1 Introduction

Drought is considered as the most severe and least understood natural hazard, resulting in prolonged dry period in natural climate cycle, that can have serious impacts on crop yield, environmental conservation, socioeconomic development, infrastructure, health, and Gross Domestic Product (GDP) of a country (Kao and Govindaraju 2010; Zargar et al. 2011; Hao and Aghakouchak 2014; Zhang et al. 2015; Ullah et al. 2017). It is broadly defined as an extended period of below-average precipitation, leading to a severe water shortage in an ecological system (Gonzalez-Hidalgo et al. 2009). Palmer (1968) defined drought as a meteorological phenomenon that severely affects an area and causes moisture deficiency. The drought can be categorized into the following three major types: meteorological, hydrological, and agricultural. According to Wilhite (2000), drought can occur anywhere in the world and in any climatic region (arid, semi-arid, or humid), but their spatio-temporal characteristics vary significantly among regions. The interannual fluctuations in precipitation often place the arid regions at higher drought risk due to greater likelihood of abnormally low precipitation (Le Houérou 1996; Smakhtin and Schipper 2008). Since most of Pakistan lies in an arid climate zone where droughts have been a recurring event (Haider and Adnan 2014), the study of historical droughts will help in the delineation of drought-prone areas, and thereby, contingency plans can be developed by the government authorities to cope with the hazardous effects of this natural disaster (Adnan et al. 2018).

Owing to the socio-economic effects of droughts, many studies have been carried out across different regions of the world, and different indices and techniques have been employed to monitor the frequency, duration, and severity of drought (Zhang and Zhou 2015; Tabari et al. 2012; Pietzsch and Bissolli 2011). According to Mendicino et al. (2008), drought indices are reliable and useful tools for monitoring and forecasting drought conditions. Hisdal et al. (2001) studied the hydrological droughts using threshold-level approach and found that the variability in precipitation can explain the trends of drought occurrence, frequency, and duration. Narasimhan and Srinivasan (2005) developed two drought indices, Soil Moisture Deficit Index (SMDI) and the Evapotranspiration Deficit Index (ETDI), for the monitoring of agricultural drought using the Soil and Water Assessment Tool (SWAT) model. Adnan et al. (2018) applied different drought indices and reported that, among 15 different indices used, the Standardized Precipitation Index (SPI) (McKee et al. 1993), Standardized Precipitation Evapotranspiration Index (SPEI) (Vicente-Serrano et al. 2010), and Reconnaissance Drought Index (RDI) (Tsakiris and Vangelis 2005; Tsakiris et al. 2007) perform well in assessing drought conditions over Pakistan and adjacent areas. Other common drought indices that are being used extensively by the national hydro-meteorological organizations include the Palmer Drought Severity Index (PDSI) (Palmer 1965), Deciles Index (DI) (Gibbs and Maher 1967), China Z Index (CZI) (Wu et al. 2001), Crop Moisture Index (CMI) (Palmer 1968), and Surface Water Supply Index (SWSI) (Shafer and Dezman 1982).

Pakistan is ranked on the seventh position among the most affected countries by the impacts of extreme weather events during 1996 to 2015 (Kreft et al. 2013). Variability of droughts in Pakistan has captured the attention of many researchers in the recent past. Pasha et al. (2015) studied droughts over the Sindh Province of Pakistan and showed that human-induced warming has increased drought risk in the country. Adnan et al. (2018) evaluated the performance and efficiency of various drought indices over Pakistan using the data of 58 meteorological stations for the period 1951–2014. Their results obtained by most of the indices showed significantly increasing (towards wetness) trends. Adnan and Haider (2012) studied the classification and assessment of aridity in Pakistan for the period 1960–2009 and reported that the southern parts of the country are more vulnerable to drought conditions compared with the northern areas. The southern parts of Pakistan have also been identified as the most drought-vulnerable region by Anjum et al. (2010). Haroon et al. (2016) employed MODIS-based Drought Severity Index (DSI) to investigate drought conditions in Pakistan and found severe drought during 1998–2002 in the region. Their results further revealed that the MODIS-Terra NDVI values can be used as a tool to identify wet and dry conditions in large-scale studies; however, the technique is not suitable for regional scale studies, as it is hard to detect drought signal just from NDVI anomaly values alone. Adnan et al. (2016) conducted drought analysis on monthly, annual, and decadal time-scales using different indices during 1951 to 2010 in the South Central Asia (SCA) region, with particular focus on Pakistan. They found two major drought periods, i.e., 1971 and 2000–02, where 2001 was reported as an extremely dry year in the SCA region. The drought onset, end, and severity are difficult to predict, and it still poses a challenge, especially over the rainfed region of Potwar Plateau, which needs to be addressed to get a more insight on this costly natural event.

