Study site

The study was carried out in the Cuatro Ciénegas Basin (CCB; 26° 45’- 27° 00’ N and 101° 48’- 102° 17’ W) in central-northern Mexico, which is part of the Chihuahuan Desert (Figure 1). The CCB has an area of 150,000 km², the study area had an elevation of 740 masl and it is completely surrounded by mountains. These geographical barriers are favored that the climate dynamic differentiated from the rest of the Chihuahuan Desert (^{Archer & Predick, 2008}). The climate is seasonally arid with two contrasting seasons and an average annual temperature of 21.2 °C with 252.5 mm of annual precipitation (^{SMN-Conagua, 2018}). The first season, from November to April, is cold and dry with minimum and maximum temperatures of 4.8 and 29.8 °C, respectively, and 65.9 mm of precipitation. The second season, from May to October, is hot (with minimum and maximum temperatures of 15.7 and 33.4 °C, respectively) and around 60 % (186.6 mm) of the total annual precipitation is concentrated within this season (^{SMN-Conagua, 2018}). Jurassic-era gypsum is the dominant parent material at the western side of the basin, while Jurassic-era limestones dominate the eastern side (^{McKee, Jones, & Long, 1990}). According to the World Reference Base for soil resources (^{WRB, 2015}) the predominant soils are Gypsisol and Calcisol at the western and eastern sides of the basin, respectively. The main vegetation types are: 1) grassland (G), dominated by Sporobolus airoides (Torr.) Torr., and Allenrolfea occidentalis (S. Watson) Kuntze; 2) microphyll scrub, dominated by Jatropha dioica Cerv., Larrea tridentata (DC) Cov., and Fouqueria sp Kunth (^{Perroni, García-Oliva, & Souza, 2014}); and 3) rosetophylous scrub (RS) dominated by Dhasylirium cedrosanum Trel., and Yucca treculeana Carriére (^{González, 2012}).

Figure 1 Location of the Cuatro Ciénegas Basin (CCB) in Coahuila, Mexico.  

Climate data set

For the present study, a data set covering a 70-year period (1941 to 2013) was analyzed. The data came from two weather stations within CCB (Cuatro Ciénegas station of the Mexican National Water Commission (http://smn.cna.gob.mx/, data from 1941 to 2013) and the “Rancho Pozas Azules” station of INIFAP (http://clima.inifap.gob.mx/redinifap/, data from 2005 to 2013; Figure 1). Daily and monthly data were used to evaluate the following parameters: precipitation (pp), mean temperature (T_mean), maximum temperature (T_max) and minimum temperature (T_min). The pp, T_max and T_min databases were subjected to quality control in order to identify possible errors (i.e., pp > 0 or T_min > T_max; (^{Moberg & Jones, 2005}). The monthly thermal oscillation was calculated from the difference between T_max and T_min (^{Tabari & Hosseinzadeh-Talaee, 2011}).

Trend analysis: temperature and precipitation

EViews version 7.0 (Quantitative Micro Software) for parametric statistics (^{López-Díaz, Conde, & Sánchez, 2013}) and Clic-MD version 2 for nonparametric statistics (^{Bautista et al., 2013}) were used for analyzing temporal trends in temperature and precipitation. For the EViews model, the following assumptions were tested before performing the trend analysis: normality, functional form, no autocorrelation, correct specification, structural permanence, multicollinearity and no homoscedasticity. To test if the slope of the relationship between the climate variable and time differed from zero, EViews adjusted this relationship to a least squares linear regression model (^{López-Díaz et al., 2013}):

Where “y” is the climate variable (temperature or precipitation), x_j the time, C a constant coefficient, a_j the regression parameter (slope), and e the residual error. An increase or decrease of the trend for the climate variable under analysis was indicated by a positive or negative “a” value, respectively. Slope intensity was analyzed by Pearson correlation analysis (^{López-Díaz et al., 2013}).

