Land Cover Changes Implications in Energy Flow and Water Cycle in São Francisco Basin, Brazil, over the Past Seven Decades

This research aimed to quantify and qualify alterations in land cover and verify the implications of these modications for variables related to energy ows and water cycle in São Francisco basin (SFB), located entirely in Brazilian territory, in the second half of the 20 th century and beginning of the 21 st . For this, statistical analyzes (descriptive, trends, seasonal and correlations) were used to quantify changes in the variables of land cover and energy/water ows, in addition to relating them. As a result, it was found that the SFB lost 65,680 km² of native vegetation (10.4% of basin area) to crops and pastures, reducing water inltration (-52%) while the rains remained stable (-2%). Water loss increased through evapotranspiration (+5%) and surface runoff (+225%). Such changes in the water cycle have entailed an 11% reduction in São Francisco river long term ow rate (Q 95 ), comparing pre and post-1990s period. In SFB, the activities that required water, such as farming activities, are those that promote hydric loss.


Introduction
Investigating Earth's energy and water cycle is essential to understand global climate dynamics and how it impacts and is in uenced by human actions (Seyoum and Milewski, 2017;Umair et al., 2019). The terrestrial climate is susceptible to radiation ows, both those of short-wave originating from solar activities and those of long-wave from solar rays' irradiation on the planet's surface (Mokhtari et al., 2018; the planet (Yuan et al., 2019). There are several other examples of studies, which can be found on LDAS website.
In this context, the objective of this research is to identify and measure changes in the energy ow and water cycle in São Francisco Basin (SFB) in the second half of the 20th century and the rst two decades of the 21st, and verifying the implications of land cover modi cations, caused by human activities in the last four decades, in changes caused in these cycles' dynamics. Although studies have already been carried out for the basin in this sense, there are still certain gaps concerning changes in the inlet and outlet of water and energy, alterations that have been in uenced by human activities.
The most recent SFB studies dealing with the water cycle focus on issues related to water availability and demand. Sun et al. (2016) identi ed a loss rate of 3.3 km³ of water per year over 13 years of assessment (between 2002 and 2015), caused mainly by the drought that started in 2012 and ended in 2017, which resulted in a loss rate of 27.63 km³ between 2012 and 2015. In this study, data on precipitation and stored water present on the earth's surface were used, and the in uence of the El Niño phenomenon on drought period occurrence.
In another research, by Koch et al. (2015), water availability and demand were assessed under two scenarios: i) a regionalized world with slow economic development, high population growth and little awareness of environmental problems; ii) a globalized world with low population growth, high growth in Gross National Product (GNP) and environmental sustainability, also adding climate change scenarios to both. As a result, they found that between 2021 and 2050 the basin will be wetter, with more intense rainy and dry periods, and increased water availability for irrigation and a drop in electricity generation.
In addition to these, some other studies highlight that hydroelectric energy production may cease in this rst half of the 21st century in drought periods (de Jong et al., 2018). Also the prices of agricultural products generated in SFB will be affected, which will depend on climate change, the production location, variety of products and production technology (Torres et al., 2012). It is as well estimated that economic values of water for use in irrigation tend to increase in the coming decades (Alcoforado de Moraes et al., 2018).
Although they are important researches, none of them deal with implications that changes in land cover, having as vectors the anthropic activities, cause in the energy ows and water cycle in SFB, a theme that will be the focus of this research.

Materials And Methods
The stages of this research were organized as follows ( Fig. 1): i) study area determination; ii) de nition of variables to be studied; iii) statistical analysis of the variables de ned in study area over the available historical series; iv) spatiotemporal analysis of the studied variables; v) evaluation of impacts on the water availability in São Francisco River caused by changes in the studied variables.

Study Area
The São Francisco Basin (SFB) (Fig. 2) is located in the Northeast and North of the Southeast of Brazil, with an area of 630,000 km² and with its main river running 3,200 km from the source, in the State of Minas Gerais, at its mouth, on the border of the States of Alagoas and Sergipe, owing into South Atlantic Ocean (Bezerra et al., 2019;de Jong et al., 2018). It is between the latitudes of 21º and 7º South, with climatic characteristics, according to the revised classi cation of Thornthwaite (Feddema, 2005), ranging from humid and wet sub-humid types between latitudes 21º and 10º South (29% of basin area), until dry sub-humid, semi-arid and arid types between latitudes 17º and 7º South (71%).
