Mechanisms for Severe Drought Occurrence in the Balsas River Basin (Mexico)

This work provides an assessment of the two most intense seasonal droughts that occurred over the Balsas River Basin (BRB) in the period 1980–2017. The detection of the drought events was performed using the 6 month scale standardized precipitation–evapotranspiration index (SPEI-6) and the 6 month standardized precipitation index (SPI-6) in October. Both indices were quite similar during the studied period, highlighting the larger contribution of precipitation deficits vs. temperature excess to the drought occurrence in the basin. The origin of the atmospheric water arriving to the BRB (1 May 1980–31 October 2017) was investigated by using a Lagrangian diagnosis method. The BRB receives moisture from the Caribbean Sea and the rest of the tropical Atlantic, the Gulf of Mexico, the eastern north Pacific and from three terrestrial evaporative sources: the region north of BRB, the south of BRB and the BRB itself. The terrestrial evaporative source of the BRB itself is by far the main moisture source. The two most intense drought events that occurred in the studied period were selected for further analysis. During the severe drought of 2005, the summertime sea surface temperature (SST) soared over the Caribbean Sea, extending eastward into a large swathe of tropical North Atlantic, which was accompanied by the record to date of hurricane activity. This heating generated a Rossby wave response with westward propagating anticyclonic/cyclonic gyres in the upper/lower troposphere. A cyclonic low-level circulation developed over the Gulf of Mexico and prevented the moisture from arriving to the BRB, with a consequent deficit in precipitation. Additionally, subsidence also prevented convection in most of the months of this drought period. During the extreme drought event of 1982, the Inter Tropical Convergence Zone (ITCZ) remained southern and stronger than the climatological mean over the eastern tropical Pacific, producing an intense regional Hadley circulation. The descent branch of this cell inhibited the development of convection over the BRB, although the moisture sources increased their contributions; however, these were bounded to the lower levels by a strong trade wind inversion.


Introduction
Numerous severe drought events have been witnessed around the world during the last decades [1], such as the 2007-2009 drought in the Fertile Crescent [2], the 2010 Amazon drought [3], the long drought of the first decade of the current century in Australia [4] or the 2011 record-setting Texas drought [5]. In the case of Mexico, the frequency and extension of (Figure 1b) [15], but due to the orography, the basin has a variety of climatic characteristics [16]. The BRB has an extension of 116,436 km 2 [17], and it is the most important Pacific slope river in the country, supplying water to big conurbation areas such as Puebla, Tlaxcala and Cuernavaca-Cuautla, among others. Additionally, the hydroelectric installed capacity in Mexico, which contributes 48% of the annual clean energy generation, is mainly concentrated in the Balsas, Lerma Santiago and Grijalva basins, with the former being the most vulnerable to suffer a lack of water availability [18].

Drought Indices
The standardized precipitation index (SPI) relies on the difference of precipitation from the mean for a specified time period divided by the standard deviation [28]. This method takes into account the fact that the time period from the arrival of precipitation until water is available in each useable form (soil moisture, ground water, snowpack, streamflow and reservoir storage) differs greatly [28]. The SPI can be calculated over different precipitation accumulation periods, and the averaging period (1,3,6,12,24 or 48 months) is selected depending on the type of drought in which user is interested (i.e., meteorological, agricultural, hydrological or ecological). A disadvantage of this simple method is that precipitation is typically not normally distributed for accumulation periods of 12 months or less, but this is overcome by applying a transformation to the distribution [28,29]. Many studies using the SPI have been undertaken, as the SPI provides information on precipitation deficit, average percentage and probability. However, in recent years, some shortcomings in the index have become apparent due to the fact that it does not take into account variables such as evapotranspiration or temperature [30,31], which can be crucial in order to assess the occurrence of droughts.
The standardized precipitation-evapotranspiration index (SPEI) takes into account the potential evapotranspiration (PET), besides the precipitation, in order to determine the occurrence of droughts [32]. The SPEI is able to capture the impact of temperature on the atmospheric evaporative demand of water [33]. The method used to compute the SPEI is similar to that for the SPI. However, while the SPI adjusts precipitation to a certain distribution, in the case of the SPEI, it is the difference between precipitation and PET that is adjusted. The SPEI can also be calculated over different time-scales for the estimation of the diverse potential impacts of water deficit accumulation occurring over time periods of different lengths.
The response of the hydrological system to precipitation is time-lagged and not known a priori [34], but in general the SPEI and SPI average accumulation periods (e.g., 3 to 12 months) can be used as an indicator for reduced stream flow and reservoir storage [35,36]. We computed 6 month scale indices in October (SPI-6 and SPEI-6) to analyze the seasonal scale droughts. The 6 month indices at the end of October provide a very good indication of the amount of precipitation that has fallen during the wet season for the BRB. In this way, we can even take into account the impact in the early stages on water reservoirs. In this work, 1 month indices (SPEI-1 and SPI-1) were also used to determine the months within the drought period with the most severe meteorological drought. Monthly rainfall data from the University of East Anglia Climate Research Unit (CRU) database [15] were used to generate the SPEI and SPI for a 38 year period (1980-2017) with a 0.5 • resolution in longitude and latitude. The PET from the CRU database, which employs the FAO (Food and Agricultural Organization) Penman-Monteith formula [15], was also used to compute SPEI.