The Clic-MD software (^{Bautista, Pacheco, & Bautista-Hernández, 2014}) included two statistical analyses: 1) a Spearman simple linear correlation was used to evaluate changes in climate variable intensities, and 2) a non-parametric Mann-Kendall test (MK-T) was used for analysis of the temporal trends of climatic variables (^{Jain & Kumar, 2012}). The MK-T tests the null hypothesis of no temporal trend (where the slope is equal to zero; (^{Tabari & Hosseinzadeh-Talaee, 2011}). The null hypothesis is rejected when Z > 1.96. Positive or negative values of Z indicate increments or decrements of the climate variable in the time series, respectively (^{Bautista et al., 2013}; ^{Jun et al., 2012}). The autocorrelation of the residuals was analyzed by the Breusch-Godfrey test (^{Breusch, 1979}). Additionally, the homoscedasticity was analyzed by the White test (^{Bautista et al., 2014}).

Analyses of Extreme Climate Events (ECE)

The ECEs were identified as those values below the 10^th (lower extremes) or above the 90^th (upper extremes) percentile distribution values in any climate parameter (^{Ben-Gai, Bitan, Manes, Alpert, & Rubin, 1999}; ^{Easterling et al., 2000}; ^{IPCC, 2012}). For this analysis, we used daily temperature data and monthly precipitation data. In order to analyze whether the frequency of temperature (T_max and T_min) and precipitation ECEs had changed over the last 36 years, the 70-year CCB dataset was divided into two time periods: a) from 1941 to 1976 and b) from 1977 to 2013. The series of 70 years was split in two to ensure that each period had at least 30 years of data according to the number of years needed to define a climate (^{WMO, 2018}). A chi-square test was used to identify changes in the frequency of lower and upper extreme events of T_max, T_min and for upper extreme events of precipitation between the two time periods. To identify monthly changes between the two time periods, we applied a residual analysis to the lower and upper extreme events of T_max, T_min and precipitation. Where the residual value was > 1.96 or < -1.96, the change in frequency was considered to be significant (^{Ben-Gai et al., 1999}; ^{Everitt, 1992}).

Precipitation analyses

We used a data set of 70 years (1941 to 2013) for calculated an index based on rainfall data: the standardized pluviometric drought index or SPDI (^{Pita, 2001}; ^{Thielen et al., 2020}). The SPDI is an index based on cumulative monthly precipitation anomalies and is used to identify the duration and the severity of both, drought (^{Alexander et al., 2006}; ^{Thielen et al., 2020}) and humidity (^{Thielen et al., 2020}). For the SPDI calculation, the equation used was:

Where the APA_i is the accumulated rainfall anomaly for month i. The APA- is the average value of the accumulated precipitation anomalies of all the months in the data series. The σAPA is the standard deviation of the accumulated precipitation anomalies of all the months in the series (^{Thielen et al., 2020}).

The indexes for the 70 years were calculated and graphed with the software CLIC-MD version 2 (^{Bautista et al., 2013}). The indexes values obtained from each month were classified into a category according to proposed by ^{Thielen et al. (2020)} for the SPDI. The index is bounded by SPDI ≤-2.0 and SPDI ≥ 2.O. The positive values indicate humid conditions and the negative values indicates dry conditions (Table 1).

We also calculated the following parameters: the amount of rain in a typical rainy month (r), the rainfall concentration or the number of rainy months (P) and the equitability, which is a relative measurement of rainfall concentration (E). We calculated these parameters using the equations proposed by ^{Ezcurra and Rodrigues (1986)} for the total period (1941 to 2013) and the two defined periods, where “x” is the amount of precipitation in one month per year and “n” is the number of months (12):

Climate type classification

For classification of climate type, analysis of data records for a period of at least 30 years is required (^{WMO, 2018}). To detect changes in the climate type within the last 36 years, the 70-year CCB database was divided into two periods: a) from 1941 to 1976 and b) from 1977 to 2013. For each period, climate type was characterized using the Köppen classification method modified for Mexico by ^{García (1981)} .

Trends analysis: Temperature, thermal oscillation and precipitation

T_min (Table 2 and Figure 2) and T_mean (Table 3 and Figure 3) presented a significant increase of approximately 2 °C from 1941 to 2013 in January to March, and in May. T_min also increased 2 °C in the summer months (from June to August) and approximately 1 °C in winter months (November to December; Table 2). Additionally, the T_mean increased of approximately 1 °C in June and 2 °C in November. We found no significant trend in the T_max from 1941 to 2013 with either of the software packages used (Figure 4). For thermal oscillation, we observed a negative trend by a decrease of ca. 2 °C in March, July, August and September (Table 4).