In humid and wet sub-humid areas, south and southwest of basin, temperatures reach an average of 20ºC (Torres et al., 2012), with potential evapotranspiration ranging from 1,241 to 4,344 mm/year, with an average of 2,300 mm/year (Beaudoing and Rodell, 2019), and rains precipitate an average of 1,500 to 2,000 mm/year (Koch et al., 2015;Sun et al., 2016). In dry sub-humid, semi-arid and arid areas, north and northeast of basin, temperatures reach an average of 26.5ºC (Torres et al., 2012), with potential evapotranspiration ranging from 1,533 to 5,110 mm/year, with an average of 2,884 mm/year (Beaudoing and Rodell, 2019), and rainfall precipitates an average of 350 mm/year (Koch et al., 2015;Sun et al., 2016).
Precipitation in humid and wet sub-humid areas is mainly associated with the South Atlantic Convergence Zone (SACZ) that operates in the southeastern region of Brazil and southwestern Bahia during the summer (between October and March) (Torres et al., 2012). In the dry sub-humid, semi-arid and arid areas, rains are concentrated during the months of March and April, caused mainly by action of the Intertropical Convergence Zone (ITCZ), which acts more intensely in this period due to weakening of the inter-hemispheric south gradient of sea surface temperature, which allows ITCZ to reach regions further south. In the rest of the year, there is a strong presence of a high-level cyclone over the Amazon, which inhibits rains formations in the region and favors dry weather conditions, which is often enhanced by strong El Niño events.
The humid and wet sub-humid areas cover almost all the sub-regions of Upper São Francisco and western part of the Middle. The main river has its contribution mainly from these areas, which were responsible for 93%, between 1951 and 1999, of total permanence ow curves (Q 95 ). That is, approximately 43% of basin was responsible for maintaining ow rate that owed through the riverbed 95% of the time. The other 57%, corresponding to the sub-regions of Low, Lower-Middle, and eastern part of the Middle São Francisco, which are mostly covered by dry sub-humid, semi-arid and arid areas, contributed with remaining 7% of Q 95 (Pereira et al., 2007;Pruski et al., 2004). This showing the importance of the areas closest to its source for maintenance of main river.
São Francisco has an average ow rate of 2,850 m³/s, ranging from 1,077 to 5,290 m³/s (Bezerra et al., 2019), with a permanence ow curve (Q 95 ) of 800.4 m³/s (Pereira et al., 2007;Pruski et al., 2004). It comes from 70% of the surface waters of entire Northeast region of Brazil (Torres et al., 2012), which has a population of over 53 million people, approximately a quarter of the Brazilian population. Besides, hydroelectric system of this river normally meets 70% of the demand for electricity in Northeast region (de Jong et al., 2018).
Added to these hydrological conditions, the basin has vegetations of Cerrado types (Savannah Formation) with a great diversity of tree vegetations (Forest Formation) in Upper and western sub-region of Middle São Francisco, as well as the vegetation of Caatinga biome, which is widely spaced and smaller, associated with drier and hotter climate in Low, Lower-Middle and east of Middle São Francisco (Creech et al., 2015).
Land use is predominantly agricultural, with crops (46%) and pastures (41%) dominating the land cover in basin. Aside from, there are remnants of large and medium-sized tree vegetation, urban areas, mining, among others. The main agricultural production is soy, with other major crops such as corn, wheat and cotton, in addition to important fruit centers (Juazeiro and Petrolina) (Correia et al., 2001) and family farming throughout the basin. Vegetations, crops and pastures are on soils of latosol (41% of the basin), podzols (11%), sandy soils (10%) and cambisols (7%) (Alcoforado de Moraes et al., 2018;Creech et al., 2015;Torres et al., 2012).
