Seasonal Ecosystem Productivity in a Seasonally Dry Tropical Forest (Caatinga) Using Flux Tower Measurements and Remote Sensing Data

Abstract

The Caatinga dry forest encompasses 11% of the total continental territory of Brazil. Nevertheless, most research on the relationship between phenology and ecosystem productivity of Brazilian tropical forests is aimed at the Amazon basin. Thus, in this study we evaluated the seasonality of ecosystem productivity (gross primary production—GPP) in a preserved Caatinga environment in northeast Brazil. Analyses were carried out using eddy covariance measurements and satellite-derived data from sensor MODIS (MODerate Resolution Imaging Spectroradiometer, MOD17 and MOD13 products). In addition to GPP, we investigated water use efficiency (WUE) and meteorological and phenological aspects through remotely sensed vegetation indices (NDVI and EVI). We verified that ecosystem productivity is limited mainly by evapotranspiration, with maximum GPP values registered in the wetter months, indicating a strong dependency on water availability. NDVI and EVI were positively associated with GPP (r = 0.69 and 0.81, respectively), suggesting a coupling between the emergence of new leaves and the phenology of local photosynthetic capacity. WUE, on the other hand, was strongly controlled by consecutive dry days and not necessarily by total precipitation amount. The vegetation indices adequately described interannual variations of the forest response to environmental factors, and GPP MODIS presented a good relationship with tower-measured GPP in dry (R² = 0.76) and wet (R² = 0.62) periods.

1. Introduction

Deforestation and land use change affect the preservation of tropical forests, demanding great efforts from the scientific community to investigate and discover the main natural controls of biophysical and phenological aspects in these ecosystems. With that, it is possible to better understand their ecological functions, which in turn allows the extrapolation of in situ observations to larger areas through remote sensing data, for example. This bigger picture provides more accurate and detailed scenarios regarding the impacts of climate change and land use change on landscape and the environment. In this perspective, studies on the characterization of the biosphere–atmosphere interactions are crucial to highlight the importance of tropical forests to the global climate balance [1,2,3,4]. Given the particularities and the variety of physiognomies within a single biome, these studies are also useful for the development of more accurate biosphere–atmosphere models [5,6,7], for the validation of remote sensing products [8,9,10,11], for the characterization of vegetation response to the natural seasonality of the environment [7,12,13], and they can also be used as proxies in the analysis of climate change impacts and implications.

In Brazil, most research in this field is concentrated in the Amazon region [14,15,16,17,18]. Thus, there is a relative scarcity of studies on the seasonality of phenological aspects of other Brazilian biomes and their atmospheric controls, or even studies regarding the validation of remote sensing data over these areas. However, recent studies indicate that other Brazilian biomes such as the Caatinga behave as efficient carbon sinks if compared to other dry forests across the globe [19]. These authors show that even in the dry season, the Caatinga uptakes from 145 to 169 gC m⁻² year⁻¹ due to the low respiration rates of the ecosystem. Nevertheless, there is little point-based experimental evidence apart from these results and, therefore, remote sensing estimates can be important sources of information on the ecological functions of the Caatinga, as suggested by the study by Ferreira [20]. Although observations at a single point are not representative for the entire ecosystem, such observations are useful to improve the understanding of ecosystem productivity and gas exchange in different biomes, being a common analysis in the literature [2,4,19] in studies using the eddy covariance technique in a single tower, given the high cost of keeping many experiments covering the same area. Such data can serve as a subsidy for several actions, such as validating remote sensing products for larger areas.

Remote sensing methods provide means to monitor carbon dioxide (CO₂) exchange and several other vegetation growth parameters in almost all kinds of ecosystems, from boreal forests [21] to the different mosaic vegetation patterns of tropical forests [9]. Furthermore, they may be associated with information collected through eddy covariance systems, providing more robust measurements to characterize energy and water fluxes in ecosystems (e.g., Maselli [22]). In addition, due to limitations in establishing long-term series in these environments, measurements in shorter periods (2 years or more) are already efficient in characterizing the seasonality of local variables and validating remote sensing products [2,3,19,20].

