Space–Time Dynamics of Mortality and Recruitment of Stems and Trees in a Seasonally Dry Tropical Forest: Effect of the 2012–2021 Droughts

Abstract

Seasonally dry tropical forests (SDTFs) represent about 41.5% of the planet’s tropical forests. The objective of this study was to characterize the annual mortality and recruitment patterns of stems and trees between the years 2012–2021 in a Caatinga remnant in the state of Pernambuco, Brazil, through geostatistical modeling, and to associate the drought events recorded in the region with vegetation dynamics. Mortality and recruitment of stems and trees were monitored in 80 permanent plots located in an SDTF remnant, counted year by year between 2012 and 2021. The standardized precipitation index (SPI) was calculated to quantify the deficit or excess of rainfall in the evaluated period. The data were then subjected to geostatistical analysis based on the calculation of classical semivariances. As a result, there was a loss of 68.33% of trees and 61.93% of stems in the forest community during 2012–2021, which were associated with the water deficit caused by drought events recorded based on precipitation data and SPI calculation for the region. The Gaussian semivariogram model better represented the spatial variability of mortality and recruitment of stems and trees. An accumulative effect of droughts on increasing mortality rates and reducing recruitment during the study period was observed. The relationship between tree and stem mortality and recruitment rates and drought events highlights the impact of water deficit on vegetation, emphasizing the importance of considering extreme climatic events in the proper management of natural resources.

1. Introduction

Seasonally dry tropical forests (SDTFs) are distributed in extensive non-contiguous areas in the African continent, Latin America, and the Asia–Pacific region, representing about 41.5% of the planet’s tropical forests [1,2,3]. SDTFs are characteristic of semiarid regions with high water deficit, irregular and poorly distributed rainfall, with dry periods lasting from 3 to 7 months of the year and mean annual temperatures above 17 °C [4,5].

The Caatinga biome is predominantly located in the Brazilian Northeast, and harbors the largest and most biodiverse nucleus of SDTFs in the world [6,7,8], featuring a wide diversity of endemic species and landscapes responsible for generating and providing various ecosystem services for the population [9]. However, anthropogenic actions mainly related to subsistence agriculture and extensive livestock farming have historically led to the degradation of its vegetation, and whose effects are mainly intensified by population growth and the expansion of urban and industrial areas [10,11,12]. Associated with the unsustainable exploitation of SDTF vegetation, there is an environmental liability to be compensated, highlighting the need for developing research aimed at understanding the dynamics of the dry forest after suffering anthropogenic disturbances. In this sense, the space–time monitoring of vegetation constitutes an essential tool to assist in the search for alternatives for the sustainable use of forest resources.

Studies on the dynamics of forest communities or populations are conducted through analyzing variation in floristic composition and forest structure in space and over time through assessments of productivity, growth, recruitment, and mortality [13,14,15]. It is important to note that the occurrence of trees and shrubs with multiple stems is a striking characteristic of Caatinga species [9,16,17]; thus, such a feature should be considered in studies of vegetation population growth dynamics so that its components (mortality, recruitment, growth, increment, and others) are the result of values related to individuals and stems.

Tree mortality is one of the main processes that influence the dynamics and functioning of ecosystems worldwide [18,19]. According to the study’s inclusion level, tree recruitment is also considered an important tool in vegetation monitoring, as this component represents the outcome of regeneration and the establishment of new trees [17,20]. Thus, it is essential to study the spatial occurrence patterns of mortality and recruitment processes in order to understand vegetation dynamics, which in turn can aid in the decision-making for the sustainable management of forest resources, whether timber-related or not.

Among spatial modeling techniques, geostatistical interpolation models are effective in characterizing spatial dynamics through kriging maps [21,22]. Geostatistical interpolation models have been widely used in the characterization of forest species, such as mapping spatial patterns and basal area dynamics for successional groups of tree species in a remnant of a mixed tropical forest in southern Brazil [23]. Other studies, such as by Buddenbaum et al. [24], who classified conifer species and age classes; by Hernández-Stefanoni et al. [25], who mapped forest species; and by Negrón [26], who related host spatial distribution to subsequent tree mortality caused by Dendroctonus ponderosa, demonstrate the effectiveness of geostatistical modeling in spatial characterization. However, studies addressing spatial characterization and analysis of tree and stem mortality dynamics in seasonally dry tropical forests are scarce.

Tree mortality is quite complex as it can be associated with various biotic stress factors (e.g., wind, fire, pests, insects) and abiotic factors (i.e., drought-induced mortality) [27,28]. However, prolonged drought events are known to be among the main causes of alterations in forest structure, affecting regeneration and increasing mortality rates of individuals in a forest [29,30]. This phenomenon can induce tree mortality directly through water scarcity or reduced carbon availability, as well as by making trees more susceptible to biotic stresses [31,32].

Thus, it is necessary to understand and monitor the occurrence of drought phenomena to identify their behavior and variability more accurately to predict and prevent potential impacts on vegetation [33]. Such monitoring can be performed with the aid of drought indices. The exploration, employability, and applicability of drought indices in various studies across different biomes in Brazil and worldwide are currently gaining momentum with the aim of characterizing the different effects (e.g., anthropogenic, natural, and others) they cause on populations and biomes. An example of this is the standardized precipitation index (SPI), one of the most widely explored drought indices [34,35,36,37].

The SPI is widely used by researchers worldwide in the fields of climatology, meteorology, agriculture, forest management, and water resources due to its ability to calculate drought phenomena for any time scale and its simplicity of calculation, depending only on precipitation data [38]. Studies such as by Oliveira-Júnior et al. [39], who used the SPI to define wet and dry periods in the state of Alagoas, Brazil, made it possible to characterize drought phenomena in the state and the impacts generated for the population and the Caatinga biome over a time scale of more than 50 years. Similar results were observed by Merabti et al. [40], who characterized the spatial and temporal variability of drought and humidity based on the SPI in northeastern Algeria. Similarly, Zarei et al. [34] in Iran and Thotli et al. [41] in India assessed changes in the spatial and temporal patterns of drought in their territories based on the SPI.

Although geostatistical modeling and the use of drought indices such as the SPI have been separately employed in ecological studies, their combined application to assess spatial patterns of tree and stem mortality and recruitment in seasonally dry tropical forests remains scarce. In particular, there is a notable gap in studies that explore how drought events, characterized through the standardized precipitation index (SPI), influence the spatial distribution and temporal patterns of tree and stem mortality and recruitment. Given the increasing frequency of droughts in semiarid regions such as the Brazilian Northeast, which account for more than 50% of drought events in the country [39,42], this type of integrated analysis is essential to provide accurate insights into the functioning and resilience of dry forest ecosystems, closely linked to population development, the economy and, mainly, the dynamics of local vegetation.

Therefore, understanding changes in tree mortality and recruitment rates in the Caatinga over the years is essential for developing predictive models of dynamics and for sustainable forest management for production. Thus, the objective of this study was to characterize the annual patterns of mortality and recruitment of stems and trees between the years 2012–2021 in a Caatinga remnant in the state of Pernambuco, Brazil, through geostatistical modeling, as well as to associate drought events recorded in the region based on the SPI with vegetation dynamics.

2. Materials and Methods

2.1. Characterization of the Study Area

The study was conducted at Fazenda Itapemirim, owned by Agrimex Agroindustrial Excelsior S.A., located in the municipality of Floresta, PE, in the mesoregion of São Francisco Pernambucano and the microregion of Itaparica, at coordinates 8°30′49″S and 37°57′44″W, zone 24S-UTM (WGS84). The total area of the farm is approximately 6000 hectares. The study was conducted in two sampling areas with different disturbance histories (Figure 1) as follows. Area I is situated near the state highway PE-360 to the south of the farm. The last change in land use and land cover in the area occurred in 1987 when the vegetation was cleared using chains for eucalyptus planting; however, the area was abandoned and has been undergoing regeneration since then. Area II is located inside the farm, and its vegetation is considered preserved as it does not have a recent history of cutting. Forest products are only occasionally exploited for maintenance the fence bordering the area. Both areas are approximately 50 hectares each and are currently extensively grazed by animals, mainly goats.

The municipality of Floresta is located within the ecoregion of the Southern Sertaneja Depression. This region is characterized as one of the most impacted by anthropogenic action, with few protected areas both in number and in the total area and protection category [42,43]. The region’s climate is classified as BSh-hot, semiarid, steppe type, marked by a dry season and a rainy season [44]. It has a mean annual temperature of 27.8 °C and a mean annual precipitation of approximately 520.7 mm, with rainfall concentrated from January to June, with the wettest months being March and April [45]. The soil in the region is characterized as shallow Chromic Luvisols, with sandy to medium surface texture [46]. The predominant practice in the region is extensive grazing of domestic animals, mainly goats [47].