E-P Backward Tracking
FLEXPART v9 (flexible particle dispersion model) was used here to identify the largescale water vapor advection toward the region of study. Although the model was originally created to determine pollutant dispersion, it was adapted also to diagnose any change of different variables along the trajectories. Over the last decades, the model has largely been used to determinate moisture transport that generates precipitation over continents based on the methodology by Stohl and James [37,38]. FLEXPART is supported by a large number of peer-reviewed publications [39] and has been validated for the atmospheric branch of the water cycle of major river catchments [38]. The model has also been used to study water vapor transport variability toward several hydrologic basins, such as the Amazon [40], Niger River basin [41] or the Danube river basin [42], among others.
FLEXPART generates the trajectories of a large number of particles (2 million) into which the atmosphere is divided; each particle moves by using a three-dimensional wind field with a constant mass. FLEXPART uses meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF) [43] every 6 h with a 1 • × 1 • resolution on 60 vertical levels to calculate the grid-scale advection. To determine moisture sources and sinks, the position of each particle is tracked backward or forward, respectively, in combination with the particle loss of specific humidity by precipitation (p) or gain by evaporation (e). The (E-P) differences of all particles in a column are amassed, and then the total (E-P) field is obtained, where (E) is the evaporation rate and (P) the precipitation rate per unit area. This method therefore diagnoses E-P, but not E or P individually [37].
In this work, the trajectories were followed in backward mode, which means that we were interested in determining the origin of the humidity over the BRB. By integrating the changes of all particles aimed towards the BRB region, we found the areas where those particles had either gained (E-P > 0) or lost moisture (E-P < 0) along their path. We limited the transport to 10 days as this is the optimal particle travel time for the BRB region during our study period using the database by Nieto and Gimeno [44]. Then, a general view of the moisture sources was attained by adding the net freshwater flux from day −1 to day −10, ((E-P) 1-10 from now on). When a long enough period is analyzed, the mean moisture sources for the region can be described from a climatological perspective.

Climatic Fields Data
The COBE (Centennial in situ Observation-Based Estimates) SST v2 monthly dataset was used from 1980 to 2017 [45,46] at 1 • latitude × 1 • longitude spatial resolution. COBE incorporates ICOADS (International Comprehensive Ocean-Atmosphere Data Set) data [47] and includes some additional observational datasets mainly for recent periods.
ERA-Interim reanalysis monthly means of horizontal wind, vertically integrated moisture flux, geopotential height and vertical velocity (omega) at several tropospheric levels [43] were used for the period 1980 to 2017 at a 1 • resolution in longitude and latitude. In addition, temperature and relative humidity were obtained from the same reanalysis database. These variables were employed to compute the equivalent potential temperature (θ E ) according to Equation (38) in [48], allowing us to assess the moisture content and thermal properties of the air column.
Seasonal anomalies were computed with respect to the climatology of May 1981-October 2010, referred to hereafter as 1981-2010 climatology.

Drought Characteristics during the Period of 1980-2017
The temporal variation of SPI-6 and SPEI-6 in October for the period 1980-2017 is depicted in Figure 2. Both indices are very similar, with the Pearson correlation coefficient between them being ρ = 0.99, and both match the same dry periods. This indicates that the contribution of precipitation deficits to the drought severity in the BRB prevails over the temperature excess, unlike other regions, such as the Mediterranean region, where the increase in the atmospheric evaporative demand driven by increasing temperatures in the last decades has enhanced droughts, independently of the precipitation evolution in the region [49,50]. In fact, the link between droughts and heat waves is widely known: high temperatures accelerate soil drying and dry soils in turn warm the atmosphere by retaining less water for evapotranspiration [51]. Studies examining the coupling between the land surface and extreme temperature have shown that the soil moisture/temperature relationship is geographically widespread [52]. This area of research has held a predominant European focus [51]. For example, it is known that the intensity of the 2003 heatwave was increased by dry conditions by up to 40% [53,54], while the record-breaking drought that affected Europe in 2016/2017 has been found to have been caused by decreased precipitation and increased sunshine duration in the northern part, whereas the main contributors to the drought in the south were thermodynamic processes, mostly associated with high temperatures [55]. Seasonal anomalies were computed with respect to the climatology of May 1981-October 2010, referred to hereafter as 1981-2010 climatology.

