runoff. The water percolation factor ranged from 0.001 to 0.7 (dimensionless), expressing the combined variation of hydraulic conductivity values and PUC scores fitted to the 0-1 range ( Figure 10 and Table  8). Areas with melanic gleysols and fluvic neosols presented the lowest water percolation factors, decreasing the groundwater recharge in these areas because of their low hydraulic conductivity.  The water percolation factor ranged from 0.001 to 0.7 (dimensionless), expressing the combined variation of hydraulic conductivity values and PUC scores fitted to the 0-1 range ( Figure 10 and Table 8). Areas with melanic gleysols and fluvic neosols presented the lowest water percolation factors, decreasing the groundwater recharge in these areas because of their low hydraulic conductivity.  The areas with lowest groundwater recharge potential are represented in Figure 11 with a red color and are located in the urban area and in areas with the presence of melanic gleysols and fluvic neosols. The reduced potential is due to the soil sealing process occurring in urban areas, and to low The map of Figure 11 shows the groundwater recharge potential of the Pandeiros River basin ranging from 0 to 122.7 mm year −1 , with a mean value of 93.99 mm year −1 . The areas with higher groundwater recharge potential are located in regions with dense tree vegetation cover, areas with flat or slightly wavy relief, and areas with developed and structured soils, where the porosity and hydraulic conductivity allow water percolation to the water table. These areas are distributed throughout the basin and are found in the three municipalities that encompass the basin.
The areas with lowest groundwater recharge potential are represented in Figure 11 with a red color and are located in the urban area and in areas with the presence of melanic gleysols and fluvic neosols. The reduced potential is due to the soil sealing process occurring in urban areas, and to low hydraulic conductivities that decrease water percolation and consequently recharge in the aforementioned soil types.
The validation of PUC-based recharge estimates is depicted in Table 9. The mean groundwater recharge is 93.99 mm year −1 which is close to the median value obtained with the hydrograph recession method (87.2 mm year −1 ). The difference between the two values is just 6.6 mm year −1 or 7.3%. The hydrograph used to calculate groundwater recharge based on base flows is shown in Figure 12. Hydrographs describe the succession of peaks representing the watershed response to a precipitation event, which are separated by baseflow segments that describe the aquifer response to drainage. Hydrograph recession curves are usually separated in the quick flow stage, which depicts the runoff and water infiltration towards the saturated zone, and the baseflow stage when only the saturated zone discharges [68]. The baseflow discharge is the most representative feature of an aquifer's global response because it is less influenced by the temporal and spatial variations on infiltration [69]. Generally, hydrographs are analyzed together with rainfall data. The peak of a hydrograph rising curve shows the highest values of stream flows, which take place in the months when rainfall values are the highest. In these periods the superficial runoff also reaches its highest values and it decreases during the recession time, marked by the red segments in Figure 12.  The areas with lowest groundwater recharge potential are represented in Figure 11 with a red color and are located in the urban area and in areas with the presence of melanic gleysols and fluvic neosols. The reduced potential is due to the soil sealing process occurring in urban areas, and to low Figure 11. Spatial distribution of groundwater recharge potential in the Pandeiros River basin.