Covering approximately 7.5% of the Brazilian territory, SFB has a population of over 17 million people (8% of Brazilian population), 21% of which are considered poor by the country's standards (Sun et al., 2016). Due to population growth and the demand for food, water, jobs and services, the municipalities in basin have had satisfactory socio-economic development in recent decades, but at the cost of natural resources, mainly native vegetation, soil and water, relative with deforestation problems, deserti cation, water scarcity, pollution and water bodies silting up (Creech et al., 2015;de Jong et al., 2018;Koch et al., 2015).
From the above, it is clear that SFB has signi cant physical-natural and socioeconomic diversity, which are dynamic in space and time. Due to these factors, the basin was de ned as a study area precisely because of its dynamics and diversity related to environmental and human aspects, which interfere with each other. The choice focused on dynamics pertinent to energy and water cycles, which are affected by the basin's socio-economic activities, but also affect them. The option for the area is also linked to the fact that the basin has one of the main rivers in Brazil, entirely within the national territory, which passes through the driest region and one of the poorest in the country. The river has great national importance due to its strong potential for economic use, mainly linked to agricultural activities, irrigation, public supply, navigation and generation of electric energy, but has been under strong pressure exactly by exploration of these same activities.

Variables De nition
The variables de ned in this research were those related to the energy ow and water cycle in SFB, as well as those that can interfere in these cycles, referring to land use and land cover, in addition to the minimum, average and maximum ows rate of São Francisco river (  The Global Land Data Assimilation System (GLDAS) is a program from the National Aeronautics and Space Administration (NASA) in which observational data from orbital images and information collected in the eld are used to model the dynamics of terrestrial phenomena linked to energy ow and the water cycle of the planet Earth. Produces results almost in real-time, with resolutions that vary between 2.5º and 1 km, with historical series between 1948 until today with periodic update (Beaudoing and Rodell, 2019;Rodell et al., 2004).
The data used in GLDAS contains relief information, soil types, vegetation and atmospheric variables such as rain and radiation ows. Data assimilation techniques are employed to incorporate hydrological products based on satellite sensor imaging that include snow cover and water equivalent, soil moisture, surface temperature and leaf area indexes (Beaudoing and Rodell, 2019;Rodell et al., 2004).
The Annual Mapping of Land Use and Land Cover in Brazil (MapBiomas) (Souza et al., 2020) is a collaborative network with specialists in Brazilian biomes, land uses, remote sensing, geographic information systems and computer science, which uses cloud processing and automatic classi ers (Random Forest) developed and operated from Google Earth Engine platform. Since its foundation (2015) it has been generating an annual historical series of land use and land cover maps for the whole of Brazil (1985 to 2018 -Collection 4.1), with a spatial resolution of 30 meters using the series of images from Landsat sensors.
The program Making Earth System Data Records for Use in Research Environments (MEaSUREs), from NASA's Land Processes Distributed Active Archive Center (LP DACC), is dedicated to the advancement of remote sensing and the scienti c use of measurements from satellite sensors to expand the understanding of terrestrial system, with production and recording of consistent, high quality and longterm data. Among several databases, those used in this research were the Vegetation Continuous Fields (VCF) (Hansen and Song, 2018) and the Vegetation Index and Phenology (VIP) (Didan and Barreto, 2016). The rst is a collection that provides global vegetation information from the Advanced Very High-Resolution Radiometer (AVHRR) long-term records between 1982 and 2014, annually, with a spatial resolution of 5,600 m containing information on the percentage of tree vegetation cover, non-tree vegetation cover and bare ground. The second is another collection containing a monthly historical series from 1981 to 2019, of vegetation and landscape phenology indexes, based on data from the Moderate-Resolution Imaging Spectroradiometer (MODIS), AVHRR and Satellite Pour l'Observation de la Terre (SPOT), also with a spatial resolution of 5,600 m.