Drought Characteristics during the Period of 1980-2017
The temporal variation of SPI-6 and SPEI-6 in October for the period 1980-2017 is depicted in Figure 2. Both indices are very similar, with the Pearson correlation coefficient between them being = 0.99, and both match the same dry periods. This indicates that the contribution of precipitation deficits to the drought severity in the BRB prevails over the temperature excess, unlike other regions, such as the Mediterranean region, where the increase in the atmospheric evaporative demand driven by increasing temperatures in the last decades has enhanced droughts, independently of the precipitation evolution in the region [49,50]. In fact, the link between droughts and heat waves is widely known: high temperatures accelerate soil drying and dry soils in turn warm the atmosphere by retaining less water for evapotranspiration [51]. Studies examining the coupling between the land surface and extreme temperature have shown that the soil moisture/temperature relationship is geographically widespread [52]. This area of research has held a predominant European focus [51]. For example, it is known that the intensity of the 2003 heatwave was increased by dry conditions by up to 40% [53,54], while the record-breaking drought that affected Europe in 2016/2017 has been found to have been caused by decreased precipitation and increased sunshine duration in the northern part, whereas the main contributors to the drought in the south were thermodynamic processes, mostly associated with high temperatures [55]. According to [28], the intensity of a drought event is arbitrarily defined for values of the SPI with the following categories: mild drought if 0 ≤ SPI ≤ −0.99, moderate drought if −1.00 ≤ SPI ≤ −1.49, severe drought if 1.50 ≤ SPI ≤ −1.99 and extreme drought for SPI ≤ −2. Recent studies-e.g., [42,[56][57][58][59]-classified drought events with identical thresholds for SPEI and SPI. Therefore, we identified seven moderate to extreme events in the years 1982, 1986, 1987, 1994, 1997, 2005 and 2009. Table 1 summarizes the main characteristics of these drought events, including the year and average intensity, based on SPEI-6 and the number of months with meteorological drought and the driest months based on SPEI-1 (in bold letter). As mentioned before, SPI values are very similar and they are not shown. The year 1982 is identified as the year with the most intense drought episode for the study period. All episodes had meteorological droughts with a duration of roughly 4 months. June was the peak month for 2005 and July was the peak month for 1986, 1994 and 2009. The droughts of 1982 and 1997 peaked in August, and that of 1987 peaked in October. It is interesting to mention that five of these According to [28], the intensity of a drought event is arbitrarily defined for values of the SPI with the following categories: mild drought if 0 ≤ SPI ≤ −0.99, moderate drought if −1.00 ≤ SPI ≤ −1.49, severe drought if 1.50 ≤ SPI ≤ −1.99 and extreme drought for SPI ≤ −2. Recent studies-e.g., [42,[56][57][58][59]-classified drought events with identical thresholds for SPEI and SPI. Therefore, we identified seven moderate to extreme events in the years 1982, 1986, 1987, 1994, 1997, 2005 and 2009. Table 1 summarizes the main characteristics of these drought events, including the year and average intensity, based on SPEI-6 and the number of months with meteorological drought and the driest months based on SPEI-1 (in bold letter). As mentioned before, SPI values are very similar and they are not shown. The year 1982 is identified as the year with the most intense drought episode for the study period. All episodes had meteorological droughts with a duration of roughly 4 months. June was the peak month for 2005 and July was the peak month for 1986, 1994 and 2009. The droughts of 1982 and 1997 peaked in August, and that of 1987 peaked in October. It is interesting to mention that five of these top-seven drought events occurred during El Niño -Southern Oscillation (ENSO) warm phases (1982,1987,1994,1997,2009) being three of them during strong El Niño years (1982,1987,1997). The other two drought events developed under ENSO neutral conditions (1986,2005). It is widely accepted that dry conditions prevail in the southern part of Mexico during El Niño summers, and wetter conditions dominate during La Niña throughout the American tropical regions including southern parts of Mexico [60][61][62].
Severe and extreme drought events that occurred in the period 1980-2017 were selected for analysis in this study, corresponding to events in 1982 and 2005. The spatial representations of SPEI-6 values in October for these two episodes show different patterns (  top-seven drought events occurred during El Niño -Southern Oscillation (ENSO) warm phases (1982,1987,1994,1997,2009) being three of them during strong El Niño years (1982,1987,1997 Severe and extreme drought events that occurred in the period 1980-2017 were selected for analysis in this study, corresponding to events in 1982 and 2005. The spatial representations of SPEI-6 values in October for these two episodes show different patterns (

Water Vapor Transport toward the Balsas River Basin
To better understand the origin of the precipitation deficits, we analyzed the anomalies of moisture transport toward the BRB during the drought events. Figure 4 shows the sum of (E-P)n > 0 up to 10 days (n = 1, …, 10) before an air mass that moved towards the BRB reached the area ((E-P)1-10 > 0 from now on) from May to October for the study period 1980-2017. Figure 4 also depicts the vertically integrated moisture flux (VIMF) and geopotential height isolines at 925 hPa for the same study period and the same months. In May, positive water vapor values were found in the northern and north-western BRB along the Pacific coast (Figure 4a), linked to the circulation of the North Pacific Subtropical High (NPSH). A smaller amount of moisture arrived following the Caribbean Low-Level