The vegetation in the area is shrubby–arboreal, with the occurrence of trees with multiple stems, cacti, herbaceous strata, and bromeliads, such as Bromelia laciniosa Mart. ex Schultes f. and Neoglaziovia variegata (Arr. Cam.) Mez., as well as species such as Commiphora leptophloeos (Mart.) J.B. Gillett, used by traditional populations to make handicrafts (Figure 2). The occurrence of deciduous species is also observed, a typical phytogeographical feature of a wooded steppe savanna [48].

2.2. Data Collection

Data were collected from 80 permanent plots (40 in each area), each measuring 20 m × 20 m (400 m²), spaced 80 m apart with a 50 m buffer, totaling 3.2 hectares. The plots, established in 2008, have been monitored annually since 2011. A Trimble Catalyst digital antenna connected to a smartphone was used to obtain coordinates of the plots, tracking GPS L1/L2C signals with sub-meter accuracy (0.3 m to 0.75 m).

Mortality and tree recruitment were monitored annually from 2012 to 2021, covering 10 years of data. It is important to note that most individuals in the studied forest fragment had multiple stems, so both stem mortality and recruitment were recorded accordingly.

In this study, a stem was considered dead when there were no visible signs of meristematic activity, with a dry stem and bark lacking living tissue, and when peeling of the bark revealed characteristics indicative of mortality. An individual tree was considered dead when all of its stems were dead. Trees and stems were considered recruits when they met the minimum inclusion criteria for being classified as arboreal, with a circumference at 1.30 m above ground level (C) ≥ 6 cm, following the permanent plot measurement protocol for Caatinga. All bifurcations below 1.30 m on the same tree meeting this criterion (C ≥ 6 cm) were considered stems.

2.3. Climate Pluviometrics and Air Temperature Characterization via TerraClimate

Monthly frequency data from the TerraClimate model were used to characterize rainfall and air temperature (Tar, °C) patterns and variability in the study region. These data are covered by water and climate balance data, with a spatial resolution of approximately 4 km (1/24°) and a temporal resolution from 1958 to the present [49]. The TerraClimate dataset is divided into primary and secondary climate variables. Primary climate variables include maximum temperature, minimum temperature, vapor pressure, rainfall accumulation, downward shortwave radiation at the surface, and wind speed. Secondary climate variables include reference evapotranspiration (ASCE Penman–Monteith standardized model), runoff, actual evapotranspiration, climatic water deficit, soil moisture, snow water equivalent, Palmer drought severity index (PDSI), and vapor pressure deficit.

Annual rainfall (mm) and mean air temperature (Tar, °C) were used for the present study. The data were specifically obtained from the Climate Engine platform (a platform for processing TerraClimate image data with georeferenced point data), available at the following address: “https://climateengine.com/ (accessed on 23 March 2024)”. The rainfall and mean temperature obtained for the study area were then associated with data on tree and stem mortality and recruitment, aiming to verify if there is a relationship between these variables.

2.4. Analysis of the Standardized Precipitation Index (SPI)

The formulation of the SPI is based on the gamma probability density function (Equation (1)), calculated on a monthly basis, where α is the shape parameter (α > 0), β is the scale parameter (β > 0), determined through the maximum likelihood method, and x represents the amount of rainfall, which can vary according to α and β. The assigned values are normalized and transformed to a normal distribution (i.e., mean zero and variance one) [50]. The SPI calculation uses only rainfall values [39,51], categorizing a drought event when the SPI presents continuous negative values, and when the SPI is positive, it marks the end of the drought event [52]. Further details on the mathematical formulations and statistical procedures used in the SPI calculation can be found in the study by McKee et al. [52].

$$\mathrm{F}\left(\mathrm{x}\right)=\underset{0}{\overset{\mathrm{x}}{\int }}\mathrm{f}\left(\mathrm{x}\right)\mathrm{dx}={\displaystyle \frac{1}{\Gamma (\alpha ){\beta }^{\alpha }}}{\int }_0^{\mathrm{x}}{\mathrm{x}}^{\alpha -1}{\mathrm{e}}^{-{\displaystyle \frac{\mathrm{x}}{\beta }}}\mathrm{dx}$$

(1)

After this calculation, the SPI was categorized according to

Table 1. Classification of the SPI according to McKee et al. [52].

SPI	Drought Intensity Class
≥2.00	Extremely wet
1.00 to 1.99	Very wet
0.50 to 0.99	Moderately wet
0.49 to −0.49	Near normal
−0.50 to −0.99	Moderately dry
−1.00 to −1.99	Severely dry
≤−2.00	Extremely dry

and analyzed on a monthly scale (SPI-1). The SCI package from the R software version 4.0.3 library was used for the SPI calculation [53].

The aim of determining the SPI was to quantify the deficit or excess of rainfall during the evaluated period (2012–2021) and associate these results with the data on tree and stem mortality and recruitment, seeking to verify if there is a relationship between them and drought events recorded in the study area.

2.5. Analysis of the Soil Moisture Index (SMI) and Analysis of the Potential Evapotranspiration (ETp)

The SMI was calculated according to Moawad [54] using the LST as follows:

$$\mathrm{SMI}={(\mathrm{L}\mathrm{S}\mathrm{T}}_{\mathrm{m}\mathrm{a}\mathrm{x}}- \mathrm{LST})/{(\mathrm{L}\mathrm{S}\mathrm{T}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{L}\mathrm{S}\mathrm{T}}_{\mathrm{m}\mathrm{i}\mathrm{n}})$$

(2)

where SMI is the soil moisture index and LST_max, LST_min, and LST are the maximum, minimum, and retrieved LST values, respectively.

The land surface temperature (LST) and potential evapotranspiration (ETp) data were obtained from observations by Landsat 7 and 8. These images were collected for the dry seasons of each year between 2012 and 2021 using the Earth Engine Evapotranspiration Flux (EEFlux, version 0.20.17), as outlined by Allen et al. [55]. The EEFlux tool is based on the METRIC (Mapping Evapotranspiration at High Resolution with Internalized Calibration) model [56]. In the EEFlux process, LST values are generated using a fixed atmospheric calibration, where temperature gradients near the surface are derived as an indexed function of the radiometric surface temperature. This method removes the need for highly precise surface or air temperature measurements. The actual evapotranspiration is derived from Landsat images representing the 24 h ETp, using EEFlux’s automated standard calibration. Within this approach, actual ETp is determined as a residual from the surface energy balance, as shown in the following equation:

$$\mathrm{LE}={\mathrm{R}}_{\mathrm{n}}-\mathrm{H}-\mathrm{G}$$

(3)

where LE is the latent heat flux (energy used in the evapotranspiration process (W m⁻²)), R_n is the net radiation (W m⁻²), G is the soil heat flux (W m⁻²), and H is the sensible heat flux (W m⁻²). The EEFlux calibration utilizes the Landsat thermal band and shortwave bands to estimate the surface energy balance, as well as to estimate vegetation amount, albedo, and surface roughness.

Version 0.20.17 of EEFlux employs automated image calibration by assigning ETo values to the “hot” and “cold” pixels of the surface temperature spectrum in the scene. The LE is estimated at the exact moment of the satellite pass for each pixel, and the instantaneous ET is then calculated by dividing the LE by the latent heat of vaporization according to the following equation:

$${\mathrm{E}\mathrm{T}}_{\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}}=3.600\mathrm{LE}/\lambda \rho w$$

(4)

where ET_inst is the instantaneous evapotranspiration (mm h⁻¹); 3.600 converts seconds into hours, λ is the latent heat of vaporization (J kg⁻¹), and ρw is the water density (∼1000 kg⁻³).

2.6. Analysis of Arboreal Community Dynamics—Mortality and Recruitment

The vegetation dynamics were expressed in demographic terms (count). Based on Sheil et al. [57], changes in population size over time intervals were assumed to be a constant proportion of the initial population size; thus, the mean annual rates of mortality (M) and recruitment (I) of individual stems and trees were calculated according to Equations (5) and (6) below.

$$\mathrm{TM}=\left[1 -{\left({\displaystyle \frac{{\mathrm{N}}_0-{ \mathrm{N}}_{\mathrm{m}}}{\mathrm{No}}}\right)}^{{\displaystyle \frac{1}{\mathrm{t}}}}\times 100\right]$$

(5)

$$\mathrm{TI}=\left[1 -{\left(1-{{\displaystyle \frac{{\mathrm{N}}_{\mathrm{i}}}{\mathrm{N}}}}_{\mathrm{t}}\right)}^{{\displaystyle \frac{1}{\mathrm{t}}}}\right]\text{×}100$$

(6)

where t is the time elapsed between inventories, N₀ and N_t are the initial and final counts of individual stems/trees, respectively, and N_m and N_i are the numbers of dead stems/trees and recruited ones, respectively.

It is important to note that for the calculation of tree and stem mortality rates in individuals with multiple stems, mortality was assessed at the individual level, defined as the death of all stems of a given tree. For example, if a tree had 10 stems, it was considered dead only when all 10 stems were dead. In the mortality analysis, only the mortality of individual trees was considered, while stem mortality was evaluated based on the total number of stems of each tree.