Carbons fluxes, including photosynthesis and respiration, are strongly coupled to the water cycle in the various terrestrial ecosystems [23,24]. Water-use efficiency (WUE) is commonly defined as the ratio between gross primary production (GPP) and evapotranspiration (ET) at the ecosystem level [25], representing an important ecophysiological index relating water and carbon cycles. At the ecosystem level, WUE is typically controlled by climate and soil variables, including precipitation, air temperature, vapor pressure deficit (VPD), and soil water content, due to their roles in energy partitioning and canopy conductance (e.g., Tong an Song [26,27]), although the magnitude and direction of the response might be different or even opposite (e.g., Liu [28]).

As observed in many semi-arid ecosystems, studies on the dynamics of CO₂ exchange in the Caatinga biome are still incipient. Regarding the role of the Caatinga in the global and regional carbon balance, there is still need to corroborate in situ GPP measurements, to comprehend phenological controls and WUE patterns in the biome, and to better understand the Caatinga response to years of extreme drought such as 2015 [29]. Therefore, this study proposes an assessment of the seasonality of ecosystem production and its correlations with atmospheric controls (meteorological variables), vegetation indices (NDVI—Normalized Difference Vegetation Index and EVI—Enhanced Vegetation Index, which are the most used greenness proxies in the literature according to Helman [30]), and GPP data estimated by remote sensing. Our objective is to answer the following questions: (i) what are the correlations, and which are the main controls and contributions of evaporative fluxes to the seasonality of Caatinga GPP? (ii) Can phenological patterns observed at the field level also be perceived at larger scales (MODIS NDVI and EVI)? (iii) What are the patterns of WUE in the Caatinga and how do they behave during climate extreme years such as 2015?

2. Materials and Methods

2.1. Description of the Study Area

The study was conducted in a preserved fragment of the Caatinga biome, which is a seasonally dry tropical forest (SDTF), located at the Seridó Ecological Station (ESEC-Seridó) (6°34′4″S, 37°15′0″W), in the Rio Grande do Norte state (Figure 1), Brazilian Semiarid region. The ESEC-Seridó is a federal conservation unit of the Caatinga biome, managed by the Chico Mendes Institute for Biodiversity Conservation (ICMBio), with an area of 1163 ha of preserved forest.

The vegetation is composed of dry, xerophyte, deciduous and semi-deciduous species, with shrublike and arboreous structure, sparsely distributed and reaching up to 8 m in height, in addition to herb patches that thrive only during the wet season [31]. The climate in the region is semiarid, with approximately 700 mm of total annual rainfall distributed during the wet season (January to May). The mean annual temperature is of approximately 25.0 °C, and the accumulated evapotranspiration rates are high all year round (1500 to 2000 mm yr⁻¹) and relative humidity of the air is of approximately 60% [32].

2.2. Eddy Covariance Climate and Carbon Flux Data

Experimental data were collected in the period from January 2014 to December 2015 through an eddy covariance system installed in an 11 m height flux tower belonging to the Brazilian National Institute of the Semiarid (INSA) and integrating the National Observatory of Water and Carbon Dynamics in the Caatinga Biome (NOWCDCB) network. The analyzed variables were daily and monthly means of air temperature (Ta, °C), net ecosystem exchange (NEE, gC m⁻² day⁻¹), gross primary production (GPP, gC m⁻² day⁻¹), ecosystem respiration (Reco, gC m⁻² day⁻¹), vapor pressure deficit (VPD, kPa), relative humidity (RH, %), rainfall (mm), latent heat flux (LE, in W m⁻²), sensible heat flux (H, in W m⁻²)m and net radiation (Rn, in W m⁻²). High (10 Hz) and low frequency data (sampled every 5 s and stored as half-hour averages) were recorded. The high frequency data consist of measurements of CO₂ concentration and water vapor and the three components of wind speed (u_x, u_y, u_z), using an Integrated CO₂/H₂O Open-Path Gas Analyzer and 3D Sonic Anemometer (IRGASON, Campbell Scientific, Inc., Logan, UT, USA). Atmospheric pressure was measured using an Enhanced Barometer PTB110 (Vaisala Corporation, Helsink, Finland) and air temperature by an HMP155A probe (Vaisala Corporation, Helsink, Finland) of low frequency include net radiation (Rn), air temperature (Tair), and relative humidity (RH), in addition to precipitation. Rn measurements were performed using a net radiometer model CNR4 (Kipp and Zonen B. V., Delft, Netherlands Ta) and RH data were measured using a model HMP45C temperature and relative humidity probe (Vaisala Corporation, Helsinki, Finland).