The Hidroweb Portal (http://www.snirh.gov.br/hidroweb/) is a Brazilian tool that is part of the National Water Resources Information System (SNIRH, acronym in Portuguese) and offers access to database that contains all information collected by the National Hydrometeorological Network (RHN, acronym in Portuguese), which gathers data on ows rate, river height, rainfall, climatology, sediment and water quality. The data used come from conventional uviometric stations with codes 45298000 (14.3º S and 43.76º W) and 49705000 (10.21 S and 38.82 W), with historical series of daily and monthly frequencies between 1928-2019 and 1959-2019, respectively. The use of these two stations is better detailed in item 3.1, as the choice of both was de ned based on some results of this research.

Statistical Analysis
The statistical procedures were broken down into three stages: i) rst performed on variables related to energy ow and water cycle; ii) later on variables related to land use and land cover; iii) and, nally, in the variables related to São Francisco river ow rate and its correlations with the analysis of the previous data.

Statistical analysis on variables related to energy and water cycles
First, the Mann-Kendall test (MK tau) was performed on variables related to energy ow and water cycle, from GLDAS database, monthly, between January 1948 and December 2019.
This rst analysis aimed to verify the spatiotemporal behavior of the variables and to search for different trend patterns throughout SFB in the mappings made by the test, pixel by pixel, using the Series Trend tool of the Earth Trend Modeler module available by Idrisi Selva 17.0 software (Eastman, 2016). With the identi cation of patterns, it was decided to divide the study area into two parts (P1 and P2). This decision is explained in results.
MK tau is a non-parametric trend indicator that measures the degree to which a trend is increasing or decreasing consistently. It was rst proposed by Mann (1945) and later studied by Kendall (1975), then enhanced by Hirsch et al. (1982) and Hirsch & Slack (1984), which made it possible to take seasonality into account. It has a range of -1 to + 1. A value of + 1 indicates a trend that continually increases and never decreases. The opposite is true when it has a value of -1. The zero value indicates that there is no consistent trend. All combinations of pair values over time are evaluated at each pixel and a count is made of the number that is increasing or decreasing over time. It is simply the relative frequency of increases minus the relative frequency of decreases (Eastman, 2016).
With P1 and P2 discriminated, descriptive statistics were made to verify variations of the average values of each variable in both areas, checking if in general there was an increase, stability or decrease in each one of them. Stationarity (KPSS) and homogeneity (Pettitt and Buishand) tests were also performed.
The stationarity test used was the KPSS (Kwiatkowski et al., 1992), which allows to check whether a series is stationary or not. The test result varies from zero to ∞ (Eta Observed Value), in which the higher the critical value (Eta Critical Value), greater the tendency for the series not to be stationary, and the opposite being also true, that is, as closer to zero and below the critical value, greater the tendency to stationarity.
The homogeneity tests used were those by Buishand (1982) and Pettitt (1979), that check if a series is homogeneous over time, or if there is a moment when a change or break in that homogeneity occurs.
Such tests were selected based on different sensitivities they present for identi cation at the change time and because they are widely used to test homogeneity in environmental data series (Bickici Arikan and Kahya, 2019; Rougé et al., 2013;Serinaldi et al., 2018;Serinaldi and Kilsby, 2015;Yozgatligil and Yazici, 2016).

Statistical analysis on variables related to LULC and MEaSUREs
To measure changes in land use and land cover in SFB, the Change Analysis tool of Land Change Modeler module, from Idrisi Selva 17.0 software, was used (Eastman, 2016). In this tool, it was possible to verify the main changes between the years 1985 and 2018.
In addition to measuring changes in land use and land cover using MapBiomas data, the MEaSUREs database was also used, which contains data on the percentage of tree cover, non-tree cover and bare ground (VCF) between 1982 and 2014, and NDVI (VIP) between 1981 and 2019, respectively (Didan and Barreto, 2016;Hansen and Song, 2018).
For MapBiomas land use and land cover data, from the measurements generated in Land Change Modeler module, all classes related to human actions (Farming, Mining, Planted Forest and Urban Areas) were grouped into a single class called Anthropogenic Disorders. In this, Mann-Kendall and homogeneity tests (Pettitt and Buishand) were performed.
In the MEaSUREs data, Mann-Kendall test was used to identify differences spatiotemporal trends in the SFB with Series Trend tool of Earth Trend Modeler module from Idrisi Selva 17.0 software aid (Eastman, 2016).

