3 Results and discussion
3.1 Time series analysis of evapotranspiration
Figure 3 shows the daily time series of potential and actual evapotranspiration, as well as precipitation. It can be seen that the annual amplitude of ET0 in the two areas is very similar: they range from 2 to 10 mm day-1 in the Caatinga forest and from 0 to 8 mm day-1 in the Pinares forest. In the Caatinga forest, however, value dispersion is higher, since the area does not have a well-defined distinction between four seasons. We also observe that ETa is much closer to ET0 in the Pinares forest, especially in the winter, when the atmospheric water demand is lower. In summer (July-September), the soil is dryer due to the combination of high potential evapotranspiration and low precipitation (Calama et al., 2019), which diminishes soil moisture and, consequently, actual evapotranspiration.
During the dry season in the Caatinga forest, actual evapotranspiration is generally much lower (low ETa) than the potential one (Figure 3a). In that period, the potential evapotranspiration is high due to a high air temperature and low air moisture, which raise the vegetative water demand. Simultaneously, water availability for the plants is limited due to a continuous absence of precipitation for several months (Medeiros & de Araújo, 2014), so that soil moisture is extremely low (in the Caatinga forest, moisture is below wilting point during more than six months annually: Costa, Lopes, Pinheiro, Araújo, & Gomes Filho, 2013). The combination of high atmospheric-water demand and low soil-water availability causes the plants to actually evapotranspirate considerably less than the potential value. During the rainy season, in contrast, soil moisture increases and potential evapotranspiration decreases, leading to high ETa values compared to potential ET (Figure 3b).
Due to the presence of clouds it was not possible to obtain images during the rainiest months of the year (February, March and April in the Caatinga forest and December in Pinares). According to Jaafar and Ahmad (2020), some limitation exists when cloudy days are frequent: estimated ETa may be higher than the actual values during winter and early spring, considering that the cloudless images are not representative of that period, and in images of cloudy days, evapotranspiration is below the period average. This is true for the Brazilian semiarid during the rainy season, when cloudiness prevails, and it explains why the curve generated in Figure 4a is incomplete. Despite data gaps, however, one can observe the annual behaviour of ETfrac in both areas and its value can be used for prospective estimates of ETa. According to Teixeira et al. (2009), the value of ETfrac during a Caatinga dry season is around 0.20. Yet, for the present study the maximum ETfrac value could not be identified, considering that no useful images were available during the months with the highest precipitation rates. In accordance with the authors op cit. , the ETfrac may surpass 0.90 during wet season in a Caatinga vegetation. This value was admitted for the months of February, March and April for the complete construction of Figure 4c.
With regards to the Pinares forest (Figure 4b), a large dispersion of ETfrac data in late autumn and winter can be perceived; it may be related to the high daily ranges of wind speed (standard deviation = 1.9 m s-1) and precipitation (standard deviation = 3.9 mm) during that period. The gradient of water potential in the soil-plant-atmosphere complex causes soil water to be carried through the plant and transpired through the leaves. When the precipitation rate is low and there is not enough soil moisture present, the water potential gradient is reduced and does not favour transpiration. Wind also influences the evapotranspiration process, as it carries moist air over the surface of leaves, soil and water bodies, renewing the air and keeping it prone to receive more steam. Thus, the daily variation of these factors influences the increase and decrease of ETa and explains the behaviour of Kc.Ks in the Pinares forest.
The box-plot graph of the daily actual evapotranspiration in Figure 5 is divided into colours corresponding to the seasons for the two study areas. As in the Brazilian semiarid zone the four typical seasons are not well-defined like in the Pinares forest, only rainy (first semester) and dry seasons (second semester) are distinguished. The individual box-plots correspond to the ETa data of each of the pixels of the images that represent the two studied sites. There is a greater amplitude in the Caatinga forest ETa data (Figure 5a) than in the Pinares forest (Figure 5b).
The Caatinga forest is an area of natural vegetation and, hence, characterized by a wider range of species and a higher irregularity in the distribution of vegetation density, if compared to the Pinares forest. It may also be highlighted that there are pixels of reservoir water or rivers and pixels of exposed soil from roads and badlands in both areas; these pixels influence the respective maximum and minimum ETa peaks. Many pixels in the Caatinga area show an ET0 equal or close to zero in almost all the images. In some cases, such as on 10/Sep/1998, 1/Jul//2007, 18/Aug/2007, 9/Jul/2016, 10/Aug/2016 the median was zero; note that the first and last of these three years (1998 and 2016) were very dry (see Figure 3a). In the Pinares forest the medians are higher in summer and spring. Autumn and winter have the lowest atmospheric demands of the year and, as a consequence, they present the lowest medians.
As there is no agricultural production around the study areas during some time spans of the year, such as the dry season at Caatinga forest and winter at the Pinares forest, it was difficult to choose a cold pixel for the execution of SEBAL; that being so, a pixel within a water body was selected. This circumstance makes it a somewhat inconvenient method to be used in natural areas. Still, the ETa data obtained using SEBAL adjust well to the ETa data assessed by Penman-Monteith times Kc and Ks (Figure 6ab). Teixeira et al. (2009) also found a good correlation (R2 = 0.87) between the daily ETa data by SEBAL and field data from a mixed agricultural and natural ecosystem in a Brazilian semiarid region.