Net Ecosystem Exchange (hereafter referred as NEE) is the sum of the eddy CO₂ flux (FCO₂), calculated as the covariance between the fluctuations of the vertical wind speed (w′) and the density of CO₂ (c′) and the rate of change of CO₂ stored in the air column below the EC measurements height (Sc). Since no concentration profile was installed at the site, we opted for the discrete approach, assuming that the CO₂ concentration inside the canopy can be estimated as an approximation [33]

CO₂ concentrations in the form of Net Ecosystem Exchange (NEE) were partitioned into Gross Primary Production (GPP) and Ecosystem Respiration (Reco) which is equivalent to the sum of autotrophic and heterotrophic respiration, using a flux partitioning method based on night time as described by Reichstein [34]. Thus, the GPP was estimated according to Equation (1):

GPP = NEE − Reco

(1)

in which Reco and GPP were calculated using the online tool provided by the Max Plank Institute (Max Planck Institute for Biogeochemistry—http://www.bgc-jena.mpg.de/~MDIwork/eddyproc/ accessed on 24 April 2022).

The NEE was modeled based on daytime data using the common rectangular hyperbolic light response curve model [35]:

NEE = (α∙β∙R_g)/(α∙R_g + β) + γ

(2)

where α (μmol C 〖J〗⁻¹) is the light utilization efficiency and represents the initial slope of the light response curve, β (μmol C m⁻² s⁻¹) is the maximum CO₂ absorption rate of the canopy at light saturation, γ (μmol C m⁻² s⁻¹) is the ecosystem respiration, and R_g (W m⁻²) is the global radiation. Full details on the instrumentation, energy and carbon fluxes partitioning and corrections, and the eddy covariance method in general can be consulted in previous literature using the same dataset [2,3,19].

Caatinga ET was estimated through latent heat as follows (Equation (1)):

$$\mathrm{ET}=86,400\cdot \frac{\mathrm{LE}}{\mathrm{L}}$$

(3)

where L (J kg⁻¹) = 103 × (2500 − 2.37 × Ta) and 86,400 is the daily integration factor.

The Water Use Efficiency was calculated as follows [28]:

$$\mathrm{WUE}=\frac{\mathrm{GPP}}{\mathrm{ET}}$$

(4)

2.3. MOD13A2 Vegetation Indices Data

Vegetation indices were obtained through the remote sensing MOD13A2 v6—Terra Vegetation Indices product from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor onboard the Terra satellite. The MOD13A2 product provides 16-day composites of NDVI and EVI data at 1 km spatial resolution, which were both used in this study [36]. Data are downloadable at the following website: https://lpdaac.usgs.gov/products/mod13a2v006/ accessed on 22 April 2022. The indices were extracted from the MODIS product at the exact location of the ESEC-Seridó flux tower.