Figure 2 Trend analysis of monthly minimum temperatures from 1941 to 2013 in the CCB. The correlation coefficient is shown. 

Figure 3 Trend analysis of monthly mean temperatures from 1941 to 2013 in the CCB. The correlation coefficient is shown. 

Figure 4 Trend analysis of monthly maximum temperatures from 1941 to 2013 in the CCB. The correlation coefficient is shown. 

For precipitation, we observed an increasing trend in July only. However, it is likely that the lack of observable trends in the other months was due to the wide variability of the data (Table 5).

Analyses of Extreme Climate Events (ECE)

The Chi-square analysis was significant (p < 0.005) in all the cases of temperature but was not significant for monthly precipitation (Table 6). For the T_min of January, February and May, we observed an increase in the frequency of lower temperature (months increasingly colder) from the first (1941-1976) to the second time period (1977-2013; Figure 5a and Figure 6). Likewise, in April, June, July, November and December, we observed a decrease in the lower extremes of T_min (months increasingly less cold) from the first to the second period (Figure 5a and Figure 6).

Figure 5 Analysis of residuals from January to December of the frequencies of Extreme Climate Events (ECE) over two periods: 1) 1941-1976 and 2) 1977-2013 in the CCB; A) T_min lower extremes, B) T_min upper extremes, C) T_max lower extremes, and D) T_max upper extremes. The asterisk (*) indicates the values above dashed lines that are significantly different to the expected value (p < 0.05). 

Figure 6 Frequency distribution analysis of minimum temperature from 1941 to 1976 (red line) and from 1977 to 2013 (blue line) in the CCB. The green color shows the 10th and 90th percentile of distribution in the data series. 

For the upper extremes of T_min in February, July, August and November, we observed an increase of frequencies (increasingly warmer T_min) from the first to the second period (Figure 5b and Figure 6). For the months of March, May, June, September and December, the frequencies of upper extremes (colder T_min) of T_min decreased from the first to the second period (Figure 5b and Figure 6).

In January, April, June and September, we observed a decrease of the frequencies of the lower extremes (increasingly warmer months) of T_max from the first to the second period. In the months of July, August, October, November and December, we observed an increase in the frequencies of lower extremes (fewer warm months) of T_max from the first to the second period (Figure 5c and Figure 7).

Figure 7 Frequency distribution analysis of maximum temperature from 1941 to 1976 (red line) and from 1977 to 2013 (blue line) in the CCB. The green color shows the 10th and 90th percentile of distribution in the data series. 

In the upper extremes of T_max, we observed in January, February, June and July an increase of frequencies (increasingly warmer months) from the first to the second period. In March, April and September, we observed a decrease of frequencies of upper extreme events (fewer warm months) of T_max from the first to the second period (Figure 5d and Figure 7).

We did not find significant differences in the frequencies of extreme monthly precipitation between the first and second periods (Table 5).

Analysis of Precipitation

We identified a trend of increasing precipitation in the month of June using the Mann Kendall test (Table 5). Before the year 1985, the SPDI values don’t show years classified with severe drought or whit extreme drought (Figure 8). After the year 1985, two years (2013 and 2012) were classified with severe drought and three years with extreme drought (1988, 1989 and 1990). Finally, 13 years classified as extremely humid (1949, 1958, 1971, 1976, 1977, 1978, 1979, 1986, 1992, 1997, 2003, 2008, 2010).

Figure 8 Standardized pluviometric drought index (SPDI), calculated from 1941 to 2013 in the CCB. 

The calculated “r”, “P” and “E” values, respectively, for the complete period (1941-2013) were 54 mm, 277 and 0.32 mm; for the first period, these values (1941-1976) were 51, 137 and 0.32 mm and for the second period (1977-2013), they were 57, 142 and 0.32 mm.

Climate type classification for two periods

For the two defined time periods, we found a change of climate type. For the first period from 1941 to 1976, the climate classification according to Köppen was BWhw(x´)(e´). This climate is defined as very dry, semi-warm, and had an annual average temperature of 21.4 °C; the temperature of the coldest month (January) was 12.3 °C, while that of the hottest month (July) was 28.4 °C. Rains were markedly seasonal (summer), the percentage of rains that fell in winter was 10.7 %. Thermal oscillation was extreme (Figure 9a).

