Long-Term Variability in the Arabian Peninsula Droughts Driven by the Atlantic Multidecadal Oscillation

Abstract

Drought is a recurring hydroclimatic extreme over the Arabian Peninsula (AP). So far, no study has examined the changes in drought characteristics in recent decades, not to mention the background mechanisms for such changes. To this end, analyzing the Standardized Precipitation Evapotranspiration Index (SPEI) mainly from the European Reanalysis (ERA5) data sets, in addition to other observational/reanalysis data sets over the period of 1951–2020, we show that droughts over the AP have increased in frequency and severity over the last two decades. We show that this drought acceleration, which was not observed in the previous 40–50 years, is a combination of decadal variability and long-term trends. Importantly, we demonstrate that the decadal SPEI variability is due to the Atlantic Multidecadal Oscillation (AMO). The unprecedented multiyear drought over the AP in recent decades is evidently associated with the current positive phase of the AMO. We also show that the recent warming of the AP is a more significant factor in the drought intensification than the concurrent weakening of local precipitation. Furthermore, we developed a machine learning model largely based on the observed AMO–SPEI relationship. This model predicts a reduced drought severity over the AP in the near future.

Introduction

Droughts are severe natural disasters that affect millions worldwide every year. In particular, droughts over the Arabian Peninsula (AP) have grave implications because of the low precipitation, high heat, large water-stressed deserts, and excessive water exploitation (Hameed et al., 2019; Odhiambo, 2017). Drought events have become increasingly frequent over the last few decades in the Kingdom of Saudi Arabia (KSA), which comprises a major portion of the AP (Almazroui, 2019; Amin et al., 2016; Syed et al., 2022). The United Arab Emirates has also experienced low precipitation since 1999 (Sherif et al., 2014), and a substantial increase in drought episodes has been observed across Oman in the last few decades (El Kenawy et al., 2020). However, while regional studies suggest an increasing temperature trend over most of the AP (Amin et al., 2016; Attada, Dasari, Chowdary, et al., 2019), there have been no such long-term changes in local precipitation (Almazroui et al., 2020; Dasari et al., 2021). The relative contributions of the temperature and precipitation to the recent exacerbation of droughts over the AP are not yet understood.

The interannual variability of the precipitation and temperature over the AP has been related to large-scale climate drivers such as the El Niño-Southern Oscillation (ENSO), ENSO Modoki, Indian Ocean Dipole (IOD), North Atlantic Oscillation (NAO), and Arctic Oscillation (AO) (e.g., Abid et al., 2018; Attada, Dasari, Chowdary, et al., 2019; Dasari et al., 2021; Kang et al., 2015; Kotwicki & Al Sulaimani, 2009). Many of these relationships are subject to decadal variations (Dasari et al., 2021; Kang et al., 2015). La Niña, which causes an anomalous precipitation deficit over the AP during the winter, has also been correlated with meteorological droughts over the region (Barlow et al., 2002; Niranjan et al., 2016). Pagano et al. (2002) suggested an association between the negative phase of the NAO and dry conditions in the Middle East. Syed et al. (2022) claimed that the Pacific Decadal Oscillation (PDO) modulates interannual drought variability in KSA. Other factors, such as shifts in the inter-tropical convergence zone, storm tracks, and so forth, have also been suggested to influence the drought variability of the AP (Barlow et al., 2016).

Several studies have noted decadal variations in the precipitation and temperature over the AP in recent decades (Almazroui, 2020; Ehsan et al., 2020; Kang et al., 2015). Ouarda et al. (2014) showed the possible potential of large-scale drivers, like PDO and AMO, for major climate shifts over the Gulf region in recent decades. El Kenawy et al. (2020) suggested that the warming trend in the Atlantic Ocean and the transitions of PDO and ENSO in 1998 have accelerated the drying of Oman since then. Notaro et al. (2015) established a relationship between drought and dust activity over the AP at the decadal scale, derived by PDO. Ehsan et al. (2020) suggested that the Atlantic Multidecadal Oscillation (AMO) is relevant to the summer temperature variations of the Middle East. Furthermore, the decadal variability of the sea surface temperature (SST) (Krokos et al., 2019) and the wind and wave conditions (Langodan et al., 2020) of the adjoining Red Sea has been suggested to be linked with the AMO. Another study suggests the combined impact of global warming signal and AMO on Red Sea SST amplifications (Alwad et al., 2020). These studies hint that the AMO may affect the AP climate, including local droughts. Anthropogenic climate change may have also increased the severity and frequency of droughts (Douville et al., 2021). Significantly, no studies have examined the long-term trends and decadal variability of droughts over the whole AP.