Statistical analysis on variables related to São Francisco river ow rate
The data series on the minimum, average and maximum ows rate of São Francisco river was analysed in two river stations, one located in P1 (code 45298000) and another in P2 (code 49705000).
Mann-Kendall, stationarity and homogeneity tests were also applied. After the application of this tests and results obtained, these were compared to the results of tests related to the variables of previous topics. This was performed to verify possible implications of changes in land cover, in energy ows and, consequently, in water cycle basin, and the impacts of these processes on ow rate dynamics of São Francisco river at P1 and P2.
To reinforce possible in uences of changes in land cover in water cycle basin, correlation analyzes between NDVI and variations related to water cycle were performed, with the aid of Correlate module of Idrisi Selva 17.0 software (Eastman, 2016). The main application of the module is to identify areas that correlate with a speci c temporal pattern of interest, calculating the correlation coe cient R between one or more predictors (independent) with a time series of images (dependent) for each pixel. In addition, such analyzes were also performed between historical series of average ows rate in P1 and P2 with rainfall, surface and subsurface runoff data, using Pearson's correlation (Barber et al., 2020).

Results And Discussion
This topic is divided into three subtopics, which are: 1) results related to spatiotemporal analysis of energy ow and water cycle variables; 2) results related to spatiotemporal analyzes of land use and land cover and; 3) results and discussion referring to consequences on the dynamics of São Francisco river ow rate resulting from changes in the variables evaluated in the two previous subtopics.

Spatiotemporal analysis of energy ow and water cycle variables
Mann-Kendall tests carried out for energy ow variables showed that there were different intensities in trends, positive and negative, of radiation ows (short and long waves) and heat (sensitive and latent), in speci c areas in SFB between January 1948 and December 2019 (Fig. 3).
Due to this distinctions, the basin was segmented into two parts, P1 and P2 (Fig. 3). Such segmentation was based on trends of radiation ows (short and long waves) (Tables 2 and 3), which will in uence the dynamics of other variables, such as heat ows (sensitive and latent), evapotranspiration and presence of moisture in the canopies and plants root zones. These results also guided the choice for used uviometric stations locations, aiming to relate the changes in trends of the variables evaluated in GLDAS database with ow rate dynamics. The de nition of the uviometric stations followed two conditions: i) close to the place further downstream from P1 and P2, seeking to cover the entire contribution basin and; ii) with su cient historical series to satisfactorily cover the period of GLDAS and MEaSUREs data.   This fact contributes to positive trends of potential evapotranspiration throughout the basin, notably in P2, in uenced by signi cant positive ow of short-wave radiation and speci c heat. Even though the potential evapotranspiration has increased considerably, the real evapotranspiration, although it has grown, has not followed it (Table 5). In P1, the average potential evapotranspiration increased from 2,043 to 2,510 mm/year, while average real evapotranspiration increased from 820 to 883 mm/year. In P2, the average potential evapotranspiration increased from 2,359 to 3,431 mm/year, while average real evapotranspiration increased from 631 to 662 mm/year. This is shown in the trend analysis (MK) results, where for potential evapotranspiration, they were 0.175 and 0.309, P1 and P2 respectively. As for the real evapotranspiration, they were 0.04 and zero, P1 and P2 respectively.
While there was an increase in potential and real evapotranspiration, pluviometric indexes remained stable (Table 6).
In P1 the average rainfall decreased from 1.324 to 1.261 mm/year, and in P2 there was no variation in the average, remaining at 694 mm/year. The Mann-Kendall analysis con rms such variations, in P1 with a result equal to zero and in P2 equal to 0.005, that is, a tendency to stability.