Water 2020, 12, x FOR PEER REVIEW 16 of 24 hydraulic conductivities that decrease water percolation and consequently recharge in the aforementioned soil types. The validation of PUC-based recharge estimates is depicted in Table 9. The mean groundwater recharge is 93.99 mm year -1 which is close to the median value obtained with the hydrograph recession method (87.2 mm year -1 ). The difference between the two values is just 6.6 mm year -1 or 7.3%. The hydrograph used to calculate groundwater recharge based on base flows is shown in Figure  12. Hydrographs describe the succession of peaks representing the watershed response to a precipitation event, which are separated by baseflow segments that describe the aquifer response to drainage. Hydrograph recession curves are usually separated in the quick flow stage, which depicts the runoff and water infiltration towards the saturated zone, and the baseflow stage when only the saturated zone discharges [68]. The baseflow discharge is the most representative feature of an aquifer's global response because it is less influenced by the temporal and spatial variations on infiltration [69]. Generally, hydrographs are analyzed together with rainfall data. The peak of a hydrograph rising curve shows the highest values of stream flows, which take place in the months when rainfall values are the highest. In these periods the superficial runoff also reaches its highest values and it decreases during the recession time, marked by the red segments in Figure 12.  In the assessment of recharge using the method of hydrograph recession analysis, we looked for straight line segments corresponding to base lows. These segments allowed us to draw the corresponding fitting lines (dashed red lines). As would be expected, the lines are all virtually parallel because the line slope is solely dependent on the aquifer characteristics and dimension, which are   In the assessment of recharge using the method of hydrograph recession analysis, we looked for straight line segments corresponding to base lows. These segments allowed us to draw the corresponding fitting lines (dashed red lines). As would be expected, the lines are all virtually parallel because the line slope is solely dependent on the aquifer characteristics and dimension, which are invariant at the timescale of a few years (present case). For all years, the point in the graph where the fitting line intercepted the hydrograph at the upper flows (left edge point) was defined as the Q 0 ,t 0 point, i.e., the point where the recession period began. On average Q 0 = 10.8 ± 1.8 m 3 s −1 . The coefficient of recession was estimated by the simplified version of Equation (5), α = 1/(0.4343t cycle ). In the Pandeiros River basin, the estimated t cycle is 789 days, and therefore α = 0,002918. Using Q 0 and α in Equation (6) results in the recharge value of V =98.8 ± 20.3 mm year −1 . Figure 13 shows the spatialization of groundwater recharge within the municipalities that constitute the studied basin. Although the municipality of Januária encompasses a large area of the Pandeiros River basin, approximately 212,956.8 ha, this municipality presents a higher proportion of melanic gleysols, fluvic neosols, and litolic neosols, which are soils with the lowest water percolation factor (PF) assessed by the model. Moreover, it presents a considerable proportion of exposed soil, sparse Cerrado vegetation, and cultivated soils in the northwest region of the basin, which increases the runoff potential. For these reasons, this municipality presented a mean annual groundwater recharge of 90.30 mm year −1 .
The municipality of Cônego Marinho presented the highest mean annual groundwater recharge (105 mm year −1 ). This municipality occupies a smaller area of the basin (approximately 26,083.86 ha) than the other municipalities, but it is in a region where soil and topography favor groundwater recharge.
Bonito de Minas presented mean annual groundwater recharge of 97.18 mm year −1 . The area of the basin within this municipality is approximately 156,987.95 ha. Despite the considerable presence of fluvic neosols and melanic gleysols in this region of the basin, the proportion is lower than that of Januária. The urban area of Bonito de Minas within the basin is small, representing 0.03% of the total area of the basin, which does not significantly affect the mean annual groundwater recharge.
The land use and cover map ( Figure 5) and the groundwater recharge map (Figure 11) showed that the areas with forest cover presented the best recharge values. This type of land use and cover combined with flat or slightly wavy relief and areas overlaid with red-yellow latosols proved preferable for groundwater recharge.
Pandeiros River basin, approximately 212,956.8 ha, this municipality presents a higher proportion of melanic gleysols, fluvic neosols, and litolic neosols, which are soils with the lowest water percolation factor (PF) assessed by the model. Moreover, it presents a considerable proportion of exposed soil, sparse Cerrado vegetation, and cultivated soils in the northwest region of the basin, which increases the runoff potential. For these reasons, this municipality presented a mean annual groundwater recharge of 90.30 mm year -1 . The municipality of Cônego Marinho presented the highest mean annual groundwater recharge (105 mm year -1 ). This municipality occupies a smaller area of the basin (approximately 26,083.86 ha) than the other municipalities, but it is in a region where soil and topography favor groundwater recharge.
Bonito de Minas presented mean annual groundwater recharge of 97.18 mm year -1 . The area of the basin within this municipality is approximately 156,987.95 ha. Despite the considerable presence of fluvic neosols and melanic gleysols in this region of the basin, the proportion is lower than that of Januária. The urban area of Bonito de Minas within the basin is small, representing 0.03% of the total area of the basin, which does not significantly affect the mean annual groundwater recharge.