Water Vapor Transport toward the Balsas River Basin
To better understand the origin of the precipitation deficits, we analyzed the anomalies of moisture transport toward the BRB during the drought events. Figure 4 shows the sum of (E-P) n > 0 up to 10 days (n = 1, . . . , 10) before an air mass that moved towards the BRB reached the area ((E-P) 1-10 > 0 from now on) from May to October for the study period 1980-2017. Figure 4 also depicts the vertically integrated moisture flux (VIMF) and geopotential height isolines at 925 hPa for the same study period and the same months. In May, positive water vapor values were found in the northern and north-western BRB along the Pacific coast (Figure 4a Caribbean Sea, extending to the tropical North Atlantic Ocean. Our results confirm that the CLLJ core south of the Greater Antilles [63,64] is an important moisture source for the BRB during boreal summer (Figure 4b), as pointed out by other previous research works [62,65]. During October, the NASH migrates north-eastward, and its western flank causes more water vapour to arrive to the BRB from the north of the Gulf of Mexico (Figure 4c). From May to October (Figure 4a-c), evaporation clearly dominates over precipitation in the air masses located over the basin itself and the surrounding areas.
Jet (CLLJ) region. From June to September (JJAS) the distribution of water vapor fluxes during each month was fairly similar; therefore, the 4 month average is shown in Figure  4b. The NPSH system moved northwestward, the influence of the North Atlantic Subtropical High (NASH) on the Caribbean was stronger and the trade winds intensified along its southern flank. Particles transporting moisture spread from the Yucatan Peninsula to the Caribbean Sea, extending to the tropical North Atlantic Ocean. Our results confirm that the CLLJ core south of the Greater Antilles [63,64] is an important moisture source for the BRB during boreal summer (Figure 4b), as pointed out by other previous research works [62,65]. During October, the NASH migrates north-eastward, and its western flank causes more water vapour to arrive to the BRB from the north of the Gulf of Mexico (Figure 4c). From May to October (Figure 4a-c), evaporation clearly dominates over precipitation in the air masses located over the basin itself and the surrounding areas. To quantify water vapor contributions, it is necessary to define the moisture source regions independently. The threshold of the 95th percentile of the (E-P)1-10 > 0 values averaged from May to October during the study period (1980-2017) was used to delimit the source regions in which the water vapor recharge was most intense. Seven main moisture sources that were potentially important for precipitation development over BRB were defined (Figure 5a), comprising four oceanic and three terrestrial sources. The moisture arriving from the ocean was divided into four regions: the Caribbean Sea (CAR), the Gulf of Mexico (GOM), the rest of the tropical Atlantic (ATL) and the Pacific (PAC) from the west. The terrestrial evaporative moisture source regions were classified into three categories: north of the BRB (NORTH), the southern BRB region including the Yucatan Peninsula (SOUTH), and finally the BRB itself (BRB). The monthly evolution of (E-P)1-10 > 0 integrated over these seven areas is shown in Figure 5b. As expected, the evaporative source of BRB itself is highly significant, representing by far the main atmospheric moisture source during the wet season (note that the y-axis of Figure 6b is in a logarithmic scale). NORTH is the second moisture source during the beginning (May) and the end (October) of the wet season. However, from June to September, the contributions of SOUTH and CAR are slightly higher.
According to these results, the importance of the available moisture in the atmosphere of the BRB itself must be decisive compared with the rest of the moisture sources. For simplicity, we limited the analysis to the source regions delineated by the 99.5th percentile of the (E-P)1-10 > 0 values (black contours in Figure 5a). Three moisture sources were obtained: the terrestrial evaporative source of the BRB, the Yucatan Peninsula inner SOUTH and the central area of CAR. To quantify water vapor contributions, it is necessary to define the moisture source regions independently. The threshold of the 95th percentile of the (E-P) 1-10 > 0 values averaged from May to October during the study period (1980-2017) was used to delimit the source regions in which the water vapor recharge was most intense. Seven main moisture sources that were potentially important for precipitation development over BRB were defined (Figure 5a), comprising four oceanic and three terrestrial sources. The moisture arriving from the ocean was divided into four regions: the Caribbean Sea (CAR), the Gulf of Mexico (GOM), the rest of the tropical Atlantic (ATL) and the Pacific (PAC) from the west. The terrestrial evaporative moisture source regions were classified into three categories: north of the BRB (NORTH), the southern BRB region including the Yucatan Peninsula (SOUTH), and finally the BRB itself (BRB). The monthly evolution of (E-P) 1-10 > 0 integrated over these seven areas is shown in Figure 5b. As expected, the evaporative source of BRB itself is highly significant, representing by far the main atmospheric moisture source during the wet season (note that the y-axis of Figure 6b is in a logarithmic scale). NORTH is the second moisture source during the beginning (May) and the end (October) of the wet season. However, from June to September, the contributions of SOUTH and CAR are slightly higher.