In both forests, the correlation between NDVI and ETawas greater in areas of less dense vegetation (Figure 7). As the most vegetated areas have a more constant NDVI throughout the year, the annual variations of ETa are more dependent on meteorological factors (related to the atmospheric demand for humidity) than on vegetation. Areas of less dense vegetation, on the other hand, show greater variation in NDVI according to the season. In the case of the Caatinga forest, NDVI is highly influenced by the variation of soil moisture between the rainy and dry months, and in the Pinares vegetation the NDVI varies depending on the four seasons of the year. It is also possible to observe greater NDVI ETa correlations in the Caatinga than in Pinares. This is due to the strong deciduousness that exists in Caatinga vegetation in the dry months, causing the NDVI to vary more throughout the year.
3.3 Temporal Stability Index (TSI) of actual ET
In Figure 8 it may be noticed that the zones of lower and higher TSI are always the same in both areas; what changes from season to season is the intensity. The areas with the highest TSI are the ones that respond strongest to differences in soil moisture. In the Caatinga forest, these are the highest regions with denser vegetation and, consequently, a huge capacity to expand leaf mass and increase evapotranspiration when soil moisture is higher after precipitation. High TSI values can also be observed in areas of exposed soil at the edges of the study area. There herbaceous vegetation grows during the rainy season, disappears during the dry season and leaves the soil bare. According to Tasumi (2019), in bare soils ET overestimations are more frequent that underestimations, since there the ETa in dry seasons should be near zero. The overestimated amount is not negligible when compared to annual precipitation.
The same observations in relation to exposed soil areas can be made in the Pinares forest, especially during winter and spring. There, however, unlike in the Caatinga forest, the areas with the highest TSI have sparse vegetation, as they produce a more substantial herbaceous layer under proper environmental conditions.
The Caatinga forest presents a greater extension with lower temporal stability than the Pinares forest (Figure 8). This is due to the greater variety of species in the Caatinga with different transpiration strategies and growth stages throughout the year, as well as to the fact that vegetation density throughout the area demonstrates a greater variation.
3.4 Trends in evapotranspiration (ET)
Evapotranspiration is a very complex process and influenced by many factors as precipitation, temperature, radiation, wind speed (Yang et al., 2019), type and stage of vegetation. That is why it is important to identify eventual existing trends in variables that influence ETa trend interpretations (Jaafar & Ahmad, 2020). In order to investigate the reasons for ETa changes, three basic factors that control ET were calculated for the study period (Figure 9 and Table 3): ET0 and temperature (that indicate the atmospheric demand); as well as precipitation, which is an indicator of water availability for evapotranspiration. According to Yang et al. (2019), surface moisture, which is affected by precipitation, is the basic material for the ETa process and also the main limiting factor in arid and semiarid regions.
When no monthly value of ETfrac produced by SEBAL was available, values from literature were used to calculate the ETa. In the Caatinga area, maximum temperature, annual Penman-Monteith ET0 and annual ETapresent a discernible positive trend over the period 1963–2017. Annual maximum air temperature increased at 0.027 °C per year (which means in arithmetic progression 1.5 °C over 55 years). There are no trends for annual rainfall (583 mm yr-1) or minimum temperature (20 °C) in the Caatinga. The ET0 displays a mean value of 5.5 mm d−1 (total of 1990 mm yr−1), a minimum of 1298 mm (during the year of 1999, Figure 9a), a maximum of 2466 mm (in the drought and hot year of 2016, Figure 9ab) and an increase of 3.5 mm yr−1, which means that it rose by 192.5 mm during the entire period. ETa showed a comparable significant positive trend with an annual mean of 877 mm yr-1, a variation that is similar to ET0 throughout the period (Figure 9c) and an annual increase of 2.2 mm (121 mm over the investigated time span).
The Pinares forest trends are similar to those observed in the Caatinga. Precipitation has a non-significant negative trend, while all the other analysed variables showed a significant positive temporal trend. The maximum temperature (17.5 °C) showed a trend of 0.023 °C per year, which supports the results of del Río, Fraile, Herrero, and Penas (2007) who worked with a historical series (1961–1997) for the Castile and Leon region and assessed a trend of 0.02 °C yr-1. The minimum temperature (5 °C) of the Pinares forest showed a trend of 0.014 °C per year; again, the authors op. cit. computed a very similar result for the same variable: a temporal trend of 0.01 °C yr-1 for the minimum temperature in Castile and Leon. ET0 showed a trend of 7.0 mm yr-1 (191 mm for all the period), higher than the trend of 10 mm decade-1 found by Vicente-Serrano et al. (2014). The mentioned authors used a larger time series for their study (1961-2011). They found a more linear ET0 pattern until 1990, after that period, coinciding with the years of our historical series, ETa had a greater annual increase. That means that we find a greater tendency for an annual increase in ETadue to the size of the historical series studied (1992-2018). In the series established for the present study, the minimum ET0 was 897 mm (during the wet year of 1997, the rainiest in the series, Figure 9d), and the maximum value was 1278 mm in 2017 (the year with the highest maximum temperature, Figure 9e). Likewise, the ETa (average 513.4 mm yr-1) presented an increasing trend of 3.9 mm yr-1 (106 mm for the period), so that the ETa variation was also similar to ET0throughout the period (Figure 9f).
In both study areas, the annual ETa was assessed to be superior to annual precipitation. The Pinares pine stone forest benefits from the moisture of the water table of the Pisuerga, Duero, Cega and Adaja rivers. In the Caatinga forest, the explanation may stem from the fact that part of a reservoir, named Beguê, is within this area and it receives input from a larger area (932 km2). Thus, the reservoir evaporation and the maintenance of a higher humidity in the surrounding soil could justify an annual ETa greater than precipitation in the area.
It is interesting to observe that annual ET0 and ETa in Caatinga Forest (10° South latitude) are almost twice than in Pinares forest (40° North latitude). This is an interesting result that must be studied in other semiarid environments with different latitudes.