Discussion
The rainfall depths (Figure 7) are consistent with Neves (2011), who evaluated a historical series  in the same region and reported a mean rainfall depth of 1050 mm year −1 [45]. However, the Brazilian Institute for Geography and Statistics (2014) considers values below 800 mm year −1 to define the semiarid region of Brazil [70], denoting that the rainfall depths of the study area were relatively high for a semiarid region. Yang et al. (2016) showed that the rainfall depth can be a determinant for evapotranspiration and temperature values [71]. Therefore, practically all water that leaves the system (output) in regions with dry climate and low rainfall, such as the Brazilian semiarid regions, is due to evapotranspiration [72]. According to Loos, Gayler, and Priesack (2007), evapotranspiration is one of the most critical variables, with a high impact on water loss in similar regions [73].
The spatial distribution of runoff factor showed that cover plants are essential to maintaining the water cycle and to protect the soil against the impacts of raindrops. Moreover, the presence of vegetation increases soil porosity and permeability by the action of roots, thereby reducing runoff, and keeping soil moisture in the vicinity of organic colloids [74].
Soils under forest are characterized by expressive plant residue layers (litter fall) and by an A horizon rich in organic matter, which enables a higher soil aggregation, preserving its porosity [75]. Soils under forest usually present significant porosity, mainly macropores due to dead roots and animal holes, which are important to facilitate water infiltration and recharge. Therefore, water infiltration capacity is usually more expressive in areas with forest vegetation [75,76] than in pastureland or cropland, as found in the present work, resulting in a lower surface runoff.
Costa et al. (2019) [10] used a groundwater recharge model and reported a higher surface runoff in urban areas than in vegetated areas. Urban areas have drainage systems, street pavements, and infrastructures that hinder infiltration of rainwater into the soil, decreasing the recharge potential [10].
Comparing the first model employed by Costa et al. (2019) with the present application, it is interesting to see how including the groundwater recharge parameter from the PUC method improved the spatialization of groundwater recharge, mainly because the first model [10] used solely total porosity values for each soil class in the Jequitiba River basin [10]. This parameter is not the best to assess drainage and water percolation, because according to Silva et al. (2013), the micro and macro porosity better reflect the movement of water through the soil profile [77]. Moreover, the PUC method uses a management approach that was not taken into consideration in the PF in the first model. Another important difference from the first model to the present one is the separation until the second category level of the neosols class, which has shown more reliable spatialization and values for the different category levels of this soil class. The differences between the physical characteristics (structure, texture and effective depth) of fluvic neosols, litolic neosols and quartzarenic neosols) make it questionable to give the same values of hydraulic conductivity and porosity to this soil class.
Regarding the validation stage of the work, the higher the slope of the recession curve, the faster the depletion of the water table reserves and the higher the demand of this system to regulate the surface system; whereas the lower the slope, the lower the time of depletion of water resources. The surface and groundwater form a system with mutual contribution, thus, any change in one will affect the other in the short or long term.
The comparison of the methods (spatialization; recession curve analysis) showed a small difference (5.2%) because of their particularities, since each method has inherent and inevitably uncertainty levels [78]. The spatialization method probably has a higher uncertainty in groundwater recharge values because the number of parameters involved with the calculations is much larger [10]. However, the recession curve analysis provides only an average groundwater recharge estimate (mean) for the entire basin, whereas the spatialization method provides one estimate for each point in the basin.
The small difference between the results obtained with the spatially distributed PUC-based and hydrograph recession methods attributes to geology a limited role in the estimation of recharge in the Pandeiros River basin, and that this process is predominantly controlled by the infiltration capacity and profile depth of the soil. The effect of vadose zone thickness on the recharge was recently reported in China [79]. Information on the characteristics of a river basin and its potentialities and limitations are essential to an adequate management of water resources [10,80,81].
The potentialities found for the municipality of Cônego Marinho include the prevalence of red-yellow latosols, which are soils that favor groundwater recharge because of their structural characteristics, such as the occurrence of macropores that increase hydraulic conductivity [43]. Moreover, the region has few cultivated areas and has a predominance of typical Cerrado vegetation, which also favor groundwater recharge, according to the model. Thus, the mean annual groundwater recharge was higher in this municipality than in the other two in the basin.