According to these results, the importance of the available moisture in the atmosphere of the BRB itself must be decisive compared with the rest of the moisture sources. For simplicity, we limited the analysis to the source regions delineated by the 99.5th percentile of the (E-P) 1-10 > 0 values (black contours in Figure 5a). Three moisture sources were obtained: the terrestrial evaporative source of the BRB, the Yucatan Peninsula inner SOUTH and the central area of CAR. The above moisture sources show interannual variability. The bars in Figure 6 represent the accumulated monthly anomalies of the water vapor supplied by these three moisture sources (BRB: blue, SOUTH: pink and CAR: red; black solid lines in Figure 5a) for the severe and the extreme drought events that occurred in the period 1980-2017. In 1982 (Figure 6a), unexpected positive anomalies of moisture supplies were found for the three sources during the 6 months, including the month of August, when the peak value of SPEI-1 occurred (see Table 1). In contrast, during 2005 (Figure 6b), all source regions showed lower contributions compared with the climatological mean; therefore, the lack of water vapor was associated with precipitation deficits. It is worth pointing out that the large-scale circulation variability in local BRB conditions was far more important than the moisture transport variability from the other two top-contributing external sources, SOUTH and CAR. To explain the drought signal in 1982 despite the positive anomalies found in moisture transport, we also analyzed the evolution of omega anomalies at 500-hPa averaged over the BRB area (Figure 6a). This drought event was associated with positive anomalies of omega from June to October, indicating a general meteorological pattern of subsidence that inhibits the upward movement needed to generate precipitation. In 2005, besides the negative anomalies of water vapor transport found, subsidence was also present at a 500 hPa level over the basin during most of the months.   The above moisture sources show interannual variability. The bars in Figure 6 represent the accumulated monthly anomalies of the water vapor supplied by these three moisture sources (BRB: blue, SOUTH: pink and CAR: red; black solid lines in Figure 5a) for the severe and the extreme drought events that occurred in the period 1980-2017. In 1982 (Figure 6a), unexpected positive anomalies of moisture supplies were found for the three sources during the 6 months, including the month of August, when the peak value of SPEI-1 occurred (see Table 1). In contrast, during 2005 (Figure 6b), all source regions showed lower contributions compared with the climatological mean; therefore, the lack of water vapor was associated with precipitation deficits. It is worth pointing out that the large-scale circulation variability in local BRB conditions was far more important than the moisture transport variability from the other two top-contributing external sources, SOUTH and CAR. To explain the drought signal in 1982 despite the positive anomalies found in moisture transport, we also analyzed the evolution of omega anomalies at 500-hPa averaged over the BRB area (Figure 6a). This drought event was associated with positive anomalies of omega from June to October, indicating a general meteorological pattern of subsidence that inhibits the upward movement needed to generate precipitation. In 2005, besides the negative anomalies of water vapor transport found, subsidence was also present at a 500 hPa level over the basin during most of the months.  The above moisture sources show interannual variability. The bars in Figure 6 represent the accumulated monthly anomalies of the water vapor supplied by these three moisture sources (BRB: blue, SOUTH: pink and CAR: red; black solid lines in Figure 5a) for the severe and the extreme drought events that occurred in the period 1980-2017. In 1982 (Figure 6a), unexpected positive anomalies of moisture supplies were found for the three sources during the 6 months, including the month of August, when the peak value of SPEI-1 occurred (see Table 1). In contrast, during 2005 (Figure 6b), all source regions showed lower contributions compared with the climatological mean; therefore, the lack of water vapor was associated with precipitation deficits. It is worth pointing out that the large-scale circulation variability in local BRB conditions was far more important than the moisture transport variability from the other two top-contributing external sources, SOUTH and CAR. To explain the drought signal in 1982 despite the positive anomalies found in moisture transport, we also analyzed the evolution of omega anomalies at 500-hPa averaged over the BRB area (Figure 6a). This drought event was associated with positive anomalies of omega from June to October, indicating a general meteorological pattern of subsidence that inhibits the upward movement needed to generate precipitation. In 2005, besides the negative anomalies of water vapor transport found, subsidence was also present at a 500 hPa level over the basin during most of the months.