2.7. Statistical Analysis

2.7.1. Validation of the TerraClimate Data

First, linear correlations from 2011 to 2021 were used based on the monthly obtained data (rainfall and Tar), utilizing the coefficient of determination (R²), Pearson’s correlation coefficient (r), and the root mean square error (RMSE, mm/°C) to assess the TerraClimate data. As an adjustment and validation factor with rainfall and temperature data from TerraClimate, the observed data were obtained from a meteorological station of Instituto Nacional de Meteorologia (INMET) in the municipality of Ibimirim, PE, the mesoregion of São Francisco Pernambucano and the microregion of Itaparica, under the coordinates 8°30′34″S and 37°42′42″W at an altitude of 434.23 m [58], and then these observed data were compared with those from TerraClimate (where the meteorological station is located). From the validation of the sensor data, the monthly rainfall and Tar values of the study area were extracted from 1 January 2011 to 31 December 2021.

2.7.2. Confidence Interval Analysis

The confidence intervals were calculated based on the means and standard deviations of the samples at the confidence level of 95% for mortality and recruitment of stems and trees between 2012 and 2021.

2.7.3. Boxplot Analysis

A boxplot analysis was conducted to identify outliers and statistical properties to find three percentiles (median and interquartile range) and the minimum and maximum values (whiskers, represented by the lines of the standard boxplot), constituting the five-number summary, for the studied variables of mortality and recruitment of trees and stems over the 10-year period.

2.7.4. Moran’s I Index Analysis

Moran’s I test [59] was used to assess the spatial autocorrelation of the variables of interest, recruitment and mortality, for stems and trees from 2012 to 2021. The objective was to identify spatial patterns in the data, such as clustering or dispersion, and to assess the spatial dependence between the sampling units. For the analysis, the geographic coordinates (longitude and latitude) of the sampling units were used to calculate the spatial weight matrix based on the 4 nearest neighbors of each sampling unit.

The calculation of Moran’s I index was performed for each variable of interest and year using the moran.test() function of the spdep package. The index measures the intensity of spatial autocorrelation, ranging from –1 (opposite distribution) to +1 (perfect clustering), with values close to 0 indicating spatial randomness. To assess the significance of the results, a randomization test was performed, in which the values of the variables of interest were randomly permuted and Moran’s I index was recalculated for each permutation. The p-value was obtained by comparing the value calculated for the original data with the values calculated in the permutations, and it was considered significant when the p-value was less than 0.05.

The application of this test allowed us to verify whether the patterns of recruitment and mortality of stems and trees were spatially grouped. Based on the results obtained in the Moran’s I test, which indicated significant spatial autocorrelation for the variables over the years, a geostatistical analysis was performed to model and predict the values of these variables in non-sampled locations.

2.7.5. Geostatistical Modeling and Kriging Maps Analysis

Data on tree and stem mortality and recruitment obtained from the 80 permanent plots in the study areas over a period of 10 years were used for the geostatistical analyses. The data underwent geostatistical analysis based on the calculation of classical semivariances (Equation (7)), which estimates the spatial structure and dependence among pairs of observations.

$$\gamma \left(\mathrm{h}\right)={\displaystyle \frac{1}{2\mathrm{N}\left(\mathrm{h}\right)}}\sum _{i=1}^{\mathrm{N}\left(\mathrm{h}\right)}{\left[\mathrm{Z}\left({\mathrm{X}}_i\right)-\mathrm{Z}({\mathrm{X}}_i+\mathrm{h})\right]}^2$$

(7)

where γ(h) is the estimator of experimental semivariance obtained from the sampled values Z(X_i) and Z(X_i + h), N(h) is the number of pairs of measured values separated by the lag vector or distance, h is the distance between sample pairs (i.e., it is the distance between two samples), and Z(X_i) and Z(X_i + h) are the values of the ith observation of the regionalized variable collected at points X_i and X_i + h (i = 1, …, n), separated by vector h.

Spatial dependence was analyzed by fitting the semivariogram based on the semivariance estimation using the GS+ program version 7.0 (Gamma Design Software). The data were fitted to the spherical, exponential, and Gaussian models (Equations (8)–(10)), respectively, according to Deutsch et al. [60]. The spherical, exponential, and Gaussian models are referred to in the literature as transitive theoretical models and are more common for fitting semivariograms [61].

Spherical model:

$$\gamma \left(\mathrm{h}\right)=\left\{\begin{array}{ll}{\mathrm{C}}_0+\mathrm{C}\text{×}\left[1.5{\displaystyle \frac{\mathrm{h}}{a}}- 0.5{\left({\displaystyle \frac{\mathrm{h}}{a}}\right)}^3\right],& \mathrm{for} 0 \text{≤} \mathrm{h} \text{≤} a\\ {\mathrm{C}}_0+\mathrm{C},& \mathrm{for}\hspace{1em}\mathrm{h} \text{>} a\end{array}\right.$$

(8)

Exponential model:

$$\gamma \left(\mathrm{h}\right)={\mathrm{C}}_0+\mathrm{C} \text{×}⌊1-\exp \left(-{\displaystyle \frac{3\mathrm{h}}{a}}\right)⌋$$

(9)

Gaussian model:

$$\gamma \left(\mathrm{h}\right)={\mathrm{C}}_0+\mathrm{C} \text{×}\left[1-\exp \left(-{\displaystyle \frac{{\left(3\mathrm{h}\right)}^2}{{a }^2}}\right)\right]$$

(10)

where γ(h) is the estimator of the experimental semivariance, C₀ + C is the sill (i.e., the nugget effect plus the variance scatter, given by the acronyms C₀ and C, respectively), h is the distance between sample pairs, and is the range (m).

The best-fitting models of the adjusted semivariograms were validated using jackknifing cross-validation, where the mean should be close to zero and the standard deviation close to 1 [62]. The software used for this analysis was GEO-EAS^® [63]. The degree of spatial dependence (DSD) was classified according to Cambardella et al. [64], suggesting strong dependence (Sd) (DSD < 25%), moderate dependence (Md) (DSD between 25 and 75%), or weak dependence (Wd) (DSD > 75%), as represented in Equation (11).

$$\mathrm{DSD}(\%)={\displaystyle \frac{{\mathrm{C}}_0}{{\mathrm{C}}_0+{\mathrm{C}}_1}}$$

(11)

The Quantum GIS (QGis) program version 3.12 (Quantum Geographic Information System) was used to create the kriging maps.

3. Results and Discussion

3.1. Validation of TerraClimate Rainfall and Temperature

Figure 3 presents the validation of observed data with the estimated data based on monthly and yearly accumulated data of rainfall (mm) and mean air temperature (°C) obtained from a meteorological station of Instituto Nacional de Meteorologia (INMET) and the data estimated by the TerraClimate model. The calibration and validation of data estimated by global models for the Brazilian territory, such as the TerraClimate model, have been addressed and applied in studies of climatic variables, as indicated in the study by Filgueiras et al. [65]. The validation of these data allows us to accurately explore climatic data in regions where there are no meteorological stations.

Through the coefficient of determination (R²), it was found that the fit of the station with TerraClimate rainfall data was satisfactory, with values of 0.48 and 0.88 for monthly and annual data, respectively. Pearson’s correlation coefficient was 0.69 and 0.94 for monthly and annual data, respectively, showing an adequate fit and enabling the exploration and applicability of the estimated data. These results are consistent with those of Silva et al. [12], who evaluated the spatiotemporal dynamics of vegetation cover and soil degradation in the microregion of the Ipojuca Valley in the state of Pernambuco, Brazil’s semiarid region, through vegetation indices, in which the authors validated rainfall data estimated by TerraClimate with an INMET meteorological station in order to characterize the effects of rainfall on vegetation dynamics, obtaining an R² of 0.59 and Pearson’s r of 0.77 for the monthly data explored. The values observed in this study for the root mean square error (RMSE) were 35.42 and 63.99 mm for monthly and annual data, respectively, corroborating with an RMSE of 26.86 mm observed in the study by Silva et al. [12].

R² showed a satisfactory fit for air temperature data with values of approximately 0.90 and 0.89 for monthly and annual validations, respectively, followed by Pearson’s r of 0.95 and 0.94 and RMSE of 0.66 and 0.14 °C. Similar results were observed by Filgueiras et al. [65] when comparing and calibrating climate variables estimated by the TerraClimate model with INMET stations for the entire Brazilian territory from 2000 to 2017, obtaining R² and RMSE values on the order of 0.72 and 2.04, respectively.

The lower R² values for rainfall data are explained by the high variability of this variable over space and time, and mainly by the overestimation of data estimated by global sensors over a longer timeframe, directly influencing significant reductions in R². Arias and Barriga [66] evaluated the performance of rainfall data using CHIRPS and TerraClimate in a Colombian Andean basin, emphasizing that a larger sample grid of estimated data leads to a greater overestimation in the dataset. Thus, reducing the grid results in a better fit, consequently increasing the reliability of the estimated data. Therefore, the results for annual data correlations were satisfactory due to the smaller sample grids for correlating observed and estimated datasets.

