Evapotranspiration and Precipitation over Pasture and Soybean Areas in the Xingu River Basin, an Expanding Amazonian Agricultural Frontier

The conversion from primary forest to agriculture drives widespread changes that have the potential to modify the hydroclimatology of the Xingu River Basin. Moreover, climate impacts over eastern Amazonia have been strongly related to pasture and soybean expansion. This study carries out a remote-sensing, spatial-temporal approach to analyze interand intra-annual patterns in evapotranspiration (ET) and precipitation (PPT) over pasture and soybean areas in the Xingu River Basin during a 13-year period. We used ET estimates from the Moderate Resolution Imaging Spectroradiometer (MODIS) and PPT estimates from the Tropical Rainfall Measurement Mission (TRMM) satellite. Our results showed that the annual average ET in the pasture was ~20% lower than the annual average in soybean areas. We show that PPT is notably higher in the northern part of the Xingu River Basin than the drier southern part. ET, on the other hand, appears to be strongly linked to land-use and land-cover (LULC) patterns in the Xingu River Basin. Lower annual ET averages occur in southern areas where dominant LULC is savanna, pasture, and soybean, while more intense ET is observed over primary forests (northern portion of the basin). The primary finding of our study is related to the fact that the seasonality patterns of ET can be strongly linked to LULC in the Xingu River Basin. Further studies should focus on the relationship between ET, gross primary productivity, and water-use efficiency in order to better understand the coupling between water and carbon cycling over this expanding Amazonian agricultural frontier.

The portion of the Xingu River Basin located at the Mato Grosso state is surrounded by infrastructure and agricultural land, making this portion more accessible than the rest of the basin [30]. In Figure 1, we also visualized the environmental relevance of the Xingu River Basin and the agricultural-related activities pressure over the natural forest. Despite this increasing pressure, the Xingu River Basin contains one of the world's largest mosaics of protected areas and Indigenous Lands, partially responsible for preserving the natural landscape of this basin.

Land Use and Land Cover (LULC) Mapping
Our main base map was obtained from the TerraClass mappings for 2004 and 2014 [30]. The TerraClass project aimed to map LULC with high-spatial-resolution (30 m) in the Brazilian Amazon during a 10-year period (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014). Since these data have the highest spatial-resolution available for LULC change studies in the Amazon, we have used these two maps for identifying pasture and soybean areas in the Xingu River Basin that did not change during 2001-2013 period. As these mappings were performed only for the Legal Amazon, and part of the Xingu River Basin is located over the Cerrado biome (Figure 1b), we have also used the MapBiomas mappings for 2004 and 2014 [31] in the areas of the Xingu River Basin that were not covered by the TerraClass map.
Both LULC maps were reclassified to eight distinct LULC types: primary tropical forest, secondary succession forest, savanna, pasture, soybean, water bodies, non-observed areas, and other uses.

Precipitation Data
We have used the Tropical Rainfall Measurement Mission (TRMM) satellite monthly product (3B43, version 7) [32] to analyze PPT in the Xingu River Basin during the 2001-2013 period. Recent findings suggest that TRMM provides skillful precipitation estimates in Brazil [33]. Processing steps for the PPT data consisted of: (i) clipping the monthly images to the delimitation of the study area, (ii) reprojecting the monthly data to a geographical coordinate system (lat./long) based on the WGS84 datum, and (iii) resampling the original spatial resolution of the product (approximately 28 km) to 1 km, the same spatial resolution of the ET estimates, using the nearest neighbor resampling method. Finally, maps showing the annual and monthly spatial distribution of PPT were generated, as well as intra-annual and interannual boxplots for this variable considering the entire River Basin, and pasture areas and soybean areas located within the study area.