2.4. MOD17A2 GPP Data

Remotely sensed GPP data obtained from the MOD17A2 version 6.0 product [37], derived from the MODIS sensor onboard the Terra satellite were used. This product provides 8-day composites data at a 500 m spatial resolution and is available for download in https://modis.gsfc.nasa.gov/data/dataprod/mod17.php accessed on 22 April 2022. These data are calculated based on the concept of light use efficiency [38], through the relation between incident photosynthetically active radiation (PAR), the fraction of photosynthetically active radiation absorbed by plants (FPAR), and the actual light use efficiency (Ɛ) of vegetation [38].

$$\mathrm{GPP}=\mathsf{Ɛ}\times \mathrm{APAR}$$

(5)

where APAR is the absorbed photosynthetically active radiation, which is calculated as the product between the FPAR—derived from the MOD15A2H product [39]—and the PAR. PAR values are obtained from the Global Modeling Assimilation Office (GMAO) reanalysis [40], set and corresponding to 45% of the total incident solar energy in the visible spectrum (0.4 to 0.7 μm). In the MOD17A2 product, Ɛ values are derived from the attenuation of its maximum value (εmax) due to two environmental stresses: (1) minimum temperature, which can inhibit photosynthesis, reducing enzymatic activity, and (2) VPD, because high VPD can reduce stomatal conductance [41].

2.5. Statistical Processing

Statistical analysis and data processing were carried out using the R software, version 4.0.5 [42]. High frequency data were processed as 30-min averages using the EdiRe software (version 1.3.2.11, University of Edinburgh, Edinburgh, Scotland). Pearson’s correlation coefficient (r) and simple linear regression models were used to assess the relationship between pairs of variables. The coefficients of the models were evaluated individually through Student’s t-test at the 5% significance level (p-value < 0.05) and the 95% confidence intervals were used to display the confidence limits of the fitted model. Additionally, the quality of the model’s fit was evaluated through the proportion of the variance of the response variable explained by the model via coefficient of determination (R²) and the significance (p-value < 0.05) of the model when compared to the fit containing only the intercept. Principal component analysis (PCA) was also applied between meteorological parameters and the components of CO₂ fluxes (GPP, Reco and NEE) to explain the variance and covariance structure of data through non-correlated linear combinations of the original dataset. In this study, we opted to use the correlation matrix to create these combinations, since PCA is sensible to variables with different scales and units [43].

3. Results

3.1. Seasonal Dynamics of Climate, Vegetation Indices, Carbon, and Energy Fluxes

Figure 2 shows the mean monthly accumulated evapotranspiration, vapor pressure deficit (Figure 2a), and air temperature and relative humidity averages (Figure 2b), as well as monthly accumulated rainfall. Evapotranspiration, relative humidity of the air, and rainfall are inversely related to VPD and air temperature, which peaked during the dry season of 2015 (29.8 °C, Figure 2b). Air temperature varied 11% between monthly minimum and maximum values, while relative humidity varied more sharply, roughly 40% between minimum and maximum values. ET was higher in April (94.4 ± 8.5 kg H₂O m⁻² month⁻¹), coinciding with the highest rainfall values. Mean ET during the wet season was 280.9 ± 23.9 kg H₂O m⁻², which is 225% higher than in the dry season (86.4 ± 8.4 kg H₂O m⁻²). At the annual scale, mean VPD in 2015 was 2.6 kPa while in 2014, it was 2.2 kPa, which may be explained by the slightly lower precipitation (513 and 466.5 mm in 2014 and 2015, respectively) and higher air temperature, with a consequent decrease in relative humidity of the air between these years (

Table 1. Annual average and standard deviation of climate and energy flux variables measured during 2014 and 2015. Rainfall refers to accumulated values.

Variable	2014	2015
Tair (°C)	27.8 ± 3.8	28.4 ± 4.1
RH (%)	54.8 ± 17.9	48.9 ± 17.1
VPD (kPa)	2.2 ± 1.18	2.6 ± 1.20
ET (kg H2O m−2 day−1)	1.2 ± 1.05	0.8 ± 0.93
Rainfall (mm)	513	466.5
Rn (W m−2)	161.0 ± 219.9	160.3 ± 220.2
H (W m−2)	83.9 ± 142.0	93.2 ± 152.2
LE (W m−2)	35.5 ± 66.7	25.5 ± 56.6

). In the wet season, 83% of mean annual rainfall occurred, which accounted for 53 days in which rainfall occurred in 2014 and 26 days in 2015.