The Potwar Plateau of Pakistan (PPP) lies in the dominant precipitation region of Pakistan (see, for instance, Latif and Syed 2016; Latif et al. 2017), which receives precipitation not only during summer season but also in winter through southwest monsoon and western disturbance weather systems, respectively. Agricultural productivity of PPP relies heavily on adequate water supply, which depends solely on the timely rainfall (Rashid and Rasul 2011). A slight variability in the rainfall and water stress at any crop phonological stage can affect the grain yield (Cakir 2004). Although many studies (e.g., Sajjad and Ghaffar 2018; Ahmad and Hussain 2017; Ali et al. 2016) have investigated climatic extremes related to temperature and precipitation in Pakistan, drought assessment and its related large-scale atmospheric dynamics as well as the future projections using Regional Climate Model (RCM) simulations for major cropping seasons have not been studied over the PPP. In this study, we aim to fill the above-mentioned gaps by investigating the past and future drought as well as related dynamics in the Potwar Plateau.

The structures of the paper are organized as follows: Study area, data, and methods are described in Section 2. The results and discussions are presented in Section 3, which include the analysis of drought frequency, severity, and trends; large-scale dynamics associated with extremely dry/wet cases; and future drought projections using RCMs. Finally, the summary and conclusions of this study are presented in Section 4.

2 Data and methods

2.1 Study area

The Potwar Plateau (32.5°–34°N and 72°–74°E) is located in the north-east Pakistan with a total area of about 5000 miles2 (Fig. 1). The Potwar region includes four districts—Jhelum, Chakwal, Rawalpindi, and Attock—and Islamabad, the capital territory (Rashid and Rasul 2011). The climate of most of the area is semi-arid to humid with annual rainfall ranging from 900 to 1900 mm (Adnan et al. 2009; Adnan and Khan 2009). Generally, two major weather systems, i.e., southwest monsoon (Latif and Syed 2016) and western disturbances (Ahmed et al. 2019a, b; Hussain and Lee 2009), cause rainfall over the PPP during the summer (July–September) and winter (December–March) seasons, respectively. The mean annual maximum and minimum temperature ranges between 26 °C and 13 °C, respectively (Adnan et al. 2017). Kharif (May–September) and Rabi (October–April) are the two major crop growing seasons in the PPP. The spatial distribution pattern of the mean seasonal (cropping seasons) and annual precipitation and temperature over Pakistan is shown in Fig. 2. The sowing and harvesting time of Kharif crops (rice, maize, cotton, soybean, jowar, moong, ground nut, etc.) are May–June and October–November, whereas Rabi crops (wheat, onion, tomato, potato, mustard, carrot, oat, isabgol, barley, etc.) are usually sown during November–December and harvested in March–April. The crop yield of the PPP contributes to about 10% of the total agriculture production (Ashraf 2004). The summer monsoon rainfall alleviates moisture stress conditions during the Kharif season and fulfills the water requirements for Rabi crops (Adnan et al. 2018). The spatio-temporal variability of precipitation during both seasons can affect negatively the GDP of the country, as agriculture sector accounts for about 24% of GDP alone (Latif and Syed 2016; IUCN 2009).

Fig. 1
figure 1

Study area map of Potwar Plateau of Pakistan. Black circles show the location of PMD stations, whereas the shading represents the elevation in meters

Full size image
Fig. 2
figure 2

Left panels (a–c): precipitation climatology (units: mm/month) during Kharif, Rabi, and annual time-scales, respectively, for the period (1981–2010) using GPCC gridded datasets. Right panels (d–f): same as left panels except for temperature (units: °C) using CRU datasets. Rectangular boxed region in panel a shows the study region