MODIS-Based Evapotranspiration Estimates
ET in the Xingu River Basin was obtained via the MODIS monthly ET product (MOD16A2 collection 06) [34]. This product is generated at a 1 km spatial resolution based on the Penman-Monteith approach. Monthly images during the 2001-2013 period from MODIS tile h12v09 were acquired, and then rescaled and clipped to the delimitation of the Xingu River Basin. Finally, the images were reprojected to the same coordinate system defined for the PPT data. Maps and boxplots were generated as described in Section 2.2.
The MODIS-based estimates of ET described here were validated for the same region (state of Para, Brazil) in a study developed by de Oliveira et al. [23]. The estimates were evaluated based on flux tower observations from a previously forested area that had been converted to agriculture. The authors found that the MODIS estimates were able to confidently capture the spatial and temporal variability of ET in the eastern Amazon region. The comparison between ET obtained from MODIS products and ground measurements at the LBA sites showed mean relative errors of 13% (ET). Here we note that a recent study by Paca et al. [35] has found reliable results in the estimation of ET for the entire Amazon region using MODIS data.

Land Use and Land Cover
As observed in Figure 2, most of the pasture areas in the Xingu River Basin are located from northeastern to the southeastern portions of the basin, especially in the Pará state. In contrast, the soybean areas concentrate in the south of the basin, bordering with the state of Mato Grosso. This is in agreement with the LULC pattern described by [36,37]. Most of the primary tropical forest areas are located in the Central and Northern portions, consisting of the most preserved areas of this basin. However, these more preserved areas in the Xingu River Basin and the entire Brazilian Amazon have been under increasing deforestation pressure in recent years, especially over protected areas such as Indigenous Lands [38]. There are 27 Indigenous Lands within the Xingu River Basin, with some of them presenting a deforestation boom, such as Apyterewa and Ituna/Itatá, the second and third Indigenous Lands of the Legal Amazon with more deforestation increment in 2019, respectively [14].
Agronomy 2020, 10, x FOR PEER REVIEW 5 of 16 in agreement with the LULC pattern described by [36,37]. Most of the primary tropical forest areas are located in the Central and Northern portions, consisting of the most preserved areas of this basin. However, these more preserved areas in the Xingu River Basin and the entire Brazilian Amazon have been under increasing deforestation pressure in recent years, especially over protected areas such as Indigenous Lands [38]. There are 27 Indigenous Lands within the Xingu River Basin, with some of them presenting a deforestation boom, such as Apyterewa and Ituna/Itatá, the second and third Indigenous Lands of the Legal Amazon with more deforestation increment in 2019, respectively [14]. From Table 1, we can see that 16,866.80 km 2 of primary tropical forest and 997.97 km 2 of savanna were converted during the 2004-2014 period. On the other hand, soybean and pasture areas increased by 15,749.97 km 2 and 3930.08 km 2 , respectively. The loss of primary tropical forests may be explained by the deforestation process in the Xingu River Basin, which led to the increase of soybean and pasture. Most of the savanna was converted into soybean in the southern portion of the basin, while pasture expanded, especially over the eastern and northern portions of the basin. Figure 3 shows the pasture and soybean areas that did not change during the 2001-2013 period, obtained by using the TerraClass and MapBiomas mappings and visual inspection based on TM/Landsat 5 images. The total area for pasture and soybean during the 2001-2013 period corresponds to 38,318.83 km 2 and 7064.46 km 2 , respectively.  From Table 1, we can see that 16,866.80 km 2 of primary tropical forest and 997.97 km 2 of savanna were converted during the 2004-2014 period. On the other hand, soybean and pasture areas increased by 15,749.97 km 2 and 3930.08 km 2 , respectively. The loss of primary tropical forests may be explained by the deforestation process in the Xingu River Basin, which led to the increase of soybean and pasture.
Most of the savanna was converted into soybean in the southern portion of the basin, while pasture expanded, especially over the eastern and northern portions of the basin.   Generally, we can observe that October-April represents the rainy season in the Xingu River Basin, while the May-September period represents the dry one. This pattern is similar to the one observed for the Cerrado biome [40], where part of this basin is located. Monthly average PPT during the dry and rainy seasons were, respectively, 48 mm month −1 and 252 mm month −1 , showing that the monthly average PPT during the rainy season was approximately five times higher than that in the dry season. It is also noteworthy that monthly PPT had a higher variation during the rainy season months (Figure 4b), especially for February, when monthly average PPT ranged between 164 mm (2001) and 432 mm (2004). A smaller variation occurred in July, with a difference between the highest and lowest PPT of 28 mm. Figure 5 shows the spatial distribution and temporal dynamics of PPT in the Xingu River Basin. We can see an increase in the annual average ET from south to north, with lower annual averages concentrated in southern areas where LULC is savanna or soybean. In comparison, higher annual averages are located in the northern of the basin over primary tropical forest areas. The monthly average ET during the rainy season is 109 mm month −1 and during the dry season, it is 89 mm month −1 , indicating that the monthly average ET in the rainy season is approximately two-times higher than the dry season. From the monthly average ET and LULC maps, we observe that ET in savanna, soybean, and pasture areas follow the seasonality of PPT, with values starting to decrease in May, reaching lowest values in August/September and increasing in October. However, we observe a distinct pattern for primary tropical forest areas (north of the basin), where ET is higher at the end of the dry season (August/September).