If on average the rainfall rates were stable, which led to an increase in real evapotranspiration in P1 and P2, even if not following the potential evapotranspiration, it was the increase in vegetation transpiration (Table 7), which increased in P1 from 265 to 441 mm/year, and in P2 from 290 to 366 mm/year. The increase in evapotranspiration was accompanied by a reduction in the presence of water accumulated in plants' canopies and a decrease in soil moisture in the root zone (Table 8). In P1, there was a water dip in plants' canopies from 0.19 to 0.12 mm, and a decrease in soil moisture in root zone from 244 to 228 mm. In P2, there was a water dip in plants' canopies from 0.11 to 0.07 mm, and a decrease in soil moisture in root zone from 210 to 179 mm. Additionally, there was an increase in surface runoff and a decrease in subsurface runoff (Table 9). In P1, there was an increase in surface runoff from 16 to 80 mm/year and a decrease in subsurface runoff from 445.6 to 283.2 mm/year. In P2, there was an increase in surface runoff from 0.8 to 38.4 mm/year, and a drop in subsurface runoff from 126.4 to 38.4 mm/year.  Figure 4 shows in general how the trends of each variable from energy ow and water cycle behaved over the period evaluated. The results in Mann-Kendall test show that there were increases in heat and radiation ows across SFB, although on average the latent heat ow is stable at P2, in addition to stability in rainfall indexes. The KPSS test con rms such trends, showing that rainfall has a stationary behavior, both in P1 and in P2, over time, whereas for radiation and heat ows the critical value has been reaching, mainly for speci c heat and short-wave radiation, showing a tendency towards non-stationarity throughout the series.
As a consequence of increase in radiation and heat ows, trends in potential and real evapotranspiration are positive, except for real evapotranspiration in P2, tending to stationarity.
Although actual evapotranspiration did not keep up with the increase in potential, it was positive due to the plant transpiration growth, as the water evaporation directly from plants' canopies reduced, and evaporation directly from bare soils was stable. Even though in P2 the KPSS test points out that there is no stationarity in the series, with a possible reduction.
The trends for the presence of moisture in plants' canopies and their root zones are decreasing, rati ed by non-stationarity in KPSS test. The reduction in humidity in these places is probably due to the increase in evapotranspiration, but, mainly, to the rise in surface runoff, which has a positive tendency, and to subsurface runoff reduction, which has a downward trend. This shows less water in ltration in the soil, providing less moisture to plants' roots, and less water containment by vegetation, re ected by less water maintenance in their canopies and increase in water loss due to surface runoff.
All variables that showed a trend, positive or negative, with signi cant p-values, obtained results in Pettitt and Buishand tests that showed a break in homogeneity mainly in the 1990s, varying between 1995 and 2000 in P1, and between 1986 and 2000 in P2. The variables latent heat, precipitation, evapotranspiration, direct water evaporation from bare soil and subsurface runoff, whether in P1 and P2 or one of them, were those that obtained meanings that were more distant from zero, being exactly those that showed a break in homogeneity in different years to the rest of variables. This means that there was no break in their series, as the results in Mann-Kendall and KPSS tests showed that the variables with most distant values of signi cance were exactly those with no de ned trend, that is, tending to stationarity.

Spatiotemporal analyzes of land use and land cover
Between 1985 and 2018 at SFB there was a reduction of 68,631 km² of native vegetation formations, equivalent to 10.8% of the basin total area. From this reduction, 95.7% was caused by agricultural activities (perennial and semi-perennial crops, planted forests and pastures), and the remainder, 2,951 km² (4.3%), due to the advance of urban areas, mining and other uses (Fig. 5). The most degraded vegetation formations were savannas and grasslands, medium and small sizes vegetations from Cerrado and Caatinga biomes and transition between them. The third most affected were forests, of medium and large size, originating mainly in Atlantic Forest biome, but also present in transition areas between these, Cerrado and Caatinga.
This reduction in native vegetation formations and their substitution by areas for agricultural use caused a fall in tree cover (Table 10), both in P1 and in P2. As for the non-tree cover, in P1 and P2 there was an increase, re ecting the pasture and crop expansion in detriment of forest and savanna areas, medium to large vegetation. The bare soil, at the same time, was reduced in P1, even with native vegetation degradation, due to territorial growth of agricultural cover. Although there were dynamics similar to P1 in P2, that is, the replacement of native vegetation with pastures and crops, there was a signi cant increase in bare soil, in contrast to what happened in P1. Between 1982 and 2014, in P1, tree cover reduced, on average, from 18.5-15%, and non-tree cover increased from 77.1-81.3%. Bare soil fell from 5.8-4%. In P2, tree cover dropped, on average, from 22-13.3%, and non-tree cover increased from 72.7-76.8%. Bare soil rose from 5.3-9.9%.