Full size image

2.2 Observed and reanalysis datasets

In order to examine drought conditions over the PPP in the past three decades (1981–2010), six monthly observational temperature and precipitation datasets at 0.5 × 0.5° horizontal resolution are used in this study. These datasets are obtained from the following different sources: (1) Climate Data Processing Centre (CDPC), Pakistan Meteorological Department (PMD); (2) Global Precipitation Climatology Centre (GPCC) V6 (Schneider et al. 2015); (3) University of East Anglia, UK Climatic Research Unit (CRU) TSv.4.01 (Harris et al. 2014); (4) Asian Precipitation-Highly Resolved Observational Data Integration Towards Evaluation of Water Resources (APHRODITE) V1101 (Yatagai et al. 2012) by Research Institute for Humanity and Nature (RIHN) and Meteorological Research Institute of Japan Meteorological Agency; (5) University of Delaware (UDEL) global gridded high-resolution station (land) data (Willmott and Matsuura 2005); and (6) NAA Precipitation Reconstruction over Land (PREC/L) (Chen et al. 2002). Era-interim reanalysis data product (Ma et al. 2018) by the European Center for Medium-range Weather Forecast (ECMWF) is also used to check for consistency with station and observed datasets.

In order to study the large-scale circulation anomalies associated with Extremely Dry (ED) and Extremely Wet (EW) events, also referred as drought and wetness conditions, NCEP (National Center for Environmental prediction)/NCAR (National Center for Atmospheric Research) reanalysis data of geopotential height, u & v winds, and vertical velocity (ω) with 2.5° × 2.5° spatial resolution (Kalnay et al. 1996) are used.

2.3 Model datasets

The fourth version of the Rossby Centre regional atmospheric model (RCA4) at the Swedish Meteorological and Hydrological Institute (SMHI) and RegCM4.4 at the Abdus Salam International Center for Theoretical Physics (ICTP) RCM outputs with different Global Climate Models (GCMs) as Lateral Boundary Conditions (LBCs) are obtained from COordinated Regional climate Downscaling EXperiment (CORDEX) framework for South Asian domain with 0.44° × 0.44° (approximately 50 × 50 km grid spacing) horizontal resolution (Table 1). The CORDEX experiment is developed by the World Climate Research Program (WCRP) which aims to produce high-resolution regional climate projections for the future climatic analysis at regional scale (Giorgi et al. 2009). RCM simulations with ERA-Interim as LBCs (Dee et al. 2011) are used for model evaluation of the recent past climate. The historical simulations (1961–2005) are based on the assumption of anthropogenic gas emission and aerosol ejection into the atmosphere (Latif et al. 2018). Whereas, future (2006–2100) simulations are based on two Representative Concentration Pathway (RCP) scenarios (Moss et al. 2008), i.e., RCP4.5: medium-to-high CO2 emission scenario and RCP8.5: high CO2 emission scenario. In this study, we used both RCP4.5 and RCP8.5 scenarios which represent the categories of “medium” and “high” greenhouse gas emission scenarios, respectively, and prescribe the future radiative forcing level of 4.5 and 8.5 W/m2, respectively, by the end of the twenty-first century.

Table 1 List of CORDEX South Asia regional climate model (RCM) experiments
Full size table

2.4 Methodology

The performance of observational temperature and precipitation gridded datasets is evaluated first for both Kharif and Rabi seasons as well as on annual time-scales using the Taylor diagram (Taylor 2001). A Taylor diagram can provide concise statistical comparison between observed or reanalysis field and reference data in terms of analyzing the ratio of their variances, correlation coefficient, and root-mean-square difference. The best performing observed data is then used to check for consistency with station data for the study of droughts over the region. The percentile grading criteria is employed by averaging the monthly precipitation of all five meteorological stations that lie in the Potwar region (Fig. 1), to identify the EW and ED cases for both cropping seasons for the study period.

The percentile rank approach is given and calculated as:

$$ R=\frac{P}{100\left(N+1\right)} $$
(1)

Where R is the rank order, percentile rank is represented by P, and N is the total number of values in the distribution.

We further applied two drought indices viz. SPI and RDI, to categorize the drought severity and also to check for consistency with the EW/ED cases as identified using percentile rank approach. SPI is calculated at different 3-, 6-, 9-, and 12-month aggregation time-scales. RDI is based on cumulative values of both precipitation and potential evapotranspiration (PET), where the values range between 1.5 to 3.0 and − 1.5 to − 3.0, which represent the EW and ED conditions, respectively. The percentile grading, SPI, and RDI criteria are given in Table 2.