The seasonal dynamics of the vegetation indices (EVI and NDVI) show the pattern of surface phenology regarding canopy structure and the development of the vegetated area (Figure 3a). We can observe that the seasonality of vegetation is directly related to local precipitation patterns (Figure 2). The NDVI and EVI presented higher values at the peak of the wet season (NDVI > 0.6; EVI > 0.4), subsequently reducing in magnitude with the establishment of the dry season, reaching NDVI < 0.4 and EVI < 0.2 and indicating the beginning of leaf senescence in the Caatinga. The occasional occurrence of rainfall events in the dry–wet transition season were sufficient to increase vegetation activity in November 2014 (NDVI > 0.5; EVI > 0.3), which led to an increase in latent heat flux, suggesting that the onset of the wet season exerts a strong control in the production of biomass in this ecosystem. These data are strongly related to energy fluxes, with higher NDVI and EVI values occurring in the wet season when evapotranspiration (and LE) is also higher (Figure 3b).

Net radiation peaked at roughly 200 W m⁻² day⁻¹ in November. Sensible heat flux accounts for most of the energy balance partitioning during the dry season, while latent heat flux is superior during the wet season. Table 1 shows the mean and deviation of the studied variables in each experiment year.

3.2. Seasonal Dynamics of the Carbon Cycle, WUE, Tower-Measured GPP, and Remotely Sensed GPP

Carbon fluxes (GPP, Reco, and NEE) and WUE are presented in Figure 4. The seasonality of these parameters is evident, as well as a relative coupling with observed rainfall pulses. GPP peaked in April 2014 and the entire period monthly mean was of 1.89 ± 0.87 gC m⁻² month⁻¹. MODIS GPP presented a similar behavior, but with slightly higher GPP values, with a maximum of 10.0 ± 0.84 gC m⁻² month⁻¹ in April 2014. Monthly mean GPP ranged from 1.0 to 3.7 gC m⁻² month⁻¹. On average, accumulated GPP during the wet season (163.9 gC m⁻² month⁻¹) was 126% higher than in the dry season (72.5 gC m⁻² month⁻¹). The lowest carbon fluxes occurred in the dry season, when rainfall accounts for roughly 17% of the total annual expected amount of precipitation. Total accumulated ecosystem productivity (estimated through GPP) was 414.7 gC m⁻² yr⁻¹ in 2014 and 334.0 gC m⁻² yr⁻¹ in 2015. The seasonal pattern of NEE was also modulated by local hydrometeorological conditions, and the Caatinga site behaved as a carbon sink (NEE < 0). Higher carbon uptake was observed during the wetter months (Figure 4a). With the end of the wet season, uptake diminished, but the Caatinga still behaved as a carbon sink until the end of each year. Carbon uptake was different between the wet season of both studied years and this is probably not associated with rainfall differences between years since it was roughly 10%. The most plausible explanation is the difference in the number of rain days in each year (Figure 4b), with a 49% decrease from 2014 to 2015. This hypothesis agrees with the higher WUE observed in 2015 (1085.2 gC kg⁻¹ H₂O) in relation to 2014 (827.3 gC kg⁻¹ H₂O). Indeed, WUE represents the trade-off between carbon gains and water consumption during photosynthesis, and the ecosystem tends to shift its behavior towards a more efficient use of water in drier conditions (Figure 4b). In both years, WUE was inversely related to ET. The computation of WUE is important to characterize specific water cycles and the effect of drought on the water balance and carbon sequestration. Data show that WUE in the studied ecosystem decreased with the increase of rain days, and WUE values rose sharply with long sequences of consecutive dry days.