Spatio-Temporal Dynamics of Precipitation and Evapotranspiration in the Xingu River Basin
Here, we note that ET in primary tropical forest areas of the Xingu River Basin, as in other areas of the Amazon Rainforest that are constituted of primary tropical forest, is not mostly driven only by PPT, but from the interaction among solar radiation, soil water storage, and PPT [24], with solar radiation being the most important driver [41]. The incidence of solar radiation is higher during the dry season because of the lower cloud coverage. Therefore, higher solar radiation in these areas leads to higher ET estimates in primary tropical forest areas [41]. Even during the dry season, the deeper root system of primary tropical forest enables access to water in deepest soil layers [24]. On the other hand, ET on the savanna, pasture, and soybean areas suffer a direct influence of drought because of the more superficial root system, making it more difficult to access water stored in deeper soil layers during the dry season. This is in agreement with [42], who found that ET in forest areas of the state of Mato Grosso was consistently higher than ET in all other LULC types during the 2007 and 2010 drought events, possibly indicating that forests can access deeper soil moisture during droughts.
We highlight that further studies must be developed in order to specifically understand the behavior of ET of primary tropical forests in the study area, focusing on the different vegetation types found in the region such as the non-flooded forests ("terra firme"), inundated forests ("igapó"), and montane forests ("campina"). Figure 6 shows the interannual and intra-annual variability of monthly average PPT and ET in pasture and soybean areas of the Xingu River Basin during the 2001-2013 period. Pasture and soybean areas had annual average PPT values of 1965 mm year −1 and 1830 mm year −1 , respectively, representing a difference of~7%. The higher annual average PPT observed over the pasture in comparison to soybean was expected since the pasture areas are more densely distributed up to the north of the Xingu River Basin (Figure 2), where annual PPT estimates are higher, as opposed to the southern part, which has the lowest values of PPT ( Figure 4). The year 2013 was the wettest in both LULC types (2332 mm year −1 (pasture) and 2089 mm year −1 (soybean)), while the driest ones for pasture and soybean areas were 2002 (1500 mm year −1 ) and 2010 (1539 mm year −1 ), respectively. This indicates that the southern portion of the Xingu River Basin was more affected by the 2010 drought event that occurred in the Amazon [43] than the northern portion of this basin. Monthly average PPT range was similar for pasture (3.8 mm month −1 in July to 360 mm month −1 in January) and soybean areas (4.6 mm month −1 in July to 347 mm month −1 in January). Minimum and maximum (0.3 mm month −1 and 489 mm month −1 ) monthly PPT for pasture areas occurred, respectively, in July 2013 and January 2004, while minimum and maximum (0.30 mm/510.20 mm) monthly PPT for soybean areas occurred in August 2010 and January 2004, respectively.