Table 2 The wet and dry classes and grades based on percentile score (Kant et al. 2014), SPI, and RDI (Adnan et al. 2015) criteria
Full size table

The composite-anomaly patterns of geopotential height and winds (850 hPa), vertical velocity (500 hPa), precipitable water, vertically integrated (surface to 300 hPa) moisture transport (see, for instance, Latif et al. 2018 and Latif et al. 2017), and integrated moisture flux convergence/divergence (Ullah and Shouting 2012) are presented to study the large-scale atmospheric dynamics associated with EW and ED cases. Monthly anomalies are calculated by subtracting climatological monthly means from the data.

To study the future drought conditions over the PPP, the RegCM4.4 and RCA4 precipitation simulations, with Era-Interim and different GCMs as LBCs (Table 1), are compared with reference (station) data for selecting the best model for the analysis. The period 1981–2005 is used as the baseline (reference) period for model evaluation, whereas the time slice 2006–2100 is used for the future under RCP4.5 and RCP8.5 scenarios. RCM simulations driven by ERA-Interim as LBCs are used for assessing the model performance in the recent past climate. A non-parametric statistical test, Sen’s slope estimator (Sen 1968), is used to identify the trend in drought indices (SPI and RDI), whereas Mann-Kendall test (Mann 1945; Kendall 1975; Gilbert 1987) is employed to ascertain the presence of statistical significance (5% significance level) of the trend in the time series.

3 Results and discussions

3.1 Performance of observed and reanalysis datasets

The pattern statistics between reanalysis/observed and PMD rain-gauge measurements for the normalized mean (1981–2010) precipitation and temperature for seasonal (Kharif and Rabi) and annual time-scales over the PPP is presented in the Taylor diagram (Fig. 3). The PMD station data are used as the reference dataset to evaluate the performance of reanalysis (Era-Interim) and observed (GPCC, CRU, APHRODITE, UDEL, and Precip/L) datasets over the study area. Taylor diagram shows 2-D plot based on three statistical parameters, i.e., Root Mean Square Error (RMSE), standard deviation (amplitude of the variations), and correlation coefficient. These parameters provide the best possible evaluation results for the selection of appropriate datasets. Based on the statistical comparison between six observational/reanalysis and station datasets, our results show that the GPCC and CRU (for precipitation) and APHRODITE (for temperature) approximate well with the station datasets during both Kharif and Rabi seasons. It is also noted that the Era-interim reanalysis does not perform well for both precipitation (Fig. 3a–c) and temperature (Fig. 3d–f) cases. The correlation of GPCC (CRU) precipitation (temperature) lies between 0.9 and 0.95, with smaller RMSE. For temperature, although APHRODITE seems more realistic and suitable (the correlation is above 0.9, standard deviation is near to reference line with small RMSE) for the analysis, the dataset is not used for further analysis due to the limited time period constraint. Ahmed et al. (2019a, b) tested the performance of four widely used gauged-based gridded precipitation datasets in the semi-arid, arid, and hyper-arid regions of Balochistan Province of Pakistan and reported that the GPCC performs much better compared with other gridded datasets. Cheema and Hanif (2013) also evaluated the performance of GPCC datasets over the Punjab Province of Pakistan and found strong correlation (R2 = 0.93) with in-situ area-weighted precipitation. Our results regarding the performance of GPCC are in good agreement with previous findings; we have therefore chosen the GPCC in this study to check for consistency with the station (reference) data.

Fig. 3
figure 3

Left panels (a–c): Taylor diagram representing a statistical comparison of gridded observed/reanalysis precipitation (units: mm/month) datasets with PMD station data for Kharif, Rabi, and annual time-scales, respectively, for the period (1981–2010). Right panels (d–f): same as left panels except for temperature

Full size image

By employing percentile score, SPI, and RDI criteria (Table 2) and assigning different grades, we found that frequency of classes differs significantly between EW and ED years (Table 3). It is observed that SPI-based cases of precipitation deficit and surplus are almost equal in Kharif (5 dry and 5 wet) and Rabi (3 dry and 4 wet) seasons; however, surplus cases on annual time scale are slightly higher than deficit (3 dry and 5 wet). The results of RDI show that the deficit cases in Kharif are higher than surplus (4 dry and 3 wet), whereas the deficit cases are higher in Rabi compared with Kharif (5 dry and 4 wet).

Table 3 Details of wet and dry classes of precipitation with their respective grades based on percentile score, SPI, and RDI criteria
Full size table

On annual basis, RDI shows higher number of deficit cases than surplus (5 dry and 3 wet). The seasonal (Kharif and Rabi) and annual analyses also identify extreme surplus and deficit cases, which have been used in Section 3.3 to see if these can explain the possible linkage with the large-scale dynamics.