3.3. Principal Component Analysis and GPP Correlations

PCA shows that GPP and ET are positively correlated in the wet season (Figure 5b), but they are not correlated during the dry season (Figure 5a). Furthermore, it clearly shows that WUE and the variables related to heating (H) and drying (VPD) of the atmosphere are positively correlated and near each other in the unit circle. This means that these variables have a similar temporal behavior. This correlation is particularly strong in the dry season (Figure 5a). A similar relationship can also be observed between Rn and Tair, probably due to soil moisture depletion, with most of available energy being directed to the heating of the air, increasing air temperature, and, consequently, VPD. MODIS GPP satisfactorily represents tower-measured GPP both in the dry season (Figure 6a) and the wet (Figure 6b) season (R² = 0.76 and R² = 0.62, respectively). We also assessed the biophysical performance of vegetation indices in the Caatinga regarding GPP dynamics. Figure 7 shows the correlation between experimentally derived GPP and the remaining meteorological and carbon variables, while Figure 8 shows the linear fit between GPP and vegetation indices (NDVI and EVI). Tower GPP was better correlated with the EVI (r = 0.81, p < 0.01) than with the NDVI (r = 0.69, p < 0.01).

4. Discussion

We observed that GPP, Reco, and NEE showed marked seasonal variation in sync with precipitation. The GPP (positive values) and NEE (negative values) increased after leaf expansion (higher EVI and NDVI) at the beginning of the rainy season, increasing the soil water availability necessary to supply the photosynthetic processes of the plants, as a consequence the rates of GPP overcame the Reco and the study area in the Caatinga acted as a carbon sink. Reco showed similar seasonality to NEE and GPP, following the pattern of rainfall. Mekonnen [44] reports that the greater availability of water in the soil is associated with high microbial activity, contributing to an increase in autotrophic respiration, and consequently in ecosystem respiration. Results showed that evapotranspiration strongly controls GPP, varying with the local rainfall pattern throughout the year. ET rates reach up to 100 kg H₂O m⁻² month⁻¹ in the wet season and remain lower than 10 kg H₂O m⁻² month⁻¹ in the dry season. Overall, the studied Caatinga site behaves as a carbon sink, as previously shown in other studies [19], with productivity pulses directly related to water availability, with the highest rainfall and evapotranspiration rates occurring in April. Both tower data and remote sensing data indicate that this ecosystem strongly responds to local rainfall, but water availability may not be the only factor limiting vegetation growth since GPP was similar in both studied years despite 2015 being a slightly drier year. Thus, even if our results comprise only one site, we hypothesize that available energy may also be a limiting factor for this type of vegetation, which would explain the correlation patterns shown in Figure 5a (dry season), with GPP and Rn closely located at the same unit circle, and in Figure 5b (wet season) where they are located at opposite circles. Similar results were reported in seasonal tropical forests in Brazil, where rainfall seasonality and soil water content were not relevant for the energy partitioning because vegetation growth was more limited by available energy than water [45,46]. This is a plausible explanation since high precipitation rates are usually associated with increased cloud cover and, therefore, reduced incident solar radiation.

The MOD17A2 product demonstrated dexterity in describing the average monthly seasonal variability of the GPP in the Caatinga. The results show a consistent correlation between gross primary production (GPP Tower) and MODIS gross primary production (GPP MODIS), R² = 0.76 in the dry season and R² = 0.66 in the rainy season, where the GPP MODIS overestimated the GPP Tower data (Figure 4 and Figure 6). This overestimation is most evident during the period from February to May 2014 and February to April 2015, which corresponds to the wettest period in the years 2014–2015. On the other hand, during the dry periods (especially from July to November) the GPP MODIS was more related to the observed values. The relationship between tower GPP and MODIS GPP agrees with results previously reported in the literature [20,47,48,49], with stronger correlations if compared to similar analysis carried out in tropical forests [49].

