Assessing Drought Vulnerability in the Brazilian Atlantic Forest Using High Frequency Data

1. Introduction

The Brazilian Atlantic Forest, once sprawling across 150 million hectares and dominating the country’s coastal regions, has undergone significant transformations over the centuries. Originally contiguous, the forest has evolved into fragmented patches due to historical land use changes and urban expansion, especially along the Brazilian coast[36].This fragmentation has placed a substantial portion of the forest’s diverse biodiversity at risk of extinction, particularly in smaller fragments of less than 50 hectares [37,38,39] as fragmented forests often experience edge effects driven by abiotic factors (e.g. water, wind, and temperature) [26], pushing plant communities toward early successional stages, which invariably leads to a reduction in the forest’s overall biodiversity [40,41,42,43]. Besides the historical land use changes, driven by urban expansion, the Brazilian Atlantic Forest faces escalating threats due to abiotic factors, with climate change posing an additional challenge, as the biome’s intricate web of life and ecological significance is confronted with shifting aridity patterns [24,25].

Many researchers have extensively studied aridity patterns using air temperature and precipitation data. For instance, a study conducted in Iran investigated time series variations in air temperature indices, De Martonne aridity index (IDM), and total precipitation using long-term meteorological data from 1960 to 2019. The study revealed that the climate in Iran has generally become warmer and drier over the past 60 years [44]. Similarly, a comparative study in Mongolia explored spatiotemporal variations of aridity using four different equations: de Martonne aridity index (IDM), Thornthwaite aridity index (AI), moisture coefficient by V. I. Mezentsev (MI), and Hydrothermal coefficient by Selyaninov (HTC) [45]. The study found that the total area of drylands determined by IDM, MI, AI, and HTC is approximately 64.1%, 70.7%, 85%, and 98%, respectively. Time series analysis of AI and MI both showed a decreasing tendency for the period of 1961–2015.

The response of forests to shifts in aridity patterns has also been the subject of extensive research. A study carried out in the Forest Reserve of Fina, Mali, it used remotely sensed data to reveal significant vegetation response to fluctuations in precipitation and air temperature. There has been observed a consistent rise in NDVI levels occurring between 15 to 30 days after rainfall following a period of drought [23]. Another study underscored the intricate interplay between precipitation, air temperature, and forests. It demonstrates, for example, how the deep roots of tropical trees can tap into water reserves, thereby sustaining their cooling role within the ecosystem, even in regions experiencing lower rainfall [22]. These studies provide an understanding of how changes in precipitation and air temperature patterns affect forests, although the specific impacts can vary depending on the geographical location, type of forest, and other environmental factors. Long-term forecasts signal a decline in precipitation and an increase in air temperature levels [21]. Such changes could significantly impact the balance of the Brazilian Atlantic Forest [27,28]. Aridity indices, commonly used to assess the dryness of a region, are typically derived from daily [12] and/or aggregated monthly data [11,13]. These statistics, due to their longer time scale, tend not to exhibit significant changes over short periods [14,15]. However, this research overlooks more immediate fluctuations in aridity that can have substantial impacts on terrestrial ecosystems as we offer a perspective on the aridity status of the Atlantic Forest by utilizing hourly data. This higher resolution data provides a more sensitive measure of short-term changes, offering a more nuanced understanding of the shifts in aridity. To do that, we use meteorological indicators derived from high resolution data, to explore the emerging trends and identify specific regions experiencing shifts in aridity conditions. Then, we established a vulnerability ranking to pinpoint regions undergoing immediate alterations in aridity conditions.

Subsequent sections provide a description of the methodology, data, and outcomes, shedding light on the complex relationship between climate variability and the status of the Brazilian Atlantic Forest.

2. Materials and Methods

We adopted the boundaries of the Brazilian Atlantic forest to be congruent with those delineated for the Atlantic forest biome by the Instituto Brasileiro de Geografia e Estatística (IBGE). The definition of the Atlantic forest biome by IBGE was based on an extensive process involving literature review, interinstitutional validation, and field surveys to verify the presence of the physical-biotic environment and historical evidence of the Atlantic forest within the demarcated boundaries [29]. These boundaries were sourced from TerraBrasilis, a platform managed by the Instituto Nacional de Pesquisas Espaciais (INPE), which facilitates data organization, accessibility, and utilization for Brazil’s environmental monitoring efforts.

To assess the exposure of plant species to extreme drought events in the Brazilian Atlantic forest, we relied on meteorological indicators. These indicators were derived from hourly data of precipitation, maximum and minimum air temperature recorded at 205 automatic weather stations, accessible from the Instituto Nacional de Meteorologia (INMET) database [https://bdmep.inmet.gov.br, accessed in August 2023]. An overview of these datasets, including station codes, sample sizes, monitoring periods, and the percentage of missing data, is provided in Table 1. The geographic distribution of the 205 automatic weather stations as well as the boundaries for the Atlantic Forest used in this study are presented in Figure 1. This dataset stands as an extensive and robust compilation, with a total of 10,299,236 data entries. Spanning a broad temporal window, from distinct station-specific start dates to a standardized endpoint on December 31, 2022, the dataset exhibits a relatively modest percentage of missing data.