We analyzed various observation-based gridded and reanalysis data sets to investigate whether droughts in the AP exhibit any decadal variations and long-term trends. In this study, by decadal variability, we refer to the variability band of 30–40 years. In particular, we explored the potential relation of AMO to decadal drought variability. We introduced a prediction model for future drought evolution based on the established causal relationship between AMO and SPEI. A similar concept for developing a prediction model has also been applied in several studies (Chylek et al., 2014; Krokos et al., 2019). Apart from this, we also examined the drivers of interannual drought variability over the AP.

In light of the above, the objectives of this study are: (a) to analyze spatiotemporal drought variability over the AP; (b) to identify the drivers of drought variability from annual to decadal scale; and (c) to explore the potential relation of AMO to decadal drought variability and develop a prediction model for future drought evolution over the AP.

Data Set and Methods Data Sets

We analyzed the following gridded data sets: the Global Precipitation Climatology Centre (GPCC) precipitation data set at 0.5° for 1980–2019 (Becker et al., 2013), Climate Research Unit (CRU) precipitation and temperature data sets at 0.5° for 1951–2020 (Harris et al., 2014), the Fifth Generation of ECMWF Reanalysis (ERA5; hereafter ERA) precipitation and temperature data sets at 0.25° for 1951–2020 (Hersbach et al., 2020), and Multi-Source Weighted-Ensemble Precipitation (MSWEP) precipitation data set at 0.25° for 1980–2020 (Beck et al., 2017). To quantify the droughts and their variability and trends, we utilized the Standardized Precipitation Evapotranspiration Index (SPEI) (Vicente-Serrano et al., 2010). The SPEI considers not only precipitation but also potential evapotranspiration (PET), which is derived from temperature. Unlike the SPEI, the Standardized Precipitation Index (SPI) is derived solely from precipitation (Mckee et al., 1993). Given the prevailing high temperature and meager precipitation over the AP, the SPEI is a more suitable index for regional drought analysis. We also examined the evolution of SPI to show the drought pattern if only precipitation is considered for drought assessment. Further details on the computational methodologies for the SPEI and SPI are provided in Text S1 of Supporting Information S1.

Most of our analysis, particularly pertaining to the SPEI variations, was based on the ERA data sets. While the CRU data sets also included both precipitation and temperature, our analysis showed that the CRU precipitation data over the southwest AP from the late 1990s did not show any variations commensurate with those observed at stations (Figure S1 in Supporting Information S1). However, the precipitation trends from the ERA data set reflected the station-level precipitation variation and showed a significant correlation with other precipitation data sets such as GPCC and MSWEP (Figures S1 and S2 in Supporting Information S1). The details of various indices used to represent the variability of large-scale drivers such as ENSO are presented in Table S1 of Supporting Information S1.

Methods

We applied Sen's slope method (Sen, 1968) and the Mann-Kendall test (Kendall, 1975; Mann, 1945) to determine monotonic trends in time series such as the SPEI and their statistical significance. We identified abrupt changes in SPEI variability by using the change point model (CPM) (Ross, 2015), which is commonly employed in hydroclimatic studies (e.g., Saharwardi et al., 2022; Shirvani, 2017). We further applied the Granger causality test (Granger, 1969) to identify the causal relationship between decadal SPEI variability and AMO.

We applied a singular spectrum analysis (SSA) (Vautard et al., 1992) to identify dominant periodicities, which involves decomposing the original time series of the area-averaged SPEI and AMO. We trained a linear recurrent formula (LRF) model (e.g., Sulandari et al., 2020) on the low-frequency AMO signal over the period of 1951–2020 and then applied it to simulate the future evolution of this signal until 2100. This combination of SSA and LRF is widely used in time-series forecasting (Sulandari et al., 2020).