Concerning the NDVI (Table 11), even with drop in tree cover in P1, there was mostly an increase, which on average rose 0.06, from 0.50 to 0.56. In P2, in general, it remained stable, with a drop of 0.01, from 0.52 to 0.51, approximately half of the area falling and the remainder rising. The tests by Pettitt and Buishand (Fig. 5), show that the break in homogeneity in historical series of land use and land cover occurred in the late 1990s for P1 and early 2000s for P2, very close to the results achieved for variables from the previous subtopic.

Impacts and changes in São Francisco river' ow rate dynamics
Although the drop in rainfall index in P1 was small, 5% between 1948 and 2019, the drought events of the last decade (2010-2019) affected the permanence ow curve of São Francisco river (Q 95 and Q 50 ) ( Fig. 6). It is also noteworthy that these ows were in uenced by the temporal behavior of rainfall, increasing when rainfall increases (between 1950 and 1989) and decreasing when rainfall drops (between 1990 and 2019).
Another relevant factor is the change in land cover caused by the expansion of activities and land uses, mainly linked to agriculture and livestock. Even though in P1 there was a fall in bare soils, this was accompanied by a reduction in tree vegetation cover and an increase in the non-tree vegetation cover, due exactly to increase in pastures and crops areas. This fact has implications for other variables related to energy ow and water cycle, such as the greater ow of radiation and, consequently, of heat, causing a greater evapotranspiration process, in addition to reduction in tree cover contribute to increase in surface runoff and reduction of water soil in ltration, reducing subsurface runoff and soil moisture in plants root zone.
Even with the reduction in rainfall during the 2000s and 2010s, surface runoff was greater than in previous decades, when there were similar or higher rainfall levels. This has an impact on the long-term ow rate of São Francisco river, given the precipitated water that drains super cially is reaching the watercourses faster due to reduction of friction caused by the replacement of the land cover. Thus, water soil in ltration reduced and caused less subsurface runoff, impairing the recharge of groundwater that supplies tributaries and the main river itself. This supply is even more important in drought periods.
In P2 (Fig. 7), the situation is similar to P1, although there was a greater drop in tree cover and a greater increase in bare soil. The exception is the behavior of rainfall over the decades, which is not similar to that of ows rate (Fig. 7). This occurs because in P2 the ow rate also follows the behavior of P1 rains, which are responsible for 65% of the water entering basin on average. It is also worth mentioning another factor that in uenced the ows rate, which is the construction of the Sobradinho dam, in the state of Bahia. Inaugurated in 1982, from then on the dam contributed to regularization of the São Francisco ow rate. Between 1959 and 1980, in rainy season, maximum and minimum ows rate reached values close to 4,600 and 2,800 m³/s, respectively, and 1,200 and 1,100 m³/s, respectively, in drought period. From the 1980s to 2000s, even with a certain increase in rainfall, maximum and minimum ows rate did not reach 3,600 and 2,400 m³/s, respectively, in rainy season, and exceeded 1,900 and 1,500 m³/s, respectively, in drought period, showing a control of increase in ows rate when the rains occur with greater frequency and intensity, and release when the drought occurs (Fig. 8).
The ow rate regularization shown in the uviometric station located in P2 does not occur in the one located in P1, which reveals that there was a clear drop in the maximum and minimum ows rate over the three periods evaluated, except for some similar values that occur between the periods, notably in dry season. Comparing P2 to P1, the amplitude of maximum and minimum ows rate between rainy and drought periods in 2001-2019, shows how the dams can help to regulate water ow rate in watercourses and keep their availability even higher in drought periods.
In P2, despite the reduction in rainfall between 2001 and 2019, there was an average drop of 28% and 8% in maximum and minimum ows rate, respectively. In P1, the fall was 82% and 63%, respectively. This fact shows that without the construction of the dam, the situation in P2, regarding water availability in São Francisco river, could have been more serious in drought periods.