The considerable overlay of fluvic neosols and melanic gleysols in Bonito de Minas is compensated for by a high proportion of typical Cerrado and dense Cerrado vegetation and by the presence of few cultivated areas, which decreased the runoff potential: Consequently, the mean annual water recharge in this municipality was higher than that in Januária. The presence of native vegetation favors the groundwater recharge, because the forest reduces the surface runoff, favors water percolation, and maintains the soil physical and mechanical stability, assisting in the storage of water and the supply of groundwater [10]. Several studies evaluated the effect of land use and cover and the benefits of forested areas to groundwater recharge [42,[82][83][84][85][86][87][88].
Considering these results, the method used in the present work provides aid in the better management of river basins by considering their potentialities and limitations, not generalizing a mean value for the whole area evaluated. Therefore, preferred areas can be identified for recharge, thus directing public policies and conservationist actions for each area, according to their needs.

Conclusions
The groundwater recharge potential was higher in areas covered with forests, located in plains or slightly wavy relief areas, or overlaid with red-yellow latosol. These are the areas to protect in the watershed management plan if recharge is to be favored or restored. The soil classes and their structural attributes, as well as the land use and cover types were considered as key factors for groundwater recharge. The results showed that areas with higher groundwater recharge potential were concentrated in the municipality of Cônego Marinho, followed by Bonito de Minas, and Januária. Areas with a presence of melanic gleysols and fluvic neosols presented the worst responses in the model.
As made evident from the results, this study should be used as a tool for the management of water resources in the Pandeiros River basin, because the preferred recharge areas could be successfully identified. It is urgent therefore that public policies and conservationist actions are enforced in these areas to improve natural groundwater recharge, and hence increase the accessible water volume to the local population. The adjustment of irrigation methods, adoption of soil preservation practices to improve water infiltration, seasonal storage of surface water in areas of low recharge potential and the preservation of forest vegetation, are examples of feasible actions. Moreover, this work provided subsidies for further studies that seek methods for the spatialization of groundwater recharge potential in river basins with a key role assigned to management practices. Funding: This study was funded by the research project "Sustentabilidade da Bacia do Rio Pandeiros" (Sustainability of Pandeiros River Basin), sponsored by grant APQ-03773-14 from FAPEMIG -Fundação de Amparo à Pesquisa do Estado de Minas Gerais that included a research scholarship for the author Maíse Soares de Moura. The manuscript translation from Portuguese to English was financed by the CAPES -Coordenação de Aperfeiçoamento de Pessoal de Nível Superior. For the author integrated in the CITAB research centre, the research was financed by National Funds of FCT-Portuguese Foundation for Science and Technology, under the project UIDB/04033/2020. For the author integrated in the CQVR, the research was supported by National Funds of FCT-Portuguese Foundation for Science and Technology, under the projects UIDB/00616/2020 and UIDP/00616/2020.

Acknowledgments:
This study was conducted within the work plan of research group "GEISS -Grupo de Estudos Integrado em Solos e Sustentabilidade", in its research line on "Gestão Sustentável de Recursos Hídricos e Segurança Hídrica". We wish to thank the support of the team of Soil and Environment Laboratory of the Federal University of Minas Gerais for the help in the analyses made in the study.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript or in the decision to publish the results.

Nomenclature
The list of mathematical symbols and their measurement units as used in the present article are listed below in alphabetical order: A recession coefficient; C runoff coefficient (dimensionless); E basis of Neperian logarithm (2.71828) ET R mean evapotranspiration (mm year −1 ); Ks FUZZY (dimensionless) soil hydraulic conductivity fitted to the 0 to 1 range by the fuzzy logic algorithm; LS FUZZY slope length and steepness factor; P mean annual rainfall depth (mm year −1 ); PF water percolation factor (dimensionless); PUC Conservative Use Potential; Q T stream flow discharge at time t (m 3 s −1 ); Q 0 stream flow discharge at the beginning of a recession (m 3 s −1 ); RF runoff factor (dimensionless); RPot groundwater recharge potential (m 3 ha −1 year −1 ); T time (day); t' converter of t measurement unit (days into seconds; 86,400)