Past, current, and future drought conditions were quantified using several meteorological indicators, including the number of consecutive hours without precipitation, the number of dry and wet days per month, the Aridity Index, and the Mann-Kendall test statistic (Z) for trend analysis in evapotranspiration (ET) and maximum air temperature ($T_{max}$). The choice of these indicators was motivated by their relevance to recent severe drought events [34].

We opted against employing imputation methods to replace missing data, as this approach can introduce systematic bias into the meteorological indicators [30,31]. While excluding missing data rather than employing imputation methods can have implications for the results of the analysis by introducing systematic bias from imputation methods, we exclude missing data entirely from the analysis to maintain data integrity, transparency, mitigate the risk of bias and because of the importance of spatial density. We acknowledge that this approach reduces the sample size and potentially limits the scope of analysis. However, by including all available stations, it ensures that crucial spatial coverage isn’t sacrificed as each meteorological station provides valuable data points that contribute to capturing the intricate local variations in weather patterns. This is particularly significant considering the need to investigate edge effects due to droughts in the Brazilian Atlantic Forest [10]. Thus, all missing data were excluded from the analysis, and the accumulated or averaged values resulting from any missing hourly data from the 205 automatic weather stations were disregarded.

The Aridity Index was computed as the ratio of monthly precipitation to evapotranspiration. Monthly precipitation values were determined by aggregating the available hourly precipitation data, while monthly evapotranspiration values were calculated by aggregating daily evapotranspiration estimates for a reference crop. These daily estimates were generated using the Hargreaves and Samani equation [32], which considers measured hourly maximum and minimum air temperatures, elevation, and latitude to compute the reference crop evapotranspiration rates under standard meteorological conditions, as follows:

$$E{T}_0=0.0023\times (T_{max}+T_{min})\times \sqrt{T_{max}-T_{min}}\times (T_{avg}+17.8)$$

(1)

, in which $E{T}_0$ represents the reference crop evapotranspiration (ET) in millimeters per day, $T_{max}$ is the daily maximum temperature in Celsius, $T_{min}$ is the daily minimum temperature in Celsius and $T_{avg}$ is the daily average temperature in Celsius. We considered this approach for estimating evapotranspiration (ET) appropriate for drought analysis in the Brazilian Atlantic forest, where the ecosystem is known to be especially vulnerable to droughts along its edges due to the prevalence of C3 and C4 grass-like vegetation, which shares functional characteristics with Samani’s reference crop. Moreover, the Hargreaves and Samani equation is particularly advantageous as it incorporates ET estimation based on both maximum and minimum hourly temperature data. This approach enables the capture of immediate fluctuations in aridity.

The choice of the Aridity Index for drought analysis instead of other standardized indices was deliberate, considering that standardized indices primarily focus on analyzing the cumulative probability of annual and monthly data. Typically, these datasets follow a gamma distribution [9], which may not be suitable for hourly datasets, such as the one utilized here, due to their distinct probability distributions.

The classification of wet and dry days per month was performed using a precipitation threshold of 0 mm per weather station. Days with precisely 0 mm of precipitation were categorized as dry, while those with precipitation exceeding 0 mm were classified as wet. This threshold is very close to the one established by [20], who assumed a threshold of 1 mm for dry days. The rationale behind our choice stems from the unique climatic characteristics of the Brazilian Atlantic Forest, which extends along Brazil’s Atlantic coastline, encompassing regions with varying levels of aridity. By setting a threshold of 0 mm, we aimed to capture the slightest presence or absence of rainfall, recognizing that even minimal precipitation, or its absence, can yield significant ecological implications. Subsequently, the median values for the number of wet and dry days per month were employed as meteorological indicators to gauge drought exposure.

To determine the duration of consecutive hours without precipitation, we analyzed sequences of hours with no recorded precipitation. In instances of missing data within a sequence, we adopted a conservative approach by considering two distinct sequences. This approach was necessary as missing data could inadvertently elongate the periods to be without precipitation.

Mann-Kendall (MK) trend analysis was employed to examine the daily evapotranspiration estimates and maximum air temperature data at each weather station. A trend was considered statistically significant when the associated p-value was less than 0.05. The MK test statistics, Z and S, were used to assess both the presence and the strength of a monotonic upward or downward trend in daily evapotranspiration and maximum air temperature. The MK test statistic Z was further utilized as an indicator for evaluating potential future drought exposure.

Lastly, we conducted a hazard assessment by ranking the meteorological indicators calculated for each weather station, which allowed us to identify the regions most exposed to drought events. This hazard assessment was developed, taking into consideration environmental factors that influence tree growth dynamics in the Brazilian Atlantic forest under drought conditions [33], as well as the amount of data available after the data filtering process for missing data.