Given the causal relationship between the AMO and SPEI, we aimed to predict drought variability in the future. Because a linear model was not fitted well, we used the nonlinear component of the support vector regression (SVR), a machine learning model, to quantify the future evolution of the SPEI until the end of the 21st century based on the simulated future AMO signal. SVR is a nonparametric regression algorithm that utilizes the radial basis function kernel to fit the best model to two variables (Javed et al., 2009; Smola et al., 2004). The details of SSA and SVR are summarized in Text S2 and S3 of Supporting Information S1.

We computed anomaly correlations not only for interannual time series but also between decadal signals of the AMO and SPEI. The statistical significance of the correlations was obtained by a two-tailed Student's t test and a bootstrap method based on 1,000 random realizations. In both methods, we accounted for autocorrelations. The statistical significance was found at a 95% confidence level unless mentioned otherwise.

Results Drought Variability Over the AP

As shown in Figure 1a, the area-averaged SPEI time series over the AP for the period of 1951–2020 exhibited interannual variations as well as a long-term trend. Interestingly, the SPEI demonstrated a decadal variation. The SPEI was predominantly positive for 1951–1997, which suggests a relatively wet regime, and was predominantly negative for 1998–2020, which suggests frequent and severe droughts. This sudden decadal shift in the SPEI after 1997 was also confirmed through CPM analysis. A few earlier drought studies for Oman (El Kenawy et al., 2020) and Iran (Nouri & Homaee, 2020) suggested similar findings. These results indicate that the recent drought acceleration over the AP is not just a long-term trend, but can also be attributed to decadal variability. As shown in Figure 1b, the annual SPEI for the period of 1951–2020 had a strong correlation of 0.85 with the concurrent SPI. However, while both indices showed decreasing trends, this trend was statistically significant only for the SPEI. The higher SPEI trend can be attributed to its computational procedure, which includes the temperature and thus the recent warming.

Figure 1. (a) Area-averaged Standardized Precipitation Evapotranspiration Index (SPEI) interannual variability over the Arabian Peninsula (AP) for the period of 1951–2020. The black line shows the 10-year moving average, whereas the dashed red line denotes the linear trend for the whole period. (b) Area-averaged Standardized Precipitation Index (SPI) and SPEI time series and trends over the AP for the period of 1951–2020.

Relative Impacts of Precipitation and PET on Drought

Figure 2a depicts the interannual area-averaged variability and trend of the precipitation and PET over the AP for 1951–2020. A significant and strong negative correlation of −0.71 was observed between the two variables. Such a relationship is expected over arid land that is consistently dry (Trenberth & Shea, 2005). Together, the decrease in precipitation and increase in PET favor the enhanced drought activity currently observed over the AP. However, while the increasing trend for PET and decreasing trend for precipitation are both statistically significant, the trend for the former is stronger than the latter. The strengthening PET trend can be attributed to not only the weakening precipitation trend but also the increasing temperature (not shown), to which it had a significant positive correlation of 0.82. The strong relationship between PET and temperature has also been reported in Europe and Australia (Estévez et al., 2009; Guo et al., 2017). Notably, we found a strong negative interannual correlation between the precipitation and PET variability before 1998. However, this coherent relationship has weakened since then, and both time series evolved in different directions. This can be attributed to a greater increase in PET relative to any contribution from the decreasing precipitation. These recent decadal changes are responsible for the accelerated drought activity over the AP in the last two decades. Our findings for the precipitation and PET trends over the AP are consistent with those of AlSarmi and Washington (2011) and Donat et al. (2014). However, this is the first study to discuss the relevance of the changes in these hydrometeorological variables for increasing droughts over the AP.

Figure 2. (a) Area-averaged interannual variability and trend of precipitation and potential evapotranspiration (PET) anomalies over the AP for the period of 1951–2020, where the dotted lines depict the 11-year moving average of the same. (b–d) Correlation between Standardized Precipitation Evapotranspiration Index (SPEI) and precipitation anomalies for annual, summer, and winter seasons. (e–g) Correlation between SPEI and PET anomalies for annual, summer, and winter seasons (values are multiplied by −1 for comparative assessment). The dotted value denotes statistical significance at the 95% confidence level.