3.2 Analysis of drought intensity and trends

The percentile (10th and 90th) rank approach and two drought indices, i.e., SPI and RDI, are used to determine the ED and EW cases over the PPP during 1981–2010 for both seasonal (Rabi and Kharif) and annual bases (Fig. 4). The percentiles are calculated both for PMD station data as well as GPCC gridded precipitation data. This approach is quite useful, and therefore, it is being used widely by many researchers to bifurcate the dry and wet cases over different parts of the world (Hao et al. 2016; Raible et al. 2017; Sajjad and Ghaffar 2018). In Fig. 4, the driest cases (at 10th percentile) are observed in years 1986, 1987, 2002, and 2009, whereas the wet cases at 90th percentile are noted during 1987, 1994, and 1997. As discussed earlier, SPI and RDI are the two renowned drought indices that have been used for identifying the drought intensity as well as for the consistency measures with percentile method over the same period (1981–2010). The choice of an appropriate drought index is critical for the effective monitoring of drought over the region. To study the drought over Pakistan, Adnan et al. (2018) recommended SPI and RDI as the best performing drought indices over the country. Both of the indices are highly correlated to each other as well as with percentile for both PMD station and GPCC gridded datasets. High interannual variability of the two drought indices has been observed for both Kharif and Rabi seasons, which may be attributed to the high interannual variability in the mean rainfall during summer (monsoon) or winter (western disturbances) seasons. The deficiency in seasonal rainfall may result in a drought situation over the PPP. It is also observed that the SPI index shows a positive slope (0.008) on both seasonal and annual time-scales for GPCC datasets (Fig. 4a,b), whereas negative (positive) is observed in Kharif (Rabi) season for station data. Moreover, the slope of RDI is positive (0.002) for Kharif and negative for both Rabi (− 0.010) and annual (− 0.009) bases. The positive and negative slopes of SPI and RDI drought indices represent wetter and drier episodes over the Potwar region.

Fig. 4
figure 4

(a): Time series comparison of SPI (blue) and RDI (red) drought indices for Kharif season for the period (1981–2010) using PMD (thick blue line) and GPCC (dotted blue line) datasets. Vertical bars on the secondary axis indicate percentile score using PMD (black) and GPCC (gray) datasets, whereas the horizontal dotted lines represent the 10th and 90th percentiles. The values in parenthesis show slope of trend line of SPI and RDI using Sen’s slope estimator. The trend in time series is checked at 5% significance level using MK test. (b,c): Same as (a) except for Rabi and Annual cases, respectively

Full size image

SPI is further calculated at 3-, 6-, 9-, and 12-month aggregation time-scales on seasonal (Kharif and Rabi) and annual bases (Fig. 5). Drought intensity usually increases with increasing time-scales; therefore, an intense drought may not be determined on shorter timescale. Furthermore, SPI 3 is considered the most appropriate for short-term rainfall deficit or surplus cases, SPI 6 and SPI 9 are suitable for the monitoring of meteorological and agricultural droughts, and SPI 12 is more effective for water resource management (Pashiardis and Michaelides 2008; Raziei et al. 2009; Adnan et al. 2015). High interannual variability of SPI is again observed during Kharif, Rabi, and annual basis; however, the Kharif season experiences the highest drought intensity (Fig. 5a). The negative slope of SPI on both seasonal and annual time-scales clearly indicates the decreasing trends (towards drought conditions). Due to the high intra-seasonal rainfall variability of PPP and its unique geographical location (receives highest amount of rainfall both in summer (Kharif) and winter (Rabi) seasons), drought period usually remains for a shorter time period and less intense as compared with other parts of the country. Since the Potwar region lies in the core precipitation zone of Pakistan, the drought or wetness events could be linked with the large-scale atmospheric dynamics to see if any physical linkages can be established.