3.2. Annual Rainfall and Temperature Dynamics from 2000 to 2021

Based on the validation of the TerraClimate data, pixel values within the experimental plot region were extracted for the total rainfall and mean temperatures between 2000 and 2021, as shown in Figure 4. It is noted that both rainfall and mean temperatures were close to the climatological normal recorded for the municipality (27.8 °C and 520.7 mm, respectively) until the year 2011. However, there has been a downward trend in rainfall since 2012, while the annual mean temperature has increased considerably.

There were years of severe droughts in the Brazilian Northeast during the study period (2012–2021). Particularly in the semiarid region of Northeast Brazil, extensive areas were affected by drought events recorded between 2012 and 2016, resulting in a significant water deficit and consequent stress on vegetation due to reduced water supply [67,68,69]. According to Marengo et al. [70], the hydrological year 2012–2013 showed the highest intensity of droughts, coinciding with the data obtained for the study region, in which the year 2012 was considered the most critical, with rainfall below 250 mm (Figure 4).

It can also be noted that the year 2014 experienced a peak rainfall of 619 mm during the drought period, which is explained by Mariano et al. [71] and Cunha et al. [72], as these years are considered transitional between the major droughts that occurred in the Brazilian Northeast. However, the years 2015, 2016, 2017, 2018, 2019, and 2021 had rainfall below 500 mm in the study area, corroborating the findings of Silva et al. [12], who assessed degradation in a microregion in the semiarid Brazilian Northeast using biophysical indices and rainfall maps, observing major drought events after 2016, with precipitation below 400 mm.

3.3. Standardized Precipitation Index (SPI) Monthly from 2000 to 2021

It is possible to observe in the SPI-12 time series between the years 2000 and 2021 in the experimental areas (Figure 5) and confirm with the rainfall data that there were drought records between the years 2012 and 2019, coinciding with the period of historic drought in the Northeast region of Brazil [70]. The SPI-12 value for 2014 categorized it as moderately wet, which is consistent with the findings of Mariano et al. [71] and Cunha et al. [72], who consider that year as a transition between droughts in the region due to rainfall. It is also noted that these results are similar to the mean rainfall obtained for the other years in the study area (Figure 4), where 2012 was considered the most critical year regarding water deficit, categorized as extremely dry (below −2.00) using the aforementioned index. In quantifying the occurrence of drought periods with SPI indices for the municipality of Ibimirim, PE, Brazil, Costa Júnior et al. [45] highlighted that there were 3 moderately dry episodes and 1 extremely dry episode in the 2010s, similar to the results obtained in this study.

There was an observed incidence of normal conditions for the region of this study between the years 2000 and 2006 due to the neutral phase of ENSO. On the other hand, the year 2008 was characterized as a year of heavy rainfall in the region (Figure 5), mainly due to the La Niña phenomenon in the Brazilian Northeast in 2007, resulting in increased rainfall as well as its frequency in the Brazilian Northeast. Since La Niña can last from 9 to 12 months, its effects and impacts can be felt in the following year, as was the case in the municipality of Floresta, impacted by heavy rains in 2008. The results found in this study are similar to those obtained by Jardim et al. [73] in analyzing the spatiotemporal climatic dynamics of municipalities along a longitudinal gradient from east to northeast in the state of Pernambuco, Northeast Brazil, from 1993 to 2018. They highlight that the effects of La Niña led to heavy rains in the municipality of Ibimirim (40 km from the study area) in 2008.

Analyzing the effects of droughts and rainfall from milder to extreme conditions becomes essential when studying such a biome as Caatinga, mainly constituted by xerophytic vegetation, characterized by small trees with twisted trunks and thorns and loss of leaves by trees during the dry season, leaving only the presence of white trunks with a characteristically dry vegetation [8,13].

3.4. Soil Moisture Index (SMI) and Potential Evapotranspiration (ETp)

The interannual variation of the soil moisture index (SMI) and the potential evapotranspiration (ET_p) for the 2012–2021 time series is presented in Figure 6. The SMI values are shown on the left vertical axis, ranging from 0.00 to 0.60, while the right vertical axis displays the ET_p values, with a scale from 0.00 to 9.00 mm year⁻¹. The year 2015 corresponds to the year with the highest soil moisture observed in the experimental area. However, moisture incidence is noted for the experimental areas of the farm, as the ET_p rates are relatively high, approaching 7 mm year⁻¹.

Corroborating the results observed in this study, Oliveira et al. [74] assessed drought conditions in the Northeast of Brazil (NEB) using drought indices and the SMI, highlighting the relationship between the SMI and the dynamics of droughts in semiarid regions. However, the authors suggest that for a more detailed analysis of soil moisture in a specific year, such as 2015, it would be important to analyze the SMI in conjunction with drought indices, such as the SPEI. In line with this recommendation, we observe that the SMI values are directly related to those of the SPI, the drought index used in this work. It is evident that the years of more extreme droughts, and their respective subsequent years, also coincide with the lowest SMI values.

There is significant variability in the SMI over the years. ET_p also exhibits annual variations, indicating changes in the atmospheric demand for water. In some years (e.g., 2013 and 2019), ET_p showed relatively high values, suggesting a higher water demand for evaporation and transpiration. In contrast, other years (e.g., 2014 and 2018) exhibited lower ET₀ values.

When analyzing the relationship between the SMI and ET_p, a complex dynamic is observed. In some periods, such as 2015, a peak in the SMI coincided with a relatively high ET_p, indicating that the soil water availability was sufficient to meet the atmospheric demand. On the other hand, in years such as 2017, a low SMI coincided with a moderate ET_p, indicating a possible water stress for the vegetation, where the atmospheric demand exceeded the soil water availability. The difference between ET_p and the SMI could be interpreted as an indication of potential water deficit. When ET_p is relatively higher than the SMI, vegetation may face difficulties in maintaining transpiration, which can lead to reduced growth and increased vulnerability.

3.5. Dynamics of the Shrubby–Arboreal Community

Mortality and recruitment of 513 and 361 trees were recorded in the period from 2012 to 2021 in Area I. Mortality for stems was 1430, while recruitment was 1054 stems. A total of 1270 trees and 4083 stems died in Area II over the 10-year period, respectively, and a total of 348 trees and 855 stems were recruited (

Table 2. Shrubby–arboreal community dynamics (C ≥ 6 cm) of two areas with different exploitation histories in the Caatinga remnant, Floresta, Pernambuco, Brazil, quantified for each area and expressed in the number of stems and trees.

Dynamics	Area I	Area II	Total
Sampling
Number of plots	40	40	80
Number of trees
Initial (no. of trees)	1096	2294	3390
Final (no. of trees)	944	1372	2316
Initial absolute density (trees ha−1)	685	1433.75	2118.75
Final absolute density (trees ha−1)	590	857.5	1447.5
Dead (no. of trees)	513	1270	1783
Recruits (no. of trees)	361	348	709
Survivors (no. of trees)	583	1024	1607
Mortality rate (% year−1)	6.12	7.75	7.19
Recruitment rate (% year−1)	4.71	2.88	3.59
Stem number
Initial (no. of stems)	3050	6418	9468
Final (no. of stems)	2674	3190	5864
Initial absolute density (stem ha−1)	1906.25	4011.25	5917.5
Final absolute density (stem ha−1)	1671.25	1993.75	3665
Dead (no. of stems)	1430	4083	5513
Recruits (no. of stems)	1054	855	1909
Survivors (no. of stems)	1620	2335	3955
Mortality rate (% year−1)	6.13	9.62	8.36
Recruitment rate (% year−1)	4.89	3.07	3.86

). Overall, there was a gradual decrease in tree density from 2118.5 to 1447.5 trees.ha⁻¹, as well as a reduction from 5917.5 to 3665 stems.ha⁻¹, respectively, corresponding to a loss of 68.33% of trees and 61.93% of stems between 2012 and 2021. The high loss percentage of trees and stems of the species in the mentioned areas coincides with the period of reduced annual rainfall and high mean temperatures (Figure 4), which may have resulted in higher evapotranspiration rates and, consequently, a negative water balance. In this sense, when evaluating the influence of climate on the dynamics of forest carbon accumulation in dry tropical forests in Costa Rica and Brazil, Calvo-Rodrigues et al. [32] found a significant correlation between seasonal climate variables such as temperature, rainfall, and potential evapotranspiration and annual mortality and carbon loss in forests, corroborating the results of the present study.

The tree mortality rate for the community was 7.19%, while it was 8.36% for stems. These values are higher than the recruitment rates, which were less than 4% (Table 2). These results indicate that mortality was nearly twice as high as recruitment in this fragment of dry forest during the years 2012–2021. In dry forests in India, Suresh et al. [75] recorded an average annual mortality rate for vegetation of 6.9% over a period of 19 years, yielding similar values to those obtained in the present study. Furthermore, according to Marengo et al. [70], the hydrological year 2012–2013 had the highest drought intensity in the semiarid region of Northeast Brazil, and the water deficit persisted throughout the semiarid region until 2016, coinciding with the study period and the highest mortality records, partly explaining the negative results obtained for the vegetation.