We further compared the spatial distributions of the precipitation, PET, and SPEI over the periods of 1951–1997 and 1998–2020 and their respective differences (Figure S3 in Supporting Information S1). An increase in PET was observed throughout the AP during the recent period. The precipitation showed a continuous reduction over most of the AP, especially in the south. However, there were pockets in the north and southeast where the precipitation and PET trends were opposite. Therefore, we conjecture that the strong signal of enhanced drought severity over the southern AP in recent decades is likely due to the increasing local temperature in concert with the decreasing precipitation.

Notably, the SPEI showed a weaker correlation with the precipitation (Figure 2b) than the PET (Figure 2e) throughout the study period. Although both correlations were statistically significant, the correlation was −0.97 between the SPEI and PET and 0.79 between SPEI and the precipitation. The local SPEI showed a strong negative correlation with the PET across the AP, but the SPEI was only strongly correlated with the precipitation in the north, not in the south. On a seasonal basis, SPEI showed a strong correlation with precipitation in the winter and a weak correlation in the summer. The low correlation in the summer can be attributed to the low precipitation over the AP during this season, except in the south. In contrast, the SPEI strongly correlated with the PET in the summer. The strong negative correlation between the SPEI and PET in the summer can be attributed to the dominant temperature contribution compared to precipitation during this season. Thus, SPEI captures subtle seasonal changes in the local hydrometeorological variables, making it a useful index for regional drought studies over the AP.

Drought Teleconnections Annual Scales

To understand the drivers of interannual drought variability, we plotted the ENSO, NAO, and IOD indices along with the area-averaged SPEI (Figure 3a). NINO3.4 index showed an insignificant correlation with SPEI on annual and seasonal timescales (not shown). However, among the nine La Niña years that occurred in 1951–2020, most of the AP experienced negative SPEI in at least 7 years (i.e., 1955, 1970, 1973, 1998, 1999, 2007, and 2010) (Figure 3a). La Niña events cause anomalously strong Indian monsoons, which in turn enhance subsidence over the AP through anomalous overturning and upper atmospheric sinking, which leads to local warming and, therefore dryness (Attada, Dasari, Parekh, et al., 2019). However, we did not find such a consistent association between SPEI and concurrent El Niño years. This asymmetry explains the weak correlation between the area-averaged SPEI and NINO3.4. In contrast, NINO3.4 showed a significant correlation of −0.33 with SPI, which only considers precipitation (Figure S4 in Supporting Information S1). Precipitation in the AP is known to increase anomalously during concurrent El Niño years (Abid et al., 2018; Dasari et al., 2021). However, it is noteworthy that ENSO did not influence SPEI over the AP as it did SPI. Our analysis also suggested that Dipole mode index (DMI) (a standard index for defining the IOD) did not strongly influence SPEI. This contradicts the results of several studies (e.g., Chakraborty et al., 2006) that have suggested that strong IOD events modulate precipitation over the AP. Even the seasonally stratified analysis shows that the SPEI over the AP, in general, is not significantly associated with the variations of the ENSO or/and IOD in most of the seasons, except for a moderate correlation with ENSO in JJA and IOD in DJF (Table S3 in Supporting Information S1). Importantly, SPEI showed a moderate correlation of 0.22 with NAO (Figure 3a), which is significant at the 90% confidence level from the two-tailed Student's t test. This correlation can be attributed to the impact of NAO on both the temperature and regional circulations, as discussed previously by Saeed et al. (2022). This correlation can be attributed to the impact of NAO on both the temperature and regional circulations, as discussed previously by Saeed et al. (2022).

Figure 3. (a) Interannual variability of Standardized Precipitation Evapotranspiration Index (SPEI) and large-scale circulations (ENSO, NAO, and DMI) for the period of 1951–2020. (b) Decadal variability of SPEI (solid line) and Atlantic Multidecadal Oscillation (AMO) (black line) periodicity for the period of 1960–2016.