Climatic Fields Anomalies during the Droughts of 1982 and 2005
A more detailed examination of the underlying causes of moisture advection and subsidence anomalies during the droughts of 1982 and 2005 could be performed through the assessment of the atmospheric circulation and vertical motions.

Drought Event in 1982
Anomalies of horizontal wind at 850 hPa and SST for the drought of 1982 are shown in Figure 7a. The SST pattern shows El Niño-like conditions over the tropical Pacific Ocean (Figure 7a). SST anomalies over the eastern tropical Pacific and those over the tropical North Atlantic (TNA) showed opposite signs. The SST gradients between the tropical Pacific and the TNA favor a stronger southern branch of the CLLJ via Walker circulation [66]. Other authors have previously shown similar results; that is, a strong southern branch of the CLLJ occurring in conjunction with warm Pacific SST anomalies and cold SST anomalies in the TNA [64,[66][67][68][69][70][71][72][73]. This configuration favors greater moisture transport toward the BRB, although it is important to keep in mind that CLLJ is a secondary moisture source for the BRB, as stated in the previous section.
Anomalies of horizontal wind at 850 hPa and SST for the drought of 1982 are shown in Figure 7a. The SST pattern shows El Niño-like conditions over the tropical Pacific Ocean (Figure 7a). SST anomalies over the eastern tropical Pacific and those over the tropical North Atlantic (TNA) showed opposite signs. The SST gradients between the tropical Pacific and the TNA favor a stronger southern branch of the CLLJ via Walker circulation [66]. Other authors have previously shown similar results; that is, a strong southern branch of the CLLJ occurring in conjunction with warm Pacific SST anomalies and cold SST anomalies in the TNA [64,[66][67][68][69][70][71][72][73]. This configuration favors greater moisture transport toward the BRB, although it is important to keep in mind that CLLJ is a secondary moisture source for the BRB, as stated in the previous section.
Negative vertical velocity (upward movement, in green) anomalies at 600 hPa (Figure 7b) were found over the southern part of the Pacific Inter Tropical Convergence Zone (ITCZ-the ITCZ is marked approximately by the blue line in Figure 7b). Positive anomalies (subsidence, in pink) were present in the Atlantic ITCZ and northern South America. Over the northern areas of the eastern Pacific ITCZ, positive anomalies were also present and reached the BRB region, which is consistent with a southward displacement of the Pacific ITCZ. Other authors have shown similar results for El Niño years. For example, global ENSO-altered Walker and Hadley circulations have been shown by analyzing atmospheric reanalysis products [74,75], with the eastern Pacific ITCZ moving closer to the equator during El Niño summers. Intense convection in the eastern tropical Pacific around 5°N has been reported to produce an intense regional Hadley cell with anomalously strong subsidence over Mexico that inhibits the development of deep convective activity [43]. Nevertheless, it is unclear how an increase of moisture and simultaneous subsidence over the BRB can be compatible. The Atlantic trade wind areas are sometimes dominated by extensive inversions that form in the descending branches of the Hadley circulation, resulting from the interaction between this large-scale subsiding air and convection-driven rising air from the lower levels [76][77][78][79]. Trade wind inversions (TWIs) limit vertical development, confining the humidity in the lower levels. Criteria for inversion identification have been traditionally based on temperature, relative humidity and mixing ratio profiles in tropical and subtropical areas [78]. Equivalent potential temperature (θE) profiles have been recently used to capture the signature of TWIs in Puerto Rico [25,80]. Figure 8 depicts the 1982 seasonal Negative vertical velocity (upward movement, in green) anomalies at 600 hPa ( Figure 7b) were found over the southern part of the Pacific Inter Tropical Convergence Zone (ITCZthe ITCZ is marked approximately by the blue line in Figure 7b). Positive anomalies (subsidence, in pink) were present in the Atlantic ITCZ and northern South America. Over the northern areas of the eastern Pacific ITCZ, positive anomalies were also present and reached the BRB region, which is consistent with a southward displacement of the Pacific ITCZ. Other authors have shown similar results for El Niño years. For example, global ENSO-altered Walker and Hadley circulations have been shown by analyzing atmospheric reanalysis products [74,75], with the eastern Pacific ITCZ moving closer to the equator during El Niño summers. Intense convection in the eastern tropical Pacific around 5 • N has been reported to produce an intense regional Hadley cell with anomalously strong subsidence over Mexico that inhibits the development of deep convective activity [43]. Nevertheless, it is unclear how an increase of moisture and simultaneous subsidence over the BRB can be compatible.
The Atlantic trade wind areas are sometimes dominated by extensive inversions that form in the descending branches of the Hadley circulation, resulting from the interaction between this large-scale subsiding air and convection-driven rising air from the lower levels [76][77][78][79]. Trade wind inversions (TWIs) limit vertical development, confining the humidity in the lower levels. Criteria for inversion identification have been traditionally based on temperature, relative humidity and mixing ratio profiles in tropical and subtropical areas [78]. Equivalent potential temperature (θ E ) profiles have been recently used to capture the signature of TWIs in Puerto Rico [25,80]. Figure 8 depicts the 1982 seasonal anomalies of θ E , temperature and relative humidity profiles averaged over the BRB region. The value of θ E sharply decreases with height from 850 to 500 hPa due to the strong decrease in relative humidity with height compared with the subtle increase in temperature across the inversion. To the best of our knowledge, no previous studies of TWIs exist in Mexico, making it difficult to assess the intensity of this seasonal mean inversion. IN comparison with the Caribbean, where the strength of the inversions during July ranges between 0.5 to 1.0 • C [76], the mean value of ∆T = 0.2 • C ( Figure 8) seems to be weak. Therefore, it could be possible that this TWI should allow some cloud development to produce moderate rains that weaken rapidly by the entrainment of dry air due to subsidence in the free troposphere [25]. Anomalies of relative humidity are positive below 700 hPa, which is consistent with the increase in the column content of moisture diagnosed by FLEXPART. Anomalous subsidence prevailed over the BRB in almost all vertical levels during almost all seasons ( Figure 8) except May, as also shown in Figure 6a for the mid troposphere. troposphere.
In summary, in 1982, despite a higher moisture contribution, convection was inhibited further over the BRB, leading to precipitation deficits. The reason was the development of a southern and intense ITCZ over the eastern tropical Pacific, producing an anomalous ascent that resulted in a strengthened regional Hadley circulation to the north. The anomalous Hadley circulation generated stronger subsidence over the BRB, inhibiting the development of deep convection. However, a greater moisture content was trapped below an anomalous trade wind inversion.
ENSO provides a clear connection with the position of the ITCZ in the equatorial eastern Pacific, typically migrating southward from its northerly position during El Niño events [74,81]. Five of the seven moderate to extreme drought events identified in the present work took place under El Niño conditions, suggesting the ENSO index to be a useful predictor of droughts over the BRB. However, over the past two decades, El Niño events have weakened, and their SST anomalies have shifted westward towards the central Pacific, impeding southward migration toward the Equator of the ITCZ [82]. In fact, although the 2015 event was classified as canonical El Niño, this event exhibited several features distinct from the previous strong El Niño, and the eastern Pacific ITCZ failed to cross south of the Equator even at the El Niño peak. Despite the large magnitude of the 2015 event, the drought over the BRB did not reach the category of moderate (see Figure 2). How ENSO phenomenon should change with anthropogenic warming is still an ongoing research [83].

