**4. Discussion**

Evapotranspiration showed well-defined seasonality in all sites, varying at the annual scale with local precipitation, radiation availability and increasing temperatures. Highest evapotranspiration values were registered in the warmer months and/or in months with more radiation or precipitation. Evapotranspiration reached maximum values in the Cerrado and Pantanal sites in wetter months, reaching up to 7 mm day−<sup>1</sup> in November, with a mean value slightly above 3 mm day−1. In the Caatinga site, even in the month with the highest evapotranspiration rates (October), overall values did not reach the same intensity of the other sites, showing the particularity of this site regarding its arid climate (BSh) according to Köppen's classification (Liang et al., 2020) [67]. Crucial studies have already been conducted to estimate the magnitude, seasonality and controls of ET at the local scale using eddy covariance measurements in Brazil (Da Rocha et al., 2009; Costa et al., 2010) [6,68]. These studies usually show that dry season evapotranspiration is higher than in the wet season, and Rn is the main ET control in tropical rainforests (such as the Amazon), while it is not true for the Caatinga, Cerrado and Pantanal sites. Our results showed that in these sites, evapotranspiration decreases throughout the dry season, reaching the lowest rates between August and September, when the Amazon features its highest values. Even during the dry season, high evapotranspiration rates can be explained by the predominance of pioneer tree species (*Combretum lanceolatum* and *Vochysia divergens*) with high photosynthesis rates and stomatal conductance (gs) (Dalmagro et al., 2013; 2016) [69,70], associated with the ability of these species to extract water from deep storages containing similar water content to that of inundation areas and wet periods (Sanches et al., 2011; Vourlitis et al., 2011; Dalmagro et al., 2013; da Silva et al., 2021) [29,69,71,72].

In the dry season of the Cerrado site, an inverse relationship was found between LE and Rn, which is typical of savanna ecosystems, where the root system does not reach deep water storages, but part of the vegetation is adapted and relies on senescence mechanisms of tree leaves and dormancy of grasses (3). A pulse in productivity (increased GPP) that is directly related to the increase in water availability (rainfall) can be observed, particularly from October on, which marks the beginning of the transition from the dry to the wet season. As expected, all sites reduced productivity in response to increased rainfall variability, with the most productive ecosystems being those with the highest precipitation rates (Amazon and Cerrado). The low rainfall rates observed in the Caatinga in October were sufficient to drive a sharp increase in productivity, which indicates that more frequent rainfall events in this environment could lead to more nutrient availability and the mitigation of water stress through leaf absorption mechanisms. A further indicator of these aspects can be observed through the remarkable variability in evapotranspiration during this month in the Caatinga. Furthermore, this biome presented the best correlations between GPP and evapotranspiration. The comparison between observed and remotely sensed data showed that both datasets represent the same seasonality, despite their weak correlation. Satellite data accurately represents the strong response of the Pantanal and Caatinga sites to local precipitation, while the results are not satisfactory regarding the dry period in the Amazon and Cerrado. The weak correlation between GPP data in tropical forests is well documented (Zhu et al., 2018) [55] and is generally attributed to light-use efficiency, which is underestimated by the MODIS GPP algorithm (Sjöström et al., 2013) [73]. Additionally, potential uncertainties regarding the FPAR might also affect MODIS GPP estimates, which have also been discussed in the literature (Liu et al., 2017) [60]. Despite its wide use, our results corroborate previous studies reporting the need to use MODIS GPP data with

caution for interannual and intra-annual studies, as suggested by Zhu et al. [55]. Our study revealed that humid biomes present lower WUE, and drier biomes present higher WUE, with decreasing values with the onset of the wet season and maximum values in the dry season. This result is consistent with previous studies in other environments (Liu et al., 2017; Singh et al., 2014; Yu et al., 2008) [60,62,74]. Ito and Inatomi [75] claim that forest ecosystems presented WUE values comparable or superior to arid (such as caatinga) and shrub biomes (such as Cerrado and Pantanal), with an intermediate value of 0.6–1.2 g C kg−<sup>1</sup> H2O for humid forest biomes. The average WUE for the Pantanal is within this range.

WUE responds differently to seasonal water availability in each biome. Overall, more productive biomes present high WUE (Xue et al., 2015) [76]. In sparsely vegetated areas, such as the Cerrado (savannas) and the Pantanal wetlands, WUE varies greatly throughout the year, despite presenting lower values than the Caatinga. From January to April, when precipitation is high in all sites (except Caatinga), the Amazon site featured an increase in WUE daily values, while the Cerrado WUE decreased, and the Pantanal WUE remained unaltered. From May, rainfall triggers a sharp decrease in WUE over these regions until September. Our results reveal important implications for the understanding of climate change effects on carbon and water exchange processes in tropical biomes, because the projected reduction in water availability over these sites due to the increasing number of dry days [77] may lead to an increase in WUE at the ecosystem level. However, increasing temperatures may further increase or reduce ecosystem WUE at the monthly scale. Consequently, changes in ecosystem WUE due to climate change will depend on the relative impact of these changes in precipitation and temperature. Caatinga WUE is highly dependent on water availability, with lower values and variability in the wetter months and higher values and variability in the drier months. The results found here corroborate other studies in other parts of the Caatinga biome [78–82], and it is important to report that owing to measurement difficulties, few studies have systematically compared global patterns of WUE of terrestrial ecosystems across different biomes or have analyzed the seasonal variability of WUE in relation to weather conditions, because ecosystem WUE is slightly different from plant WUE [61]. Plant physiologists consider WUE at leaf or stand scales and are mainly interested in relations between total or above-ground biomass, stem biomass or net CO2 uptake to transpiration or evapotranspiration (ET), and although uncertainties associated with site-to-site variation in site quality criteria, flux measurement methods, calculations and data quality control still exist, ongoing standardization and quality assurance efforts enable global integration with other tools [61].