Decadal Scales

To highlight the decadal variations of the drought over the AP, we plotted an 11-year running mean for the time series of the area-averaged SPEI over the AP along with a similarly smoothed time series of the AMO (Figure 3b). Despite the strong interannual variability of SPEI (Figure 3a), it also exhibited a long-term decadal periodicity. Figure 3b indicates that SPEI was continuously positive until nearly 2000 and then changed to a negative phase, which was also observed in the unfiltered time series (Figure 1a). Meanwhile, AMO has been in a positive phase since the late 1990s. Critically, the smoothened SPEI time series showed a strong negative correlation of −0.79 with the concurrent AMO signal that was statistically significant at the 99% confidence level after accounting for autocorrelations. Based on these results, we deduce that the decadal propensity of droughts over the AP since the late 1990s is due to the positive phase of the AMO. The lead correlations of AMO with decadal SPEI exhibited an even stronger negative correlation of −0.90 that peaked at lag 4. Our Granger causality analysis test applied to the two time series further demonstrated that AMO is a significant causal factor for multidecadal SPEI fluctuations. This strong relationship between AMO and droughts in the AP agrees with the results of earlier sub-regional studies focusing on the impact of AMO on the Red Sea SST and Oman (Ehsan et al., 2020; Krokos et al., 2019).

Recently, Syed et al. (2022) reported that PDO exhibits a strong correlation with interannual SPI from 1985 to 2020. In this context, we found that PDO has a significant correlation of 0.51 with the area-averaged decadal SPI over the AP for 1951–2020 (not shown). However, the correlation between PDO and SPEI was insignificant. This is because, on decadal time scales, the PDO is insignificantly correlated with the area-averaged precipitation as well as temperature (Figure S5 in Supporting Information S1). On the other hand, the AMO is significantly correlated with the 11-year running mean of the area-averaged temperature over the AP (Figure S6b in Supporting Information S1).

Drought Projections Based on AMO

The above findings strongly suggest that the persistent drought conditions over the AP in the last two decades are predominantly linked with the AMO cycle. Hence, we explored the relationship between AMO and SPEI to project drought conditions in the future. Figure 4a shows the 11-year running average of AMO (blue line) for the period 1861–2020, which indicates that AMO varied with a periodicity of 70–80 years. We also used SSA to reconstruct this smoothened AMO signal (dotted blue line), which showed very good agreement with the original time series. We then extended the reconstructed AMO up to the end of the 21st century by using the LRF model (Figure 4b). The projected AMO conforms to other future reconstructions of AMO obtained by different methodologies (e.g., Chylek et al., 2014, 2016). We used the same procedure to delineate the low-frequency variability in SPEI for the period of 1951–2020 and found that its evolution was opposite to that of the variability in AMO (Figure 4b). This out-of-phase relationship between the reconstructed signals conforms with the observation-based AMO and SPEI time series (Figure 3b), demonstrating the veracity of our reconstruction procedure. We then fit an SVR model for these two variables to the period of 1951–2020. The standard deviation and error (Figure S6a in Supporting Information S1) suggest a reasonable fit. Finally, we simulated the projected SPEI over the AP for 2021–2100 by applying the SVR model to the future AMO time series. By design, the projected SPEI exhibited a decadal to multidecadal periodicity (Figure 4b), and sustained co-variability with the concurrent AMO. The projected SPEI suggests that the drought severity will decrease in 2021–2040. Furthermore, SPEI is projected to be positive in 2040–2060, which suggests a wet phase albeit with a smaller magnitude relative to the current drought. Severe drought over the AP is again projected for 2060–2100.

Figure 4. (a) Actual (black line), Smoothed (blue line), and singular spectrum analysis-decomposed (dashed blue line) Atlantic Multidecadal Oscillation (AMO) for the period of 1861–2020. (b) Projected AMO (blue line) and Standardized Precipitation Evapotranspiration Index (SPEI) (red line) over the Arabian Peninsula until the end of 2100. The shaded area along the red line shows the standard error.

Summary and Conclusions

We analyzed various observation-based and reanalysis data sets to examine the drought variability over the AP and its drivers for the period of 1951–2020. Because the AP climate is primarily characterized by its temperature, and to a lesser extent, its precipitation, we selected the SPEI for analyzing the drought.