Drought Event in 2005
Regarding the 2005 wet season ( Figure 9a) the most significant signal was that SST anomalies over the Caribbean Sea and the TNA were high (>0.8 °C). Negative anomalies In summary, in 1982, despite a higher moisture contribution, convection was inhibited further over the BRB, leading to precipitation deficits. The reason was the development of a southern and intense ITCZ over the eastern tropical Pacific, producing an anomalous ascent that resulted in a strengthened regional Hadley circulation to the north. The anomalous Hadley circulation generated stronger subsidence over the BRB, inhibiting the development of deep convection. However, a greater moisture content was trapped below an anomalous trade wind inversion.
ENSO provides a clear connection with the position of the ITCZ in the equatorial eastern Pacific, typically migrating southward from its northerly position during El Niño events [74,81]. Five of the seven moderate to extreme drought events identified in the present work took place under El Niño conditions, suggesting the ENSO index to be a useful predictor of droughts over the BRB. However, over the past two decades, El Niño events have weakened, and their SST anomalies have shifted westward towards the central Pacific, impeding southward migration toward the Equator of the ITCZ [82]. In fact, although the 2015 event was classified as canonical El Niño, this event exhibited several features distinct from the previous strong El Niño, and the eastern Pacific ITCZ failed to cross south of the Equator even at the El Niño peak. Despite the large magnitude of the 2015 event, the drought over the BRB did not reach the category of moderate (see Figure 2). How ENSO phenomenon should change with anthropogenic warming is still an ongoing research [83].

Drought Event in 2005
Regarding the 2005 wet season (Figure 9a) the most significant signal was that SST anomalies over the Caribbean Sea and the TNA were high (>0.8 • C). Negative anomalies of low-level wind over the CLLJ core and a cyclonic circulation over the Gulf of Mexico were evident, inhibiting the water transport from the CAR region, which is in agreement with the lower moisture over BRB, SOUTH and CAR found during this drought event in the previous section.
Consistently, positive vertical velocity anomalies (upward motion) over the Caribbean Sea and the TNA are found in Figure 9b. In fact, the 2005 Atlantic hurricane season was the most active on record until the 2020 season, which exceeded that season [84]. The year 2005 also reached the highest SST anomalies (0.9 °C) and the largest numbers of named storms and hurricanes recorded [85]. Figure 9b also shows negative geopotential height anomalies at 850 hPa (red lines) and positive anomalies at 200 hPa (blue lines) north of the convective regions (green contours). These anomalies are consistent with Gill's model of off-equatorial heating [86], whose solution is a Rossby wave propagating westward from the forcing region with negative/positive pressure anomalies at the lower/upper levels. Gill's model also explains the occurrence of subsidence elsewhere around the rising motion associated with the warming region. Finally, it is also necessary to mention that, in this case, we found inter-monthly variations of θE profiles, with TWI-like profiles in May, June and September driven by largescale subsidence (Figure 6b), and these types of profiles were not observed during July, August and October (figures not shown).
As an overview, the 2005 drought was mainly associated with a reduced moisture supply from the main sources. A cyclonic circulation over the Gulf of Mexico resulted from the atmospheric response to a Caribbean Sea-persistent SST heating anomaly and prevented the moisture from reaching the BRB. Besides this water vapor scarcity, general subsidence took place over the BRB during the majority of the months, also inhibiting potential convection.
Data from observations and model simulations suggest that there is little historical precedent for the 2005 SST event [87]. The contribution of natural climate variability and anthropogenic forcing to this key event has been evaluated. The interaction of three independent climate teleconnections has been related to this anomalously high SST in the southern Caribbean [88], tropical North Atlantic (TNA), Atlantic multidecadal oscillation (AMO) and Atlantic meridional mode (AMM) climate indices. Donner et al. [88] found that thermal stress reaching the 2005 level would be extremely rare without any anthropogenic forcing. In this case, in contrast with the mechanism promoting the 1982 drought, the influence of global warming provides a background level that increases the risk of future drought events similar to that in 2005. High SST anomalies (Figure 9a) over the TNA increase the atmospheric water vapor content and are critical to promote disturbances that can develop into tropical storms. Consistently, positive vertical velocity anomalies (upward motion) over the Caribbean Sea and the TNA are found in Figure 9b. In fact, the 2005 Atlantic hurricane season was the most active on record until the 2020 season, which exceeded that season [84]. The year 2005 also reached the highest SST anomalies (0.9 • C) and the largest numbers of named storms and hurricanes recorded [85]. Figure 9b also shows negative geopotential height anomalies at 850 hPa (red lines) and positive anomalies at 200 hPa (blue lines) north of the convective regions (green contours). These anomalies are consistent with Gill's model of off-equatorial heating [86], whose solution is a Rossby wave propagating westward from the forcing region with negative/positive pressure anomalies at the lower/upper levels. Gill's model also explains the occurrence of subsidence elsewhere around the rising motion associated with the warming region.
Finally, it is also necessary to mention that, in this case, we found inter-monthly variations of θ E profiles, with TWI-like profiles in May, June and September driven by large-scale subsidence (Figure 6b), and these types of profiles were not observed during July, August and October (figures not shown).
As an overview, the 2005 drought was mainly associated with a reduced moisture supply from the main sources. A cyclonic circulation over the Gulf of Mexico resulted from the atmospheric response to a Caribbean Sea-persistent SST heating anomaly and prevented the moisture from reaching the BRB. Besides this water vapor scarcity, general subsidence took place over the BRB during the majority of the months, also inhibiting potential convection.
Data from observations and model simulations suggest that there is little historical precedent for the 2005 SST event [87]. The contribution of natural climate variability and anthropogenic forcing to this key event has been evaluated. The interaction of three independent climate teleconnections has been related to this anomalously high SST in the southern Caribbean [88], tropical North Atlantic (TNA), Atlantic multidecadal oscillation (AMO) and Atlantic meridional mode (AMM) climate indices. Donner et al. [88] found that thermal stress reaching the 2005 level would be extremely rare without any anthropogenic forcing. In this case, in contrast with the mechanism promoting the 1982 drought, the influence of global warming provides a background level that increases the risk of future drought events similar to that in 2005.