Mortality rates of 74.06% for stems and 71.23% for trees in the community were observed in Area II. Despite being considered “more preserved,” Area II exhibited higher mortality rates and lower recruitment rates. These findings suggest that, despite the historical use of Area I, it appears to be less impacted, with its remaining individuals demonstrating growth potential even during extended drought periods. In contrast, the effects in Area II were more pronounced. This outcome may be linked to the characteristics of natural regeneration in dry tropical forests, which often exhibit a high prevalence of sprouting, particularly from roots. This process significantly contributes to the resilience of these ecosystems [76,77,78].

The successional process in the area, driven by recent mechanized exploitation, may have led to an increase in resource availability and a reduction in interspecific competition. However, this pattern was not observed in Area II. These findings suggest that the extended period since the last exploitation in Area II has resulted in a higher tree density, which subsequently intensified intra- and interspecific competition for resources, particularly water. Such increased competition, especially during critical drought periods, has caused more pronounced fluctuations in mortality and recruitment, thereby negatively impacting vegetation dynamics. In this context, previous studies [79,80] emphasized that trees in denser stands tend to suffer greater competition with neighboring trees, which can result in higher mortality rates. Additionally, studies such as that by D’Amato et al. [81] assessing the effects of thinning on drought vulnerability in temperate forests highlighted that higher-density stands are more sensitive to changes in rainfall, especially during the growing season, and tend to have lower resistance and resilience. Therefore, the observed results suggest that competition among trees for resources, especially water, coupled with the drought periods recorded in the region, may be one of the main factors influencing mortality.

The recruitment/mortality ratio was 0.70 and 0.27 for trees, and 0.73 and 0.21 for stems in Areas I and II, respectively. It is expected that the pattern of arboreal community dynamics in mature and/or undisturbed tropical forests will reach stability over time through a balance between mortality and recruitment rates [82,83], indicating that there was a negative balance for the community in the study areas between 2012 and 2021.

The average mortality of stems was three times higher than the mortality of trees in the studied community, which was expected due to the high density of stems recorded in the areas. Thus, tillering shrubby–arboreal individuals are a regeneration strategy for characteristic species of dry forests, such as Caatinga, especially in locations subjected to natural and/or anthropogenic disturbance [9]. However, the high mortality of stems may also be related to the self-thinning of plants. As trees grow, there is a natural increase in internal competition for better conditions, leading to self-thinning in the plant itself, characterized as natural stem mortality. This process may have been intensified by the dry period recorded in the area of the present study.

The highest mortality rates for the studied dry forest remnants were recorded during the period from 2013 to 2017, accounting for 78.65% and 78.35% of the total mean dead stems and trees between 2012 and 2021, respectively (Figure 7A). The high mortality rates of species may be a consequence of the adverse conditions recorded in the area because, based on the standardized precipitation index (SPI), it coincided with the period considered the most critical for drought (Figure 5). Similarly, lower values of the soil moisture index (SMI) and the potential evapotranspiration (ET0) (Figure 6) indicate less water being lost to the atmosphere, which may occur due to various factors, such as a reduction in soil moisture or adverse climatic conditions. Additionally, the mean annual rainfall was less than 600 mm (Figure 4), except for the hydrological year 2013–2014, which is considered a transitional year between droughts [71,72].

In assessing the vegetation dynamics in the study area, Costa Júnior et al. [17] recorded a reduction in the number of individuals and stems from 2008 to 2019 and attributed these results to the vegetation’s response to severe drought conditions. These findings are also similar to studies by Suresh et al. [75] (in a dry tropical forest in India), by Aleixo et al. [84] (in the Amazon forest), and by Bradford et al. [85] (in Pinus ponderosa forests in the western United States), where the authors observed increased tree mortality during drought events.

Figure 7 also graphically presents the confidence intervals (CIs) calculated for tree mortality (Figure 7B) and stem mortality (Figure 7C) for the years 2012 to 2021. CIs are the ranges of values within which mortality is expected to fall according to a 95% confidence level. The years 2013 and 2014 stood out, with the CIs farthest from the annual mean mortality for trees and stems, higher than and non-overlapping with other years. Based on the boxplot analysis of mortality data, these respective years additionally showed the highest tree and stem mortality rates (Figure 7D,E), with an average mortality rate of 65 stems and ±17 for trees.

The higher mortality rates that occurred in these years may be attributed to the effects of severe droughts in 2012 in the study area, as shown in Figure 5. Similarly to the findings of this study, Barbosa and Kumar [86] assessed Caatinga’s response to drought using the Meteosat-SEVIRI normalized difference vegetation index for the period from 2008 to 2016 and associated the effects of droughts through the SPI, and emphasized that drought effects severely degraded the Caatinga biome, with particular emphasis on the SPI observed in 2012, classified as extreme drought, leading to a higher mortality rate of species in this biome.

It is important to highlight that, except for 2012, which was considered a year of extreme drought based on the SPI, with index values below –2 (Figure 5), no significant mortality values were recorded. This result can likely be explained by the delayed response of vegetation to drought. Native vegetation, especially in such ecosystems as dry tropical forests, tends to exhibit a certain level of resilience to water stress conditions. An example of this is the low potential evapotranspiration observed during the more intense drought years (Figure 6), indicating a reduced amount of water being lost to the atmosphere. This phenomenon may be related to the regulation of stomatal function, controlling the opening and closing of these structures to minimize water loss. These adaptation strategies may result in a gradual response to drought, with the most visible effects occurring after an extended period of stress, when the plant’s water and energy reserves begin to be depleted [87,88,89]. This behavior may be related to the presence of species adapted to drought conditions, which possess mechanisms of tolerance and water storage, allowing them to survive dry periods [90]. However, with the intensification of the drought period in the following years, the vegetation’s ability to remain healthy may be compromised, leading to a more pronounced increase in mortality, as observed in the subsequent years, 2013 and 2014 (Figure 6). Therefore, this behavior suggests that drought does not have immediate and linear effects on vegetation, with its consequences becoming more evident over time as water resources diminish and environmental stresses accumulate.

Moreover, upon analyzing the soil moisture index (SMI) (Figure 6), we observed that the lowest values occurred between the years 2013 and 2014. The SMI, derived from remote sensing data, ranges from 0 to 1, with desert areas showing low values, while agricultural areas exhibit higher values [91]. Based on this, we infer that between 2013 and 2014, the study area recorded the lowest soil moisture indices, likely as a result of the drought that began in the region in 2012. Thus, considering the adaptive capacity of species, the low soil moisture in the subsequent years may help explain the peak in tree and stem mortality observed in 2013 and 2014.

Droughts associated with high temperatures in the present study area are recognized as important drivers of tree mortality worldwide [92,93,94]. In summary, the amplification of hydrological stress reduces CO₂ uptake and water transport, consequently affecting the photosynthesis and plant respiration process [95,96]. Additionally, the high temperatures observed in the forest for the municipality of Floresta, PE, may increase the negative effect on vegetation due to the reduction in the efficiency of photosystem II, increased maintenance respiration, evaporative demand, and reduction in leaf area [97,98]. The combination of these factors likely contributed to higher mortality rates between 2013 and 2014, a period preceding the onset of the historical drought in the semiarid Northeast of Brazil [70], with low rainfall levels and increased temperatures (Figure 4), thereby negatively impacting the vegetation.

Tree mortality due to environmental disturbances functions as an important part of forest ecology. Therefore, some level of mortality in the stand is fundamental and often desirable for forming a mosaic of trees in different age classes and species compositions, resulting in greater resistance and resilience of the forest to multiple disturbances [99]. In this sense, according to Figure 7, it was observed that there were reductions in the mortality rates of stems and trees between the years 2015 and 2020, which, in addition to being associated with small variations in rainfall and temperature, may be associated with the vegetation’s resistance to environmental disturbances.

However, it is important to note that there was an increase in the mortality rates for both stems and trees in the year 2021 compared to the previous year, with values on the order of ±1 in 2020 to ±5 in the year 2021 for stems, and from ±1 in 2020 to ±5 in 2021 for trees (Figure 7D,E). This effect is mainly supported by the alignment with drought conditions based on the SPI for the year 2021 (Figure 5) and consequently low soil moisture (Figure 6). This reinforces the attention to the impacts associated with large-scale mortality associated with drought, which can negatively affect various ecological goods and services, including timber and fiber production, recreation, biodiversity, threatened species, and carbon sequestration [100].

There was significant recruitment in vegetation in 2012, as approximately 35% of stem recruitment was recorded between 2012–2021, and also 38.22% of trees (Figure 7A). It is observed that the confidence intervals and the dispersion of recruitment values in the boxplot for trees (Figure 8B,D) and stems (Figure 8C,E) reveal that the year 2012 stands out with larger confidence intervals, which are considerably higher compared to subsequent years. As evidenced in Figure 4, the annual rainfall was relatively high in the years preceding 2012, with notable rainfall in 2008, recording 789 mm in the study area. This scenario provided greater water availability for the plants, significantly contributing to the increase in tree and stem recruitment, as well as the reduction in the mortality rate in the corresponding year, as shown in Figure 7A.