Our results indicated that droughts over the AP have increased in frequency and severity since 1998. These persistent drought conditions are strongly correlated with the evolution of AMO. However, the correlations of the decadal SPEI with indices of ENSO, IOD, and NAO are, respectively, 0.01, −0.25, and 0.20. All these correlations are statistically insignificant at 95% confidence level. This is understandable because decadal changes are generally associated with slow changes brought either by multidecadal drivers such as AMO, or anthropogenic climate change. These multidecadal drivers, in turn, also modulate the interannual drivers such as ENSO (Fedorov & Philander, 2000) and their impacts. Importantly, warming during the last two decades over the AP has been a more significant factor associated with drought intensification in 1998–2020 than the concurrent weakening of local precipitation. Notably, this recent increase in AP temperature is correlated with the positive phase of AMO (Figure S6b in Supporting Information S1). This is in agreement with the results of Ehsan et al. (2020), who focused on summer temperatures. Thus, the more severe drought conditions over the AP in recent decades are due to not just long-term trends generally attributed to global warming but also the current positive phase of AMO. Krokos et al. (2019) showed that AMO constrains the global warming signature of the Red Sea SST. A paleo-record from the Red Sea also indicates multidecadal variability of North Atlantic origin in the Red Sea SST (Felis et al., 2000). Our results, along with these studies, suggest the potential significance of AMO for the climatic evolution of the AP and the conditions of its marginal seas. Apart from this, historically, proxy records also witnessed multidecadal unprecedented droughts over the AP during the sixth century CE (Fleitmann et al., 2022).

In this study, we showed for the first time the nonlinear impact of ENSO upon the interannual variability of the drought over the AP. While La Niña events strongly correlated with anomalously negative SPEI, no such correlation was observed between SPEI and El Niño. Positive NAO events were also associated with droughts over the AP. We used a statistical model based on the relationship between SPEI and AMO to project drought conditions until the end of the 21st century. Our projections showed that the drought severity will decrease in the near future and increase after 2060. However, these projections are statistical in nature. The signal may include a memory of long-term temperature trends from the training period of 1951–2020 but no correlation with future exacerbation of anthropogenic warming, which is already projected to increase over the region (Dezfuli et al., 2022). Therefore, our projected SPEI over the AP provides the minimum threshold drought statistics that can be attributed to natural variability. Thus, these projections will be useful for planning long-term strategies and megaprojects such as the Saudi Green initiative.

Acknowledgments

This research work was supported by the Climate Change Center, an initiative of the National Center for Meteorology (NCM), Kingdom of Saudi Arabia (Ref No: RGC/03/4829-01-01). The authors acknowledge the ECMWF for providing ERA5 products, CRU-UEA for CRU products, NOAA for all the climate indices, NCM for station rainfall, MSWEP, and GPCC for gridded precipitation data sets.

Data Availability Statement

All the data sets used in this study are publicly available from the different open portals. The CRU rainfall data are obtained from CRU (Harris et al., 2014). The ERA5 data set is available to registered users at the Copernicus portal (Hersbach et al., 2020). The data sets to show the stations' rainfall variability are presented in Table S2 of Supporting Information S1. The other rainfall data sets were obtained from MSWEP (Beck et al., 2017) and GPCC (Becker et al., 2013). The AMO index data set we obtained from https://psl.noaa.gov/data/correlation/amon.us.long.data, while all other climate indices (PDO, NAO, NINO3.4, and DMI) from https://psl.noaa.gov/gcos_wgsp/Timeseries/. The authors have used Python (https://www.python.org/downloads/release/python-397/) and NCAR Command Language (NCL) (https://www.ncl.ucar.edu/Download/conda.shtml) software for plotting all the figures.

Erratum

The originally published version of this article contained errors in the affiliations. Md Saquib Saharwardi, Hari Prasad Dasari, Karumuri Ashok, and Ibrahim Hoteit should be affiliated with Physical Sciences and Engineering, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia, and Climate Change Center, National Center for Meteorology, Jeddah, Saudi Arabia. Vaneet Aggarwal should be affiliated with Electrical and Mathematical Science & Engineering, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia. The errors have been corrected, and this may be considered the authoritative version of record.