The interannual variability of CO₂ uptake by terrestrial sinks is mainly associated with changes in land use and meteorological factors, as the carbon balance is strongly related to its high spatial and temporal variability. Among them, precipitation plays an important role due to its remarkable seasonality in semi-arid regions [19,50]. In these ecosystems, the availability of natural resources such as water, plant biomass, nutrient dynamics in the litter, and soil are modulated by the occurrence of rainfall. [51]. Satellite derived greenness-associated vegetation indices (such as the NDVI and EVI) have commonly been used as proxies in several biochemical characterization studies and on studies regarding their correlation with biophysical variables [30]. Our results show that there is a good correlation between GPP and EVI/NDVI patterns in the Caatinga, satisfactorily responding to surface energy and water availability variations. This kind of response is important because a prolonged leaf senescence period might actually be a reaction of the vegetation due to water availability at the end of its growing stages, which might shift a prolonged period behaving as carbon sink in years with longer than usual growing stages. During periods of abundant rainfall, the availability of nutrients in the soil is greater, resulting in a rapid and efficient uptake of nutrients by plants allocated for leaf development, reflecting higher productivity [52]. Rain events during the dry season, when leaves are still present in the canopy, may be more productive than rain events at the end of the dry season, when a second leaf emergence period might be necessary [53]. Such a combination of climate events and conditions could influence on vegetation mortality and on species dynamics during forest succession, which may be possible to monitor through satellite observations such as the products used in this study.

We also found that WUE reduces with monthly rainfall, and sharply rises with consecutive dry days. At the annual scale, WUE was higher in the drier year (2015). The relative magnitudes of GPP and ET represent the sensitivity of different biological processes to drought in different seasons and years. Although both GPP and ET reduced during the dry season, the response of the latter was much more prominent than the first, leading to a higher WUE in 2015. The expressive increase of 31% in WUE (Figure 4b) during 2015 is associated to a reduction of 49% in the number of rain days, which in turn represented a reduction of 34% in mean annual ET. These results corroborate other studies found in the literature [54,55,56,57]. Zhou [58] suggested that plants may prioritize resistance to higher temperatures via an uneven opening of stomates to improve leaf transpiration and dissipate heat while maintaining low WUE. Our study also indicates that WUE in the Caatinga is inversely related to evapotranspiration in both wet and dry seasons, which is not observed with air temperature. The negative relationship between WUE and VPD found in our study was also previously reported in the literature at the ecosystem level or even at the leaf level [59,60]. These results have important implications on the understanding of climate change effects on water and carbon exchange processes in the tropical Caatinga environment, since future climate projections for the region indicate an increase in temperature and a reduction in rainfall [61]. Such a scenario, coupled with the observed trend of longer periods with consecutive dry days, would lead to more frequent and more intense drought events with the potential total aridification of the region [29], which would lead to an increase in WUE of the Caatinga at the ecosystem level.

5. Conclusions

The cycles of ecosystem productivity and evapotranspiration are strongly associated in the Caatinga biome, where the number of dry days is a more limiting factor for WUE than the actual amount of precipitated water. The seasonality of productivity pulses is directly related to variations in local precipitation. The characterization of canopy phenology showed a good correlation between the EVI and NDVI seasonality, and energy fluxes and local productivity. Our results showed that both vegetation indices peaked a few days before the end of the wet season, which was also observed through GPP data. GPP was positively associated with the variation in rain days, suggesting a coupling between the production of new leaves and photosynthetic capacity phenology, with a pattern of higher productivity and maximum carbon assimilation at the peak of the wet season, under non-limiting water availability conditions. The NDVI and EVI managed to accurately describe interannual variations in vegetation response to environmental factors, corroborating the fact that water availability (also associated with days with/without rain) is the main phenological factor of this ecosystem. Indeed, it can alter vegetation phenology during extreme years and consequently affect ecosystem productivity. Despite GPP MODIS data considerably overrating tower-measured GPP, our study retrieved better correlations between datasets if compared to other similar studies conducted in tropical forests, highlighting the potential usability of these remote sensing tools for the monitoring of Caatinga phenology.