Figure 9 Monthly minimum, average and maximum values for temperature and precipitation recorded over two periods: a) From 1941 to 1976 and b) From 1977 to 2013. 

In the second period, from 1977 to 2013, the climate classification was BWhwx´(w)(e´). This climate is defined as very dry, semi-warm, and presented an annual average temperature of 21.9 °C. The temperature of the coldest month (January) was 12.9 °C, while that of the hottest month (July) was 28.8 °C. Summer rains concentrated around 90 % of the annual precipitation and the percentage of winter rain was 9.3 %. Thermal oscillation was very extreme (Figure 9b).

The change in the climate type classification from first to the second period was therefore mainly caused by a decrease in the temperature of the coldest month (January), an increase in the temperature of the warmest month (July), higher annual thermal oscillation and a reduction in the percentage of winter rain.

Temperature

Our working hypotheses were increases in atmospheric temperature and the frequency of temperature ECEs, as proposed for the Chihuahuan desert by ^{Loarie et al. (2009)} and by the IPCC scenarios (^{IPCC, 2012}). For our study site, T_min increased in almost all the months but T_mean increased only in the summer months. Moreover, the frequency of lower T_min increased in the winter months, while the frequency of upper event extremes increased during the summer months, as well as the extreme events of T_max. This means that the winters were colder, and the summer months were warmer with higher T_min and T_max extreme events, increasing the frequency of heat waves over the last 36 years. The heat wave is defined by the increment of frequencies in the upper extremes (90^th percentile of distribution) of T_max of June and July (the warmer months in the CCB) as proposed by several authors (^{Meehl & Tebaldi, 2004}; ^{Peng et al., 2011}). Additionally, the increment of frequencies in the upper extremes (90th percentile of distribution) of T_min on July and August in the second period, suggests an increase in warm nights, suggesting an heat wave as reported by previous authors (^{Meehl & Tebaldi, 2004}; ^{Peng et al., 2011}).

Positive trends of T_mean in the summer months have also been found in other desert ecosystems, such as the Sahara desert in Libya (^{Mamtimin, Et-Tantawi, Schaefer, Meixner, & Domroes, 2011}) and Almeria in Spain (^{Del-Río, Herrero, Pinto-Gomes, & Penas, 2011}); while other studies have reported T_mean increments throughout the year in other desert sites such as Jerusalem, Tripoli (^{Hasanean, 2001}) and Iran (^{Tabari & Hosseinzadeh-Talaee, 2011}; ^{Tabari, Somee, & Zadeh, 2011}).

As expected, we detected a positive trend in T_min, implying that T_min has become warmer in recent years for most months of the year. This was found in other desert ecosystems in Iran (^{Ben-Gai et al., 1999}; ^{Tabari et al., 2011}), Jordan, (^{Hamdi, Abu-Allaban, Elshaieb, Jaber, & Momani, 2009}) and North Carolina in the USA (^{Boyles & Raman, 2003}) and was also observed in other regions of the world, such as Italy (^{Brunetti, Buffoni, Maugeri, & Nanni, 2000}), Turkey (^{Türkes, Sümer, & Kiliç, 1996}). It was also observed at global level by ^{Easterling et al. (1997)} , and ^{Vose, Easterling and Gleason (2005)} . Several studies in other ecosystems (^{Boyles & Raman, 2003}; ^{Brunetti et al., 2000}; ^{Easterling et al., 1997}; ^{Boyles & Raman, 2003}; ^{Brunetti et al., 2000}; ^{Easterling et al., 1997}; ^{Vose et al., 2005}) have attributed the summer increase of the T_min in specific years to abnormalities in the ENSO combined with an increase of the positive phase of NAO and is probable that in our study site the increase in T_min may be related with this phenomena. However, there was an increased frequency of lower extreme events for T_min during the winter months, a finding also reported for Israel (^{Ben-Gai et al., 1999}), Utah in the USA (^{Santos, 2011}) and Tlaxcala in Mexico (^{López-Díaz et al., 2013}). This result indicates that winters with colder nocturnal events in the CCB have been more frequent during the last 36 years. The increase in T_min of winter months was explained by the negative phase of NAO during the winter months in other ecosystems (^{Boyles & Raman, 2003}; ^{Brunetti et al., 2000}; ^{Easterling et al., 1997}; ^{Vose et al., 2005}) we propose that this phenomenon could also cause colder winters in specific years in our study site.

