Droughts in the Amazon: Identification, Characterization and Dynamical Mechanisms Associated

In this study, we used the Standard Precipitation Index (SPI) to identify and to characterize the dry extreme events in the Amazon region. The results showed that the drought of 1998 was the most intense (SPI average equal to −1.69) in the period from 1979 to 2014. However, some papers have characterized the years 2005 and 2010 as the two largest droughts of the century. Thus, it was also carried out a comparative study of these droughts. The results showed that the drought 1998 was more intensive and extreme than the droughts of 2005 and 2010, although droughts of 2005 and 2010 lasted longer than 1998 due to oceanic and atmospheric conditions with influencing to cause greater social and economic impacts. Furthermore, it is suggested that the impact of the 2005 and 2010 droughts is a response from the dry occurred in 1993-1994 and 1997-1998.


Introduction
Studies showed that the Brazil is vulnerable to climate change, especially climatic extremes. The most vulnerable areas include the Amazon and Northeast Brazil, as demonstrated in recent studies [1]- [6]. In latest decades, the Amazon has experienced some extreme dry events; this attracted the attention of the scientists, not only because of the impact over local populations, but also because of the dry in the Amazon may have important impacts on the global climate [7] [8] [9].
When the drought is compared to other extreme recurring events, such as floods, it is recognized as a phenomenon that has equally adverse impacts, although they can be effectively mitigated. The occurrence of drought and its consequences may require some time to be perceived by the socio-economic systems, because they tend to evolve slowly. In addition, the extreme dry events in the Amazon may be associated with the following factors: 1) the occurrence of intense El Niño events; 2) anomalous warming of sea surface temperatures in the tropical North Atlantic during the austral winter-spring; or 3) a combination of these two effects [9]- [14].
One way to improve the studies on extreme dry events developed in this region would be through the application of techniques able to characterize and to define spatial and temporal potential patterns of these events. [15] proposed the Standard Precipitation Index (SPI) to quantify the precipitation deficit on different time scales. The versatility of SPI is in the simplicity of its calculation, furthermore, it identifies the dry events on several time scales (3, 6, 12 months, and so on). This helps to monitor the temporal dynamics of these events, i.e., the development and decline. Since it is a standardized index, the classification of the categories dry/rain can be used in any region of the world [16]. In addition, the SPI has an advantage over the other indices, e.g., Palmer Drought Severity Index, because it uses only monthly precipitation data, in other words, it is not necessary to use other meteorological variables to detect dry and rainy events [17]. Created in the United States of America, the SPI has been frequently used by researchers worldwide.
In this context, the present study has the following objectives: 1) identification and classification of the most intense drought events in the period from 1979 to 2014 using the SPI; 2) analyzing the temporal evolution of these dry events; 3) diagnosing the meteorological characteristics associated with them; 4) investigating the trends of dry and rainy events obtained by SPI in the study period.

Study Area
The study area is the Amazon region (5˚N to 15˚S and 74˚W to 44˚W) ( Figure  1). It is rich in biodiversity and fresh water, being important for the whole world, due its capacity to capture and hold carbon from the atmosphere. Moreover, the Amazon exerts a major role in South America climate by its effect on the hydrologic cycle [18].

Precipitation Data Base
In this research will be used four sets of precipitation data base. Two of these databases are based on precipitations interpolated values from surface observations provided by Global Precipitation Climatology Center (GPCC) [19] and Climate Prediction Center (CPC) [20]. The other database is based on observed precipitations that are combined with precipitation estimates generated from satellite data found in the Global Precipitation Climatology Project (GPCP) [21]. The last set of data consists of reanalysis of the Era-Interim [22].
The purpose of using these data sets is to analyze the performance of GPCC data, once it consists of the largest continuum precipitation data series (1901 to present). In addition, the GPCC data is considered as a reference point in this study. Furthermore, is important to emphasize that all sets of rainfall data were interpolated to a spatial resolution of 2.5˚ × 2.5˚ grid and adopted the period 1979-2014 as a common period between databases.

Atmospheric and Oceanic Data
In order to investigate the average and anomalous conditions of the atmosphere and oceans in extreme dry events, it was also used global atmospheric and ocean monthly data from reanalysis. The variables analysis was the zonal and meridional wind components, the vertical velocity and Sea Surface Temperature (SST). The atmospheric data were provided by NCEP/DOE AMIP-II Reanalysis (Reanalysis-2) [23], this dataset are available on global 2.5 × 2.5 grids and cover the period of 1979 to the present. The ocean data were provided by NOAA/ OAR/ESRL [24], this dataset is available on global 2 × 2 grids and cover the period of 1981 to the present.

Standard Precipitation Index (SPI)
The SPI calculation starts by determining a probability density function that de-scribes the temporal series. The Gamma Distribution (Equation (1)) it has good fit for continuous variables that have lower boundary of zero and no upper limit, for this reason it is widely used for the study of historical precipitation series [25].
To estimate the parameters α and β (Equations (1.1) and (1.2)) of the gamma distribution we used the method of maximum likelihood [26]: Thus the cumulative distribution is then transformed into normal probability distribution with mean zero and standard deviation one. Next, the cumulative probability of each monthly amount is estimated. Applies to this probability, the inverse normal function is to find the value of the SPI (Equation (1.3)). More information and/or consideration of this calculation can be found at [15] [16] [27] [28], among others. The SPI categories are illustrated in Table 1; however, to simplify the interpretation of our results there was the adaptation of the classes as suggested by [15].
Not only droughts have been characterized, but also rainy seasons. One of the great advantages of this method is to standardize the analysis, which can be used to compare completely different regions, for example, regions with more humid climates with more arid ones.