In 2001-2019 period, in addition to a small reduction in rainfall in P1 and a stable situation in P2, there was also an increase in average evapotranspiration in rainy season and a reduction in drought in both areas (Table 12). There was an increase in average surface runoff practically throughout the year, notably in greatest rainfall period, and average subsurface runoff decreased, mainly in rainy season in P1. The presence of moisture in plants' root zone also dropped in the period 2001-2019, in both areas. The average reduction of 5% of rainfall in P1 is one of the factors related to the reduction of water present in basin, but the increase in evapotranspiration and surface runoff, caused by changes in land cover, had a greater impact on water maintenance. The correlation (R) between the monthly data from January 1981 to December 2019 of NDVI with those of rain, evapotranspiration, root zone soil moisture of the plants, surface and subsurface runoff, shows, to a certain extent, the in uence of vegetation maintaining humidity in basin. Figure 9 shows how the NDVI has a higher correlation with rain, evapotranspiration and soil moisture in root zone. In other words, it shows how the seasonality of rainfall in uences the seasonality of vegetation, with vegetation increasing its vigor in the rainy seasons and decreasing in the dry seasons, expect for some areas, probably in uenced by local factors, such as relief, irrigation, among others.
The same occurs with soil moisture in root zone and with evapotranspiration, as in the rainy seasons it tends to have a greater water accumulation in the soil, increasing its humidity, and consequently a greater plant development, which also increases evapotranspiration processes. In the dry seasons, the lower occurrence of rain decreases inlet of water into basin, and with increase in energy ow (radiation and heat), it tends to increase potential evapotranspiration, removing more water from the system by this process.
Although NDVI shows the vigor and health status of vegetation with some success, there is some di culty in de ning its size, as there is no direct relationship between this and the values of this index, depending a lot on the type of vegetation evaluated, in which biome and climatic caracteristics is located.
For this reason, this index obtained lower correlations with surface and subsurface runoff, which will vary according to the occurrence of precipitation. It also depends on the relief and obstacles that favor or not water in ltration in soil and ow speed on the surface, with plant sizes as important factors in this case. Land covers (bare soil, arboreal and non-arboreal vegetation) were not used in this correlation because their information is of annual frequency, and it is not possible to assess the seasonality of the process, which is important in assessing the correlations, mainly due to the differences in the rainy and dry seasons.
Correlating average monthly ow rate of the uviometric stations located in P1 and P2, with variables of rainfall, surface and subsurface runoff (Fig. 10), it is clear their importance in the ow rate dynamics over time. However, notably, as subsurface runoff is vital for maintaining ow rate, especially in droughts. This shows how the recharge of groundwater is essential during rainy periods, and this process has been hampered by changes in land cover (Fig. 11).
As seen earlier, Buishand and Pettitt tests show that the break in homogeneity for energy ow, water cycle and land cover variables, occurred in the late 1990s and early 2000s, similar, too, the result for these tests in the cases of ows rate in P1 and P2, which were between 2000 and 2002. Thus, concerning the pre and post-1990s periods, for the entire SFB, the average reduction in rainfall volume was 2%, from 958.9 to 940.5 mm per year, down 18.4 mm. While rainfall is stable, average evapotranspiration increased by 5% (+ 38.1 mm), surface runoff increased by 225% (+ 36 mm) and water in ltration in the soil decreased by 52% (-92.5 mm) (Fig. 11). Study area location. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.
Page 29/37 Trends in variables referring to energy ow and water cycle in SFB between January 1948 and December 2019. The red dots are the uviometric stations locations with codes 45298000 (P1) and 49705000 (P2). Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.  Correlations between NDVI and water cycle variables in SFB. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 10
Correlations between average monthly ow rate with rainfall, surface and subsurface runoff data in SFB.
Note: the construction of Sobradinho dam in uenced the correlation values at the uviometric station located in P2, due to ow rate regularization in São Francisco river, which signi cantly changed its behavior compared to other variables natural dynamics.