However, the effects of drought in the region began to be observed in the vegetation recruitment later in the year 2013, extending until 2016, a period in which moderate and extreme drought events were recorded according to the SPI (Figure 5), as well as the greatest loss of soil moisture (Figure 6), which may have negatively influenced recruitment and contributed to the recording of only 35.15% of the total stems and 31.59% of the trees recruited in this period (Figure 8A). This result may be associated with the high sensitivity of natural regeneration to environmental constraints, especially in the early recruitment stages [101], as the tolerance of young individuals to environmental stress tends to be lower than that of adult trees [102,103].

Furthermore, extensive grazing, especially by goats, was not controlled in the study area, which may have also hindered the germination and establishment of new plants in the area. According to the results presented by Sousa et al. [104], grazing, when analyzed in isolation, was not responsible for significant changes in the functional diversity of woody vegetation. However, the interaction between grazing intensity and successional stage revealed relevant effects, indicating that the impact of grazing is strongly conditioned by the regeneration stage of the area. In environments under intense grazing in early stages of succession, the dominance of species with conservative strategies was observed, such as Mimosa tenuiflora, characterized by low leaf area and high tannin levels. This dominance may result in the reduction of structural and functional heterogeneity, hindering the colonization and development of other species, which may delay the advancement of ecological succession.

Areas subjected to a heavy grazing pressure show a significant reduction in the survival of tree seedlings, mainly due to the direct consumption of young structures and soil compaction caused by animal trampling, making it difficult for new individuals to establish themselves. The practice of semi-extensive or extensive livestock farming in semiarid regions becomes a factor of environmental disturbance, especially when animal density exceeds the carrying capacity of the ecosystem. In the medium term, intensive trampling compromises the soil by promoting its compaction during the rainy season and disintegration in the dry season, causing negative impacts on its physical, chemical, and biological properties. In the long term, these effects contribute to the irreversible degradation of soil and vegetation, increasing the vulnerability of the landscape to desertification processes [105,106,107].

In addition, seed germination in dry tropical forests is a seasonal process that occurs during rainfall, so the occurrence of shorter rainy seasons has the potential to reduce the germination of various species, and drought intensification can lead individuals to die before their establishment [108]. This consequently results in lower recruitment rates and higher mortality rates, as observed for the years 2020 and 2021, where there were no entries of stems or trees (Figure 8D,E) at significant rates. This phenomenon becomes concerning because the absence of entries of stems or trees in areas where mortality rates occur may be characteristic of a degradative process without area recovery.

3.6. Moran’s I Index

The spatial autocorrelation analysis using Moran’s I test was applied to assess the spatial dependence of the variables of interest, stem recruitment and stem mortality, from 2012 to 2021. The results indicated that stem recruitment showed significant positive spatial autocorrelation in several years, particularly in 2015, when Moran’s I value was 0.3279 (p-value = 1.921 × 10⁻⁷), suggesting strong spatial clustering. In 2019 and 2021, Moran’s I values for recruitment were also positive and significant, with low p-values, indicating that nearby sampling units in space exhibited similar values for this variable (

Table 3. Spatial dependence of Moran’s I index for mortality and recruitment of trees and stems in a remnant of Caatinga, Floresta, Pernambuco, Brazil.

Mortality
Year	Stems		Trees
	Moran’s I	p-Value	Moran’s I	p-Value
2012	0.3106	2.654 × 10−8 *	0.2874	1.628 × 10−8 *
2013	0.3651	8.783× 10−8 *	0.4629	3.692 × 10−11 *
2014	0.3337	5.716 × 10−7 *	0.2818	1.841 × 10−5 *
2015	0.5147	1.386 × 10−13 *	0.5166	8.267 × 10−14 *
2016	0.2529	1.360 × 10−4 *	0.1863	0.0129 *
2017	0.3429	1.360 × 10−3 *	0.1923	0.0029 *
2018	0.2518	1.359 × 10−4 *	0.2978	1.465 × 10−4 *
2019	0.2891	1.795 × 10−5 *	0.1743	4.882 × 10−3 *
2020	0.2215	0.0117 *	0.1915	0.0126 *
2021	0.2153	0.0123 *	0.2626	0.0145 *
Recruitment
Year	Stems		Trees
	Moran’s I	p-value	Moran’s I	p-value
2012	0.2192	1.268 × 10−4 *	0.1204	9.258 × 10−3 *
2013	0.2514	8.764 × 10−3 *	0.1204	9.258 × 10−3 *
2014	0.2270	2.071 × 10−3 *	0.1588	3.649 × 10−3 *
2015	0.3279	1.921 × 10−7 *	0.1656	3.441 × 10−3 *
2016	0.2499	1.853 × 10−2 *	0.2057	5.042 × 10−4 *
2017	0.1175	3.612 × 10−3 *	0.0211	5.477 × 10−2 *
2018	0.2513	1.065 × 10−2 *	0.2541	1.701 × 10−2 *
2019	0.2189	3.164 × 10−3 *	0.3129	1.776 × 10−3 *
2020	0.1911	2.005 × 10−2 *	0.1968	1.349 × 10−2 *
2021	0.2472	4.594 × 10−3 *	0.1699	6.504 × 10−4 *

Note: * special positive autocorrelation at 5% significance.

).

Regarding stem mortality, the results also showed significant positive spatial autocorrelation, especially in 2015, with Moran’s I value of 0.5147 (p-value < 1.386 × 10⁻¹³), indicating a strong tendency for spatial clustering of the sampling units (Table 3). In 2013 and 2014, Moran’s I values for mortality were also significant, with very low p-values, reinforcing the presence of a spatial pattern. These results may be associated with a high density of sampling units in specific areas, which favors the spatial clustering of mortality events.

The relationship between density and mortality in forests, especially in environments subjected to prolonged drought, can be better understood through the interaction between competition for water and the imbalance in carbon allocation. Forest communities with high density tend to trigger competition for water and, consequently, the imbalance in carbon allocation, which directly affects the survival and growth of plants, especially in water-scarce environments. These phenomena are particularly relevant in ecosystems subject to environmental stresses, such as prolonged droughts, and can have significant implications for vegetation dynamics and forest structure [109,110].

In environments subjected to prolonged drought, tree mortality is often associated with the inability to maintain a positive water and energy balance. In areas with high population density, competition for water intensifies, with plants forced to draw from the same water sources, often leading to a reduction in the availability of this resource for all species present. This can result in water stress, impairing vital plant functions such as photosynthesis, growth, and the maintenance of cellular integrity. This process arises from the interaction between the depletion of internal water and carbon reserves and the reduction in the flow of these resources relative to the metabolic demands of living tissues. The hydraulic and carbon systems are interdependent, and when compromised, they affect essential physiological functions, such as the maintenance of cellular integrity and plant survival [109,111].

Carbon is essential for energy production in plants, and in water scarcity situations, plants must allocate their resources strategically to ensure survival. Under water stress conditions, plants often redirect carbon to essential areas for maintaining life, such as roots, rather than using it for the growth of new leaves or stems [112,113]. This process is particularly evident in species adapted to dry environments, as observed in the study by Lichstein et al. [110], who demonstrated that in situations of water limitation, competitive carbon allocation strategies, such as increasing allocation to roots and leaves, are more effective than those that maximize biomass or productivity, reflecting an adaptive response to water stress.

Among the factors that increase the vulnerability of each species to water stress and failure in terms of survival and recruitment are limitations in soil-to-root conductivity, low tissue water retention capacity, water loss through cuticular transpiration, and susceptibility to failure in the xylem conduction system. Additionally, limitations in carbon assimilation, especially under stomatal closure conditions, restrict the production of energy needed for cellular repair and maintenance, further exacerbating the risk of mortality [114]. This process may lead to a reduction in biodiversity in an area, as plants that are better adapted to water scarcity and with efficient carbon allocation strategies tend to survive, while those with fewer available resources or less efficient strategies ultimately succumb.

In this context, characteristics such as wood density and leaf area exert a significant influence on the species’ ability to tolerate water scarcity, especially under high population density. Species with high wood density tend to have stiffer tissues, resistance to cellular cavitation, and deeper roots, although they store less water in their tissues [115]. In contrast, species with low wood density, although having a greater capacity for water storage, are generally more vulnerable to prolonged drought, especially when associated with large leaf areas and superficial roots, which increases water loss through transpiration and limits their ability to access deeper water, restricting them to surface water [116]. Thus, in a competitive scenario, species with smaller leaf areas, high wood density, and deeper roots tend to exhibit greater efficiency in controlling water loss, favoring their persistence under adverse environmental conditions [117,118].

3.7. Geostatistical Modeling and Kriging Maps

The experimental geostatistical semivariogram models were fitted based on the semivariance estimation, as can be seen in

Table 4. Semivariogram model and degree of spatial dependence (DSD) of stem recruitment patterns in the Caatinga remnant, Floresta, Pernambuco, Brazil.