Mann-Kendall Test and Sen Method
The non-parametric Mann-Kendall test shows that the presence of a monotonic tendency increasing or decreasing of the variable studied, then estimate the slope  [31]. In order to investigate the trends of dry and rain events at all time scales, it will be applied the non-parametric statistical Mann-Kendall test. [32] affirms that the Mann-Kendall test is the most appropriate method to analyze climatic changes in climatological series and also allows the detection and approximate location of the starting point of a given trend.
The test consists of the sum t n of the number of terms m i of the series, relative to the value X i whose previous terms (j < i) are less than the same (X j < X i ), thus: According to [31], for series with a large number of terms (N) and absence of trends (null hypothesis, Ho), t n will present normal distribution with mean [E(t n )] and variance [Var(t n )]. The test is calculated by: If the value of u(t) is less than −1.96 there is a significant trend at the 95% probability level in reducing the observed series values. When u(t) is greater than 1.96, there are significant upward trends in the series.
However, the non-parametric method of Sen [33] [34] uses a linear model to estimate slope trend and residue variance being constant over time. Missing values are allowed and the data need not obey any specific distribution. In addition, the Sen method is not greatly affected by single die errors. It carried out the estimate of the true slope of the trend, assuming that it is linear, existing in a time series, making it possible to find the magnitude of this trend. This means that: where Q is the slope and B is a constant.
To obtain the slope estimate Q, we first calculate the slopes of all pairs and data.
where j > k. If To obtain an estimate of B in Equation (3)

Case Study
After the classification of the dry and rainy events through SPI, were selected the year that presented the lowest mean value in order to characterize the atmos-   Table 2 and Table 3, the highest and lowest SPI values are associated with ENSO or Atlantic Gradient events. In addition, it is observed that in the investigated period that the dry events were more intense than the rainy. This is associated with the positive phase of the Pacific Decadal Oscillation (PDO), since in this period  there was more frequent and intense El Niño (1982-83, 1986-87, 1990-94, 1997-98).

Temporal Analysis of SPI's over the Amazon
Considering the context of the tropical climate dynamics, it is known that the Zone (ITCZ) [36]- [42], therefore modulating the regional distribution of rainfall in the Amazon and consequently dry and rainy events.
Thus, the Amazon presented increasing trends of the dry and rainy events of short (SPI-3), mean (SPI-6) and long (SPI-12) duration in GPCC and GPCP databases. On the other hand, Era-Interim and CPC differed from the other databases because they presented decreasing trends of dry and rainy events of short and medium duration in the Era-interim and long duration in the CPC.
According to the analysis of Figure 2, it is observed that the GPCC satisfacto- and 2010 were characterized as the two major droughts of the century [7] [8] [12]. Moreover, the oceanic and atmospheric conditions in large-scale associated with these dry events, the horizontal distribution of SST anomalies, the anomalous zonal and meridional circulation through vertical straight sections were also studied. These will be discussed below.

General Aspects of Dry Events Observed 1998, 2005 and 2010 Characterized by the SPI-3, 6 and 12
Positive      In terms of evolution of the spatial pattern of these events, it was noted that the drought of 1998 was the most intense in much of the Amazon, but over the months this intensity decreased along with its area of coverage. On the other hand, the droughts of 2005 and 2010 were more persistent than the one of 1998, due to the oceanic and atmospheric conditions that favored the persistence of the same ones, generating greater socioeconomic impacts as highlighted by [43] and [44].
In addition, it is suggested that the impact of the droughts of 2005 and 2010 be a response to the droughts of 1993-1994 and 1997-1998, despite normal rainfall in subsequent years, drought conditions in the 1990s were so severe that the floret was unable to fully recover. In agreement with this analysis, [44] emphasize that the extreme drought events continue at a time interval of 5 to 10 years, large areas of the Amazon will be lost florets given that the effects of drought are persistent and forest recovery is very slow.
Moreover, although these three years have been characterized as dry in the Amazon, it was noted that some regions showed moderate rainfall events (1.49 > SPI > 0.5), emphasizing the versatility of SPI to identify spatially and temporally dry or rainy events in their different time.

Conclusions
In the last decades, extreme rain and dry events have occurred in the Amazon Region and these events have been increasingly studied due to the impacts they cause in socioeconomic and environmental activities. In this regard, numerous studies have been made to quantify and characterize such events, and propose mitigation and adaptation strategies. Thus, in this study, they were characterized dry and rainy events at different time scales in the Amazon in the current climate (1979-2014).
The temporal series of SPI-3, SPI-6 and SPI-12 in the GPCP, GPCC, CPC and Era-interim, on the Amazon Region presented a good temporal concordance between them in the period from 1979 to 2014. In the three SPI scales, the years 1983, 1984, 1991, 1992, 1993, 1997, 1998 and 2005 were characterized as dry. On the other hand, the years 1986, 1994, 2000, 2007 and 2009 were characterized as rainy. It was also found that dry events were more intense than the rainy during the study period. In addition, significant upward trends were observed for the dry and rainy events of short (SPI-3), mean (SPI-6) and long (SPI-12) duration which were observed in GPCC and GPCP databases. However, Era-Interim and CPC differed from the other databases because they presented decreasing trends in dry and rainy events of short and medium duration in the Era-interim and long duration in CPC.
In The results found may provide subsidies to operational centers through knowledge of the temporal dynamics of these events, allowing preventive measures to be taken to minimize impacts caused by severe weather phenomena.