Temporal Dynamics of Precipitation and Evapotranspiration over Pasture and Soybean Areas in the Xingu River Basin
PPT seasonality follows the same pattern in both LULC types when compared to the entire Xingu River Basin. However, PPT over pasture areas had a wider variation at the beginning of the rainy season. The monthly average PPT during the dry and rainy seasons over pasture and soybean areas was 25 mm month −1 and 263 mm month −1 , and 19 mm month −1 and 248 mm month −1 , respectively. These results indicate reductions in PPT during the dry season of 10 to 13 times for pasture and soybean areas, respectively. Agronomy 2020, 10, x FOR PEER REVIEW 13 of 16

Conclusions
The results in this study demonstrate strong spatial heterogeneity in the water flux, as observed by a satellite sensor within the Xingu River Basin. Seasonal cycles of PPT and ET are shown to vary in timing and magnitude, driven by intra-annual climate variability. We show that PPT is notably higher in the northern part of the Xingu River Basin, as opposed to the drier southern part, with minima from July to September and maxima from January to March. No connection between PPT and LULC was found according to our analysis. ET, on the other hand, appears to be strongly linked to LULC patterns in the Xingu River Basin. Lower annual ET averages occur in southern areas where dominant LULC is savanna, pasture, and soybean, while more intense ET is observed over primary forests (northern portion of the basin). Our results, based on a remote-sensing, spatial-temporal approach, provide evidence that converting primary forests to pasture and soybean substantially modifies the water flux to the atmosphere in the Xingu River Basin.
We highlight the relevance of our results given the fact that meteorological observations are very sparse in the Xingu River Basin, which has notable natural and social relevance in the Amazon region, yet is a relatively unexplored area of research. Further studies should focus on the relationship between ET, gross primary productivity, and water-use efficiency, exploring the differences between agricultural areas and forest types found in the region, such as the non-flooded forests ("terra firme"), inundated forests ("igapó"), and montane forests ("campina"), in order to better understand the coupling between water and carbon cycling over this expanding Amazonian agricultural frontier.  Annual average ET in pasture areas was 594 mm year −1 , that is,~20% lower than the annual average in soybean areas (743 mm year −1 ). Both values are below the annual average for the entire Xingu River Basin (1208 mm year −1 ), where the primary tropical forest is the major LULC (Table 1). This occurred as expected since primary tropical forests have higher ET than pasture and soybean [23]. Monthly average ET was distinct for both LULC types: pasture areas ranged from 1.8 month −1 (September) to 103 mm month −1 (March), and soybean areas ranged from 5.1 mm month −1 (August) to 120 mm month −1 (January). Minimum monthly ET occurred in September/2004 (0.4 mm) for pasture areas and in September/2003 (1.8 mm) for soybean areas, maximum values for pasture and soybean areas were observed in January/2010 (125 mm) and January/2010 (140 mm), respectively. The monthly average ET during the dry season for pasture and soybean areas was 18 mm month −1 and 24 mm month −1 , and for the rainy season, the averages were 72 mm month −1 and 88 mm month −1 . Consequently, the monthly average ET during the rainy season for pasture and soybean areas was four-and three-times higher than the dry season.
As can be seen, ET was higher in soybean areas than pastures, however, PPT was lower in soybean areas in comparison to pastures. These results show how LULC affects ET and that this variable is not the major driver for PPT over agricultural areas in the Xingu River Basin.

Conclusions
The results in this study demonstrate strong spatial heterogeneity in the water flux, as observed by a satellite sensor within the Xingu River Basin. Seasonal cycles of PPT and ET are shown to vary in timing and magnitude, driven by intra-annual climate variability. We show that PPT is notably higher in the northern part of the Xingu River Basin, as opposed to the drier southern part, with minima from July to September and maxima from January to March. No connection between PPT and LULC was found according to our analysis. ET, on the other hand, appears to be strongly linked to LULC patterns in the Xingu River Basin. Lower annual ET averages occur in southern areas where dominant LULC is savanna, pasture, and soybean, while more intense ET is observed over primary forests (northern portion of the basin). Our results, based on a remote-sensing, spatial-temporal approach, provide evidence that converting primary forests to pasture and soybean substantially modifies the water flux to the atmosphere in the Xingu River Basin.
We highlight the relevance of our results given the fact that meteorological observations are very sparse in the Xingu River Basin, which has notable natural and social relevance in the Amazon region, yet is a relatively unexplored area of research. Further studies should focus on the relationship between ET, gross primary productivity, and water-use efficiency, exploring the differences between agricultural areas and forest types found in the region, such as the non-flooded forests ("terra firme"), inundated forests ("igapó"), and montane forests ("campina"), in order to better understand the coupling between water and carbon cycling over this expanding Amazonian agricultural frontier.