Stem Recruitment	Model	1 C0	2 C0 + C	3 a	4 R2	5 C0	6 DSD	Jackknifing
						(C0 + C)		Mean	8 SD
2012	Gaussian	0.100	121.000	152	0.697	0.100	7 Sd	−0.026	1.268
2013	Spherical	9.100	37.760	514	0.609	24.100	Sd	0.000	1.295
2014	Spherical	10.100	62.060	633	0.628	16.300	Sd	−0.012	1.284
2015	Gaussian	0.010	1.738	203	0.888	0.600	Sd	−0.024	1.302
2016	Gaussian	0.010	4.837	154	0.845	0.200	Sd	−0.006	1.175
2017	Gaussian	4.920	30.830	3129	0.742	16.000	Sd	0.016	1.010
2018	Exponential	2.313	14.670	717	0.790	15.800	Sd	−0.120	1.103
2019	Gaussian	4.500	46.900	146	0.979	9.600	Sd	−0.017	1.016
2020	Gaussian	2.600	20.420	192	0.851	12.200	Sd	0.003	1.222
2021	Exponential	0.710	10.410	343	0.891	6.800	Sd	−0.014	1.210

¹ C₀: nugget effect; ² C₀ + C: sill; ³ a: range (m); ⁴ R²: semivariogram adjustment; ⁵ C₀/(C₀ + C): DSD percentage; ⁶ DSD: degree of spatial dependence (%); ⁷ Sd: strong; ⁸ SD: standard deviation.

and

Table 5. Semivariogram model and degree of spatial dependence (DSD) of tree recruitment patterns in the Caatinga remnant, Floresta, Pernambuco, Brazil.

Tree Recruitment	Model	1 C0	2 C0 + C	3 a	4 R2	5 C0	6 DSD	Jackknifing
						(C0 + C)		Mean	8 SD
2012	Spherical	0.100	33.130	399	0.664	0.300	7 Sd	0.003	1.248
2013	Exponential	0.010	4.639	727	0.619	0.200	Sd	0.000	1.064
2014	Exponential	0.550	2.429	1017	0.742	22.600	Sd	−0.001	1.141
2015	Gaussian	0.050	0.206	211	0.814	24.600	Sd	−0.012	1.293
2016	Spherical	0.101	0.970	446	0.957	10.500	Sd	−0.021	1.268
2017	Gaussian	0.001	1.987	145	0.848	0.100	Sd	0.005	1.197
2018	Gaussian	0.477	2.733	194	0.864	17.500	Sd	−0.004	1.225
2019	Gaussian	0.010	4.771	169	0.879	0.200	Sd	−0.004	0.993
2020	Gaussian	0.052	0.294	151	0.897	17.500	Sd	0.038	0.823
2021	Gaussian	0.241	1.062	232	0.739	22.800	Sd	−0.001	1.132

¹ C₀: nugget effect; ² C₀ + C: sill; ³ a: range (m); ⁴ R²: semivariogram adjustment; ⁵ C₀/(C₀ + C): DSD percentage; ⁶ DSD: degree of spatial dependence (%); ⁷ Sd: strong; ⁸ SD: standard deviation.

for stem and tree recruitment, respectively, in the Caatinga remnant. It is noted that the three experimental semivariogram models (Gaussian, exponential, and spherical) for both variables fit the dataset studied, with the predominance of the Gaussian model for stem recruitment in the years 2012, 2015, 2016, 2017, 2019, and 2020 and for tree recruitment in the years 2015, 2017, 2018, 2019, 2020, and 2021, followed by the spherical model in the years 2013 and 2014 for stems and 2012 and 2016 for trees, and finally the exponential model in the years 2018 and 2021 for stems and 2013 and 2014 for trees. Consistent with the findings of this study, Pelissari et al. [23] applied a geostatistical analysis to map and correlate spatial patterns in the basal area dynamics of successional groups of tree species in a mixed tropical forest in southern Brazil, observing that the spherical and Gaussian models generally exhibited the best fits, except for the pioneer group with the exponential function.

The coefficient of determination (R²) of the experimental semivariogram models employed for stem recruitment rates (Table 4) and tree recruitment rates (Table 5) had R² values of the fits exceeding 0.60. Based on the criterion proposed by Cambardella et al. [64], the degree of spatial dependence (<25%) reflects strong semivariogram models, which was observed for the models established for stem and tree recruitment, indicating that the characterization of stem and tree recruitment variability from one point to another exhibited strong dependence, with those values being representative among neighboring points.

According to the criterion established by Vauclin et al. [62], the semivariogram models established were validated using the jackknifing technique, where the mean errors of each model should be close to zero and the standard deviation close to one, thus validating the applicability of each model representing the recruitment of stems and trees for each of the years studied. The authors emphasize that the applicability of the established models can be replicated for other regions of the globe that exhibit characteristics of seasonally dry tropical forests. The validation explored in this study is crucial and was also explored in the study by Silva et al. [21], which emphasizes the importance of the applicability of cross-validation by jackknifing for semivariogram models used in characterizing the spatial dynamics of rainfall in the coastal region of the state of Pernambuco.

Kriging maps were developed for the spatiotemporal distribution of stem recruitment (Figure 9) and tree recruitment (Figure 10) for the farm area based on the established and validated semivariogram models, with emphasis on the experimental plots. These maps enable observing the spatial variation of recruitment from 0 to >16 stems and from 0 to >10 trees in points of 400 m². The highest stem and tree recruitment values were observed for the year 2012, as highlighted in the confidence interval analysis (Figure 8B,C), and could be associated with rainfall and SPI patterns within the standards to which the region’s species had adapted in previous years (Figure 4 and Figure 5), providing better water conditions for establishing new stems and trees.

Furthermore, there were points in Area I where recruitment exceeded 16 stems and 10 trees in the year 2012. It is worth noting that this area underwent vegetation exploitation and is undergoing regeneration, which partly explains the higher recruitment values, as post-disturbance forests in early stages tend to have higher recruitment rates [119]. Additionally, the plots located to the west represented more than 50% of the stem density and 60% of the tree density in the year 2012, as in the other years. Hence, the presence of small patches with recruitment values exceeding 12 stems and 7 trees was related to higher plant densities in the nearby experimental plots and their surroundings. Consequently, it was expected that a greater number of trees would result in higher stem recruitment, and that young trees would meet the minimum inclusion criteria (C ≥ 6 cm) in these areas.

Regarding the kriging maps for stem and tree recruitment in the other years (2013–2021), it is worth noting the occurrence of subtle variations in the spatiotemporal distribution of these variables, with a predominance of homogeneous patches with recruitment in the range of 0–5 stems and 0–3 trees (Figure 9 and Figure 10). Moderately and extremely dry SPI categories were observed during this period (Figure 5), with the average annual rainfall below 400 mm and an increase in the average annual temperature (Figure 4), which may have led to stagnation in the growth of established stems and trees in the community, preventing them from reaching the minimum inclusion criteria (C ≥ 6 cm) considered to be recruited in the study. According to Taiz et al. [96], the plant closes its stomata to reduce water loss through transpiration as a way to compensate for the decrease in water potential; however this strategy also reduces leaf assimilation and CO₂ absorption, consequently decreasing photosynthesis and suppressing plant growth.

The higher stem recruitment values compared to tree recruitment over the years can be explained by a distinctive characteristic of SDTF species, which is the presence of multiple stems as an important regeneration strategy, especially when subjected to severe disturbance [9]. Additionally, the species with higher densities and frequencies in the area are Cenostigma bracteosum (Tul.) E. Gagnon and G.P. Lewis and Aspidosperma pyrifolium Mart. [17], as these species are known for their high sprouting capacity. They contributed approximately 98% of the sprout density two years after clear-cutting vegetation, as observed in the study by Lima et al. [9] conducted on the same property where the present study was conducted.

Therefore, the presence of species with regrowth or rooting capacity in these environments is crucial for the recruitment of new individuals or stems, especially when the forest is under stressful conditions. This is because the plant can temporarily eliminate the most vulnerable ontogenetic stages (i.e., seedlings and young plants) and “restart” from stumps more vigorously, triggering the recruitment process through facilitation [9,120].

The geostatistical models evaluated in the study for stem and tree mortality based on the nugget, sill, and range effects are presented in

Table 6. Semivariogram model and degree of spatial dependence (DSD) of stem mortality patterns in the Caatinga remnant, Floresta, Pernambuco, Brazil.