3. Results and discussion

We used boxplots to visualize the distribution of accumulated monthly precipitation for all 205 automatic weather stations in the Brazilian Atlantic forest. They provide range, median, and distribution of values at each weather station during their monitoring period. The boxplots are presented in Figure 2. Figure 2 reveals the presence of positive outliers across most of the weather stations, which indicates instances of extreme rainfall events. These events are characterized by precipitation values significantly higher than the majority of the data points. The consistent occurrence of positive outliers across various weather stations implies that extreme rainfall events are a recurring phenomenon in the Brazilian Atlantic forest.

The Mann-Kendall (MK) test assessed the presence and strength of monotonic trends in the time series data. We assumed that trends were statistically significant when the p-value was less than 0.05. The results of the MK trend analysis are summarized in Figure 3. A substantial majority of the 205 weather stations exhibited a statistically significant upward trend in daily maximum temperature and number of hours without precipitation. In contrast, the majority of daily precipitation exhibited a statistically significant downward trend. Unlike the unidirectional trend observed in daily maximum temperature or daily precipitation, daily minimum temperature and daily evapotranspiration presented a spectrum of behaviors, including instances of no discernible, upward and downward trends. Despite the varied outcomes in the analysis of daily minimum temperature and daily evapotranspiration, the overarching findings of the MK test indicated an overall reduction in daily precipitation and an increase in both maximum daily temperature and number of consecutive hours without precipitation across the Brazilian Atlantic forest. This suggests a scenario in which the ecosystem may encounter drier conditions in the future. It is important to note that while metorological indicators derived from annual and monthly values of precipitation or evapotranspiration provide valuable insights [16,17,18,19], they may not sufficiently capture short-term aridity extremes. This examination of hourly data for maximum temperature and consecutive hours without precipitation reveals that a significant portion of the Brazilian Atlantic Forest is already experiencing dry conditions.

In an attempt of establishing a vulnerability ranking, the outcomes of the meteorological indicators were synthesized and organized in Table 2. The ranking reflects the ascending order of vulnerability, delineating the weather stations in terms of their susceptibility to drought events within the Brazilian Atlantic Forest. There is no significant correlation between the assigned rank numbers and the latitudinal positions and elevation of the weather stations (Figure 4. This observation suggests that neither geographical location, as represented by latitude, nor elevation play a significant role in influencing the trends of dryness observed at each weather station. While the elevation can exert an overall significant influence on the characteristics and impacts of droughts. It is imperative to contextualize this within the specific geographical and climatic conditions of the Brazilian Atlantic Forest. The range of elevation within this biome is not as pronounced as those observed in regions with more mountainous topography. Furthermore, the data utilized in our study was sourced from meteorological stations that may not have been optimally positioned to accurately record the effects of elevation on drought trends. These stations are predominantly located at altitudes less than 1000 meters above sea level, which may limit the scope of elevation-related data captured.

4. Conclusions

The findings of this study provide significant insights into the dynamics of drought in the Brazilian Atlantic forest. The consistent occurrence of extreme monthly precipitation values across various weather stations suggests that they are a recurring phenomenon in the Brazilian Atlantic forest. However, in spite of the regular incidence of extreme monthly precipitation, the findings from the Mann-Kendall (MK) test demonstrate a prevailing decrease in daily precipitation alongside an elevation in both maximum daily temperature and the duration of consecutive hours without precipitation throughout the Brazilian Atlantic Forest. Additionally, the vulnerability ranking synthesized from the outcomes of the meteorological indicators does not show a significant correlation with the latitudinal positions and elevation of the weather stations, which implies that geographical location and elevation do not significantly influence the trends of dryness observed at each weather station. This observed trend hints at a prospective scenario wherein the ecosystem is likely to face drier conditions, all the while maintaining a consistent range in monthly precipitation values. It aligns with the research conducted by [35], which also highlighted the vulnerabilities of the Brazilian Atlantic forest to climate change.

Additionally, we underscored the importance of recognizing short-term aridity extremes, which are often overlooked when relying solely on monthly precipitation, evapotranspiration, or Aridity Index (AI) values. By analyzing hourly data, including maximum temperature and consecutive hours without precipitation, we discovered that a significant portion of the Brazilian Atlantic Forest is already experiencing dry conditions. Identifying these short-term extremes is crucial as they can serve as an early warning system for potential long-term drought conditions, enabling farmers, researchers, and policymakers to prepare for potential impacts, such as crop failure or water shortages [8]. They also provide insights into plant survival strategies during short-term drought, such as physiological or behavioral changes like stomatal closure to reduce water loss [7]. Short-term extremes can be used to calibrate and validate drought impact models, providing a range of conditions against which these models can be tested [5,6]; for the development of mitigation strategies, such as irrigation plans to protect crops known to be susceptible to damage under short-term drought conditions [4]; and to have significant ecological impacts, affecting plant-pollinator interactions, soil moisture levels, and even increasing susceptibility to pests and diseases [1,2,3].