In contrast, the frequency of the upper extremes for T_min and T_max during the summer increased over the last 36 years, promoting higher summer nocturnal and diurnal temperatures. Other studies in North America (^{Hasanean, 2001}; ^{López-Díaz et al., 2013}; ^{Peterson et al., 2013}) observed that in the summer months the increase in extreme temperature was produced by El Niño events in combination with the positive phase of the Pacific Decadal Oscillation (PDO), and we proposed that these phenomena could also generate the changes in the upper extremes for T_min and T_max during the summer in our study site. These results have also been reported in other studies, ^{Alexander et al. (2006)} observed a marked increase of warm nocturnal temperatures at global level throughout the year, while several authors have found a higher frequency of upper extreme temperature events of T_max during the summer months (^{Ben-Gai et al., 1999}; ^{López-Díaz et al., 2013}; ^{Santos, 2011}).

Precipitation

In the case of precipitation, our working hypothesis had been an increase in precipitation variability, but we found that the precipitation did not show any temporal trend, due to the large variability of monthly precipitation. This variability obscured any trend in the frequency of extreme precipitationevents. In desert ecosystems, precipitation events are scarce and very erratic, producing a skewed temporal distribution of the data (^{Ezcurra & Rodrigues, 1986}), as was the case in our study site. It is possible that this behavior differed from the expected trend in the CCG and ECE scenarios for the Chihuahuan Desert, explained mainly by the effect of geographical barriers on climatic dynamics within CCB. This lack of a trend in precipitation throughout the year was also observed in other desert ecosystems in Israel (^{Modarres & Silva, 2007}), Jordan, (^{Hamdi et al., 2009}) and India (^{Jain, Kumar, & Saharia, 2013}).

^{Jain and Kumar (2012)} expected that the inter-annual variability of annual precipitation could be increased by GCC during the 21^stcentury, mainly as a result of the increasing frequency and intensity of ECE (^{D'Odorico & Bhattachan, 2012}; ^{IPCC, 2012}). However, while we did not observe a significant change in the precipitation frequency of precipitation ECE, the records of annual precipitation for the second period show a higher incidence of years with annual precipitation higher than 300 mm distributed throughout the year (nine years: 1978, 1981, 1984, 1986, 1991, 1992, 1997, 2003 and 2010), in contrast to the first period where the precipitation was concentrated in the tropical cyclone seasons (5 years: 1949, 1958, 1963, 1971 and 1976).The calculated precipitation value for a typical rainy month for CCB (1941-2013) was 54 mm, and it increases from 51 mm to 57 mm from the first to the second analyzed period. We also observed a higher incidence of years with precipitation below 100 mm in the second period (6 years: 1983, 1988, 1994, 1995, 2011 and 2012) compared to the first period (4 years: 1942, 1952, 1956 and 1959). These results are according to the SPDI values, we observed an apparently increased in frequency and intensity of extreme drought and extreme precipitation events after the year 1985. The increased frequency of heavy precipitation events has been attributed to years with strong cyclones from the Gulf of Mexico, produced by a combination of La Niña and either the NAO or the PDO (^{Boyles & Raman, 2003}; ^{Brunetti et al., 2000}; ^{Vose et al., 2013}), while drought events have been associated in other ecosystems with warm subtropical anticyclones attributed to the coincidence of El Niño with either the NAO or the PDO (^{Boyles & Raman, 2003}; ^{Brunetti et al., 2000}; ^{Peterson et al., 2013}). Unfortunately, the incidences of intensive ENSO or NAO abnormalities have increased in recent decades (^{IPCC, 2013}), promoting precipitation variability.

In arid and semi-arid regions such as the CCB, changes in temperature and precipitation will affect environmental water balances, increasing the water stress experienced by organisms and leading to a significant reduction in ecosystem productivity. Unfortunately, our results suggest that winters will become colder and summers will become warmer with a high variability in the availability of water in CCB, increasing the environmental stress for organisms. For this reason, is very important to fully understand how climate is changing in order to design appropriate management strategies for adapting to such climate variability in the near future. Moreover, local climate studies, such as the present study, are critical for the calibration and development of global scenarios under GCC (^{Tabari & Hosseinzadeh-Talaee, 2011}).