Summary and Conclusions
In this paper, the 6 month standardized precipitation-evapotranspiration index (SPEI- 6) in October and the 6 month standardized precipitation index (SPI-6) in the same month were used to identify and rank drought episodes over the Balsas River Basin (BRB) from 1980 to 2017. According to the classification of McKee et al. [28], drought events were classified into mild, moderate, severe and extreme, and the most intense events (severe and extreme) were selected for study. The anomalies in moisture supply during the selected drought periods were evaluated by using a Lagrangian approach. Finally, climatic fields at several levels were analyzed in order to investigate the physical mechanisms of severe drought occurrence. The main conclusions of this study are summarized as follows: 1.
The evolution of SPEI-6 and SPI-6 averaged over the BRB are quite similar during the studied period, suggesting a larger contribution of precipitation deficit than temperature excess to the drought occurrence.

2.
Two major drought events-one severe (2005) and another of extreme intensity (1982)were found in the period 1980-2017. The differences between the spatial patterns obtained by both indices (SPEI-6 and SPI-6) were much reduced, confirming the important role of precipitation in these two events vs. atmospheric evaporative demand driven by temperature increase in the Mexican territory.

3.
The main moisture sources during the wet season (May-October) for the BRB were identified. The terrestrial evaporative source of BRB itself was shown to be by far the main moisture source during the entire wet season, followed by the southern BRB terrestrial region and the Caribbean Low-Level Jet (CLLJ) core region, which are more active from June to September.

4.
During the drought event of 2005, an anomalously warm SST emerged over the Caribbean Sea. A persistent elevated SST over the Atlantic warm pool favored hurricane activity during this year. As a response to this heating, lower-level negative pressure anomalies and upper tropospheric positive pressure anomalies developed to the northwest, consistent with Gill's model. As a consequence of the lower level cyclonic circulation over the Gulf of Mexico, negative anomalies of the CLLJ winds arose, and the BRB itself, CLLJ and SOUTH provided a decreased water vapor contribution. Additionally, this drought event was associated with subsidence that also prevented convection over the BRB in four of the six months.

5.
The mechanisms associated with the the top drought event, which occurred in 1982, were also determined. An intense and southward-shifted Inter Tropical Convergence Zone (ITCZ) over the eastern tropical Pacific produced an intense regional Hadley circulation that generated anomalously strong subsidence over the BRB. This descent branch of the Hadley cell inhibited the development of deep convection, although this drought episode was concurrent with an increase in the moisture supply. A strong trade wind inversion (TWI) driven by this large-scale subsiding air from the upper troposphere confined the moisture content in the lower layers.
A better understanding of drought causes is clearly of major scientific and social value. This work identifies the underlying physical mechanisms behind severe droughts in the BRB and generates an understanding that may help to address the major challenge of their prediction. Seasonal forecasts of severe droughts would give an advance warning to manage the detrimental impacts on crops, livestock farming and household food insecurity associated with a decline in production. An additional knowledge of the water-hydroelectric supply nexus in combination with seasonal forecasts of droughts in the BRB would also give the opportunity to manage drought impacts in the energy sector early. Finally, Mexico's vulnerability to drought by climate change has been highlighted in the literature. This work presents a starting point to study the evolution of physical mechanisms causing droughts in a changing environment.