**1. Introduction**

The Seasonally Dry Tropical Forests (SDTF) are characterised by a strong seasonal and irregular distribution of rainfall throughout the year, resulting in long dry season periods [1,2]. The Brazilian Caatinga, with an area of approximately 900,000 km2, is the largest nucleus of SDTF in the Neotropics [The tropical New World biogeographic region comprises Central America, the Caribbean, and South America] [3]. Moreover, with 3347 plant species, of which 526 are endemic, the Caatinga is the richest nucleus of SDTF in the Neotropics [4]. Its remarkable floristic diversity makes the Caatinga twice as rich as the Amazon rainforest when considering the species/area relationship [4]. Phenology studies recurring life-cycle events such as bird migration, flower blooming, or leaf emergence and senescence and the causes of their timing by biotic and abiotic forces [5,6]. The leafing patterns of Caatinga vegetation are adapted to the intense climate and water seasonality, being highly dependent on the climate interannual variability [7,8]. Most leaves fall during the dry season, and the first rainfall pulses trigger a quick leaf flush in the wet season [7–9].

Precipitation and soil moisture are the leading environmental drivers for the leaf changes in the Caatinga [8,9]. Still, the temperature can also be a driver for species presenting scheduled phenology in dry ecosystems [10]. The Intergovernmental Panel on Climate Change [11] forecasts an increase of 1.5 ◦C in the global mean air temperature for the next two decades in an optimistic scenario, considering the reduction of current emissions of CO2. Besides the temperature increases, climate changes are likely to alter the precipitation regimes in the following decades [12,13]. The consequences of these changes are, among others, the projected drying out of surface soils [12] and prolonged dry seasons, with an increase of 47% of the area of the Northeast Region of Brazil (NEB) subjected to extreme drought events until 2070 [13]. Thus, understanding the past and current vegetation's response to the environmental drivers is paramount to predicting its behaviour in climate change scenarios, allowing the detection of changes in the timing of leaf patterns and their causes.

The ability to monitor global vegetation phenology, or Land Surface Phenology (LSP), has increased with the validated Remote Sensing (RS) and modelling approaches to mapping phenology [14,15]. Long-term data from satellite products are useful tools for understanding the phenological responses of vegetation to current environmental drivers using Vegetation Indices (VIs), allowing it to predict its responses to climate change scenarios. VIs time series has received the attention given its potential to characterise interactions between climate and vegetation with broad applications in different ecosystems [16–18]. Several VIs are calculated based upon different spectral bands and, therefore, evidence of different components of the environment [19]. The Enhanced Vegetation Index (EVI) has been widely used to characterise vegetation phenology [20,21] due to its sensitivity to high biomass and reduced atmospheric and soil effects. EVI is calculated from the near-infrared (NIR), Red, and Blue bands and can be derived for different satellite platforms, such as Landsat, Sentinel, and MODIS. The use of algorithms to determine the main phenological metrics from the VIs time series has favoured the representation of the phenological stages of each cropping system, allowing a crop-type classification based on their phenological metrics [22]. However, the studies driving this on a global scale have been primarily focused on forest ecosystems, associating phenological changes in vegetation with climate patterns, particularly with rainfall data [23]. These studies showed that the phenological dynamics strongly depend on the seasonality of rainfall [21,24]. Still, the studies on a regional scale indicate that other environmental drivers also trigger phenological changes [25–28].

The LSP applied to an ecosystem scale seems to offer the best opportunities to advance understanding of environmental triggers and determinants for phenological dynamics, given the possibility to understand it on a broad scale, encompassing areas in a range of contrasting environmental conditions. For instance, the early greening or pre-rain green-up, a phenomenon where trees produce leaves before the rain starts, was registered in the woodlands and savannas of southern Africa through RS satellite techniques [29,30]. Furthermore, the application of the LSP at a continental scale and using long-term time-series

(2002–2014) allowed us to measure the variability in leaf flushing (i.e., greening) among years and to identify the photoperiod as the environmental cue for early greening [30].

The use of LSP and their drivers will be significant for SDTFs where interannual rainfall variability and rainy season duration change on a spatial and temporal scale [31,32], factors that are expected to influence the phenological strategies of plant communities in this vegetation [10]. There was also grea<sup>t</sup> regional variability and interannual fluctuation in vegetation phenology, and the overall phenological trends shifted later [21]. By following the studies for dry forests, Tong et al. [15] reported that the interannual rainfall variability was a more dominant force than fire events and land-use change in the phenological trend in tropical areas. For example, Jesus et al. [25] noticed changes in the phenological patterns for dense and open vegetation areas of the Caatinga, suggesting that factors that vary at spatial scales, such as the vegetation structure, would also be necessary for the phenological responses of the vegetation. In an experimental area in the Caatinga, the phenological response was directly related to soil water availability [9].

Despite advances in the analysis of phenological patterns and their associated environmental drivers, mainly observed in studies at the ecosystem scale, the application of long-term time series of vegetation indices in studies of the Caatinga vegetation is scarce. In addition, there are limitations to the diversity of sites studied for the Caatinga. For example, when analysing the dominant environmental drivers for the phenology of seasonally dry ecosystems (Caatinga, Cerrado), Alberton et al. [33] observed that the dominant drivers in these ecosystems were distinct, with light (measured as day-length) being more relevant in explaining leafing patterns in Cerrado communities than rainfall for Caatinga communities. Therefore, comparing sites of the same ecosystem can better define the environmental drivers associated with the phenological dynamics. There are also limitations to the number of environmental drivers analysed in the studies carried out for the Caatinga. Analysis with more environmental drivers could reveal meaningful soil–plant–atmosphere interactions, which may occur to a lesser extent.

Given the above, this study proposes to evaluate the Caatinga phenological sensitivity to environmental drivers in three Caatinga ecoregions. The seasonality of vegetation will be observed from the EVI time series over 20 years (2000–2019) and environmental drivers (precipitation, air temperature, soil moisture, and water deficit) from global databases. This study has two objectives: (i) estimate phenological parameters using an EVI time-series over 20 years, and (ii) characterise the relationship between phenologic dynamics and environmental drivers. The results will also be expected to serve as a baseline against which to compare future changes in Caatinga phenology due to natural or anthropogenic causes.

#### **2. Material and Methods**