Stem Mortality	Model	1 C0	2 C0 + C	3 a	4 R2	5 C0	6 DSD	Jackknifing
						C0 + C		Mean	8 SD
2012	Gaussian	38.00	236.70	1453.00	0.939	16.10	7 Sd	−0.016	1.032
2013	Gaussian	0.10	265.10	130.00	0.939	0.00	Sd	−0.008	1.115
2014	Gaussian	0.10	203.10	122.00	0.612	0.00	Sd	−0.006	1.218
2015	Spherical	2.40	58.38	196.00	0.975	4.10	Sd	0.055	1.109
2016	Gaussian	0.10	278.20	148.00	0.604	0.00	Sd	0.023	1.179
2017	Gaussian	120.00	748.90	1200.00	0.873	16.00	Sd	0.005	1.244
2018	Spherical	0.08	14.29	159.00	0.887	0.60	Sd	−0.007	1.002
2019	Gaussian	0.10	7.36	125.00	0.969	0.10	Sd	0.013	1.098
2020	Gaussian	0.01	5.93	160.00	0.741	0.20	Sd	−0.050	1.298
2021	Gaussian	0.01	24.27	159.00	0.750	0.00	Sd	−0.006	1.261

¹ C₀: nugget effect; ² C₀ + C: sill; ³ a: range (m); ⁴ R²: semivariogram adjustment; ⁵ C₀/(C₀ + C): DSD percentage; ⁶ DSD: degree of spatial dependence (%); ⁷ Sd: strong; ⁸ SD: standard deviation.

and

Table 7. Semivariogram model and degree of spatial dependence (DSD) of tree mortality patterns in the Caatinga remnant, Floresta, Pernambuco, Brazil.

Tree Mortality	Model	1 C0	2 C0 + C	3 a	4 R2	5 C0	6 DSD	Jackknifing
						C0 + C		Mean	8 SD
2012	Spherical	2.100	11.480	403.000	0.783	18.300	7 Sd	−0.009	1.267
2013	Exponential	0.020	23.430	222.000	0.733	0.100	Sd	−0.023	1.102
2014	Gaussian	2.010	29.780	157.000	0.841	6.700	Sd	−0.015	1.292
2015	Gaussian	0.620	3.508	230.000	0.940	9.100	Sd	0.006	1.292
2016	Gaussian	0.010	20.870	124.000	0.612	0.000	Sd	0.032	1.061
2017	Gaussian	6.010	28.230	166.000	0.726	21.300	Sd	−0.050	1.311
2018	Gaussian	0.001	1.930	126.000	0.651	0.100	Sd	0.009	1.173
2019	Exponential	0.001	2.681	126.000	0.997	0.000	Sd	0.031	1.034
2020	Exponential	0.100	0.901	207.000	0.697	0.100	Sd	−0.032	1.218
2021	Gaussian	0.100	2.041	141.000	0.840	0.000	Sd	−0.023	1.270

¹ C₀: nugget effect; ² C₀ + C: sill; ³ a: range (m); ⁴ R²: semivariogram adjustment; ⁵ C₀/(C₀+C): DSD percentage; ⁶ DSD: degree of spatial dependence (%); ⁷ Sd: strong; ⁸ SD: standard deviation.

, respectively. The Gaussian model showed the best fit for both variables, except for the years 2015 and 2018 for stem mortality and 2012 for tree mortality with the spherical function, and for 2013, 2019, and 2020 with the exponential model showing the best fit for tree mortality.

The degree of spatial dependence for the established semivariogram models for stem and tree mortality was strong (<25%) for all study years, indicating that the characterization of variability for stem mortality (Table 6) and tree mortality (Table 6) of one neighboring experimental plot to another was of strong dependence, with those values being representative between the neighborhoods [64]. According to the authors, higher values of spatial dependence indicate better spatial structure and consequently greater accuracy in mapping the evaluated properties using geostatistical techniques such as kriging.

In general, all generated geostatistical models had a coefficient of determination (R²) greater than 0.60, indicating a good fit to the experimental data. Additionally, according to the methodology proposed by Jack-Knifing [62], all adopted models were validated and showed a mean close to 0 (zero) and a standard deviation close to 1.0 (one) (Table 6 and Table 7), confirming the applicability of representative models for stem and tree mortality for each of the study years. Furthermore, the spatial range (a) for all analyzed variables had higher values than the distance between the experimental plots, demonstrating that such attributes have lower variability and spatial continuity, ensuring better accuracy in estimates in unsampled locations within the experimental areas [121].

In the thematic maps obtained for the spatial distribution of the variables under study (Figure 11 and Figure 12), it is possible to clearly observe the regions with the highest mortality of stems and trees per year of evaluation. According to these maps, the highest mortality of trees and stems occurred from 2013 to 2015, with mortalities exceeding 36 stems and 10 trees. According to Gunst et al. [122], the causes of forest mortality and the conditions which lead to outbreaks of widespread mortality are complex and difficult to predict. However, the period in which the highest tree and stem death rates were recorded in the area coincides with the onset of drought events in the region. Thus, these results may indicate a response of the vegetation to the reduced water availability for plants induced by drought, as the rainfall in the study area was only 224 mm in 2012 (Figure 4), and the SPI of the year was categorized as extremely dry (Figure 5). Several studies in forests in different phytogeographic domains of the world point to drought as one of the main causes of tree death [92,93,94], and indicate that such events can negatively impact forest production, and consequently populations that rely on these products for subsistence.

The effect of climate on drought-induced forest mortality is mediated by forest structure, as trees respond to climate and resource limitation in different ways, depending on their competitive environment [123]. According to Davis et al. [124], competition and moisture limitation are closely connected, meaning more trees, and therefore greater competition result in higher water consumption through transpiration and fewer moisture resources available to each tree. Supporting this statement, the present study generally observed higher mortality rates in area 2, especially in the early years of the drought period (Figure 11 and Figure 12), which has a tree and stem density approximately twice that of area 1 (Table 2). Furthermore, there is significant stem and tree mortality observed for 2012, 2014, and 2017 around the experimental plots located west of area 1 (Figure 11 and Figure 12), an area with higher tree density. Thus, tree competition is likely to become even more important in the coming decades, as climate variability impacts forest structure and primarily moisture availability [125], as observed herein.

Regarding the kriging maps for the subsequent years (2016–2021), there is noticeable attenuation or stabilization of stem and tree mortality in the study areas (Figure 11 and Figure 12). According to Lloret et al. [126], tree populations and communities can exhibit a set of mechanisms responsible for eco-physiological and demographic stabilization, which can compensate for eventual vegetation mortality. For the authors, the main mechanisms rely on species plasticity, tolerance, and phenotypic variability, which can attenuate and offset mortality when combined with species interactions, thus enhancing survival and/or future recruitment due to new beneficial conditions, improved biotic interactions, or resource release resulting from the death of some individuals. In this sense, as discussed in the present study, it is known that extreme climatic events, such as drought, can greatly reduce plant density, as observed in the target community of the study, which experienced a reduction in the initial tree density from 2118.75 to 1447.5 trees and from 5917.5 to 3665 stems over a 10-year period (2012–2021—Table 2). According to Dale et al. [127], the reduction in density leads to lower competition rates, increasing soil water availability, and therefore may promote survival after the event and especially in response to a new drought.

When overlaying the recruitment and mortality maps, it was observed that there were patches with higher mortality in certain years and simultaneously patches with higher recruitment values. Such a result can be readily associated with the formation of canopy gaps in the community resulting from the death of trees and/or stems. Once a canopy gap is created, the physical and biological processes of the forest are altered compared to the surrounding forest, providing less competition and greater resource availability for plants, such as sunlight incidence, thus facilitating recruitment of new plants and accelerating the growth of successional species, especially from the pioneer and secondary groups [128,129,130].

However, even with the high mortality of stems and trees during the study period, the reduction in density was not sufficient for the recruitment of species to reach the initial density, which can be directly associated with the negative effects of drought on vegetation, as discussed in the present study. Furthermore, although the 10-year period (2012–2021) may not have been sufficient for these species to reach the minimum criterion of C ≥ 6 cm, further studies are needed to address this question. Likewise, monitoring the recovery process of native vegetation after the long drought period recorded in the region is necessary, given that the rainfall for the area was only 332 mm in 2021, reinforcing the need for continued vegetation monitoring.

4. Conclusions

The mortality rates and recruitment of trees and stems are closely linked to the recorded drought events, highlighting the direct impact of water deficit on vegetation dynamics. The significant vegetation mortality and low recruitment values recorded during the study period are directly associated with water deficits, as evidenced by the analysis of rainfall data, the Standardized Precipitation Index (SPI), the Soil Moisture Index (SMI) and Potential Evapotranspiration (ET₀).

The application of the Gaussian semivariogram model has proven to be an effective tool in representing the spatial variability of the studied phenomena, providing valuable insights for planning, and implementing conservation measures. Identifying the cumulative effect of droughts on increasing mortality rates and reducing recruitment over the study period reinforces the vulnerability of these ecosystems to environmental changes, highlighting the importance of considering extreme weather events in ecosystem management.

These results indicate the vulnerability of native forest ecosystems to extreme climatic events, highlighting the need for adaptive management strategies, such as the planting of species adapted to prolonged drought periods and the recovery of degraded soils. Additionally, monitoring natural regeneration, promoting the protection of native species, restoring soil ecological functions, and reducing external pressures are crucial. Sustainable water resource utilization strategies, such as integrated management and the conservation of water bodies, are essential to ensure the resilience of forests. These strategies should be incorporated into adaptive forest management plans, considering future climatic conditions and vegetation dynamics, to ensure forest resilience in the face of environmental challenges.