Fig. 5
figure 5

(a): Time series comparison of SPI 3-, 6-, 9-, and 12-month aggregation time-scales for Kharif season for the period 1981 to 2010 using station datasets. The values in parenthesis show slope of trend line of SPI using Sen’s slope estimator. The trend in time series is checked at 5% significance level using MK test. (b,c): Same as (a) except for Rabi and Annual cases, respectively

Full size image

3.3 Drought large-scale dynamics

In this section, the large-scale atmospheric dynamics for EW and ED cases over the PPP are presented for both Kharif and Rabi cropping seasons, as well as on annual time-scale as follows:

3.3.1 Geopotential height and wind anomalies

Composites of geopotential height and wind anomalies in the lower troposphere (850 hPa) for EW cases on seasonal (Kharif and Rabi) and annual time scales are shown in Fig. 6a–c. It is observed from Fig. 6a that a strong positive height anomaly (> 6 m) is located over the Tibetan plateau during EW-Kharif season. As a result of this anticyclone (positive height anomaly), a strong convergence zone is developed along the foothills of Himalayas. This convergence zone plays a significant role in transporting lower-level moisture-laden winds to blow from Bay of Bengal towards Pakistan, causing rainfall during the EW-Kharif season over the PPP. Another positive height anomaly (5 m) over north and a weak negative anomaly (< 1 m) over the south of Pakistan together act as positive feedback to strengthen the northward flow. It should also be noted that the PPP is dominated by the southwest summer monsoon season during Kharif cropping months. During EW-Rabi season (Fig. 6b), a strong east-west-elongated negative height anomaly (<− 6 m) (cyclone) is noticed over the southwest of Pakistan, north-east Arabian Sea, Kathiyawar Peninsula, and north-west India. This negative height anomaly, associated with the eastward moving westerlies, is mainly responsible for transporting moisture from the Arabian Sea branch to the PPP during EW-Rabi season. We may conclude that during the EW-Kharif (EW-Rabi) season, Tibetan anticyclone (westerly flow) plays a crucial role in moisture incursion from the Bay of Bengal (Arabian Sea) for wet conditions over the study region. The positive and negative patterns of height anomalies can also be seen on annual time-scale (Fig. 6c).

Fig. 6
figure 6

Left panels (a–c): Geopotential height (shaded, units: meters) and wind (vector, units: m s−1) composite anomalies at 850 hPa for EW cases during Kharif, Rabi, and annual time-scales. Right panels (d–f): same as left panels except for ED cases

Full size image

The geopotential height and wind composite anomalies for ED cases on seasonal and annual time scales are shown in Fig. 6d–f. The results are the same as for ED cases, but with a reverse sign of anomalies. The outflow of winds from the PPP is observed during Kharif (Rabi) season caused by the strong positive (negative) height anomalies over most of the region (Fig. 6d,e). An outward flow of winds from the Indo-Pakistan subcontinent, during ED-Kharif/Rabi cases, is mainly responsible for dry condition over the region, which can also be seen on annual time scale (Fig. 6f).

3.3.2 Moisture transport and moisture flux convergence/divergence anomalies

Results of the composite anomalies of Vertically Integrated Moisture Transport (VIMT) and moisture flux convergence/divergence for EW cases on seasonal (Kharif and Rabi) and annual time-scales are presented in Fig. 7a–c. The integrated moisture transport (surface–300 hPa) from the NCEP/NCAR reanalysis is calculated by integrating moisture fluxes from the pressure level data. In Fig. 7a, it is observed that the strong moisture flux convergence zones (negative anomalies; <− 2 kg m−2 s−1) during the EW-Kharif season are located over the central Pakistan, along the southern slopes of central Himalayas, north Bay of Bengal, and along the north of Western Ghats. As a result of these convergence zones over the Indo-Pakistan subcontinent, particularly along the Himalayan foothills, moisture is transported mainly from the Bay of Bengal to Pakistan. It is mentioned that the development of this moisture convergence zone is associated with the anticyclone located over the Tibetan plateau (Fig. 6a). Latif and Syed (2016) and Syed et al. (2010) also described the presence and significance of this convergence zone during the south Asian summer monsoon season. The results of EW-Rabi season (Fig. 7b) clearly show the presence of moisture flux convergence (negative anomaly; about − 2 kg m−2 s−1) extending from north Pakistan along the Himalayas to the north Bay of Bengal. Another moisture convergence zone has been noticed over the north Arabian Sea. These negative anomalies are mainly responsible in advecting moisture from the Arabian Sea branch to the PPP, causing wet conditions over the region. Since eastward moving westerly troughs occasionally prevail during Rabi winter crop season, the moisture is therefore transported mainly from the west to the PPP. During annual time-scale (Fig. 7c), deeper moisture convergence progression has been observed from Bay of Bengal compared with the Arabian Sea. It can be concluded that the presence of moisture convergence anomalies over most of the Indo-Pakistan subcontinent, during seasonal and annual time-scales, is responsible in transporting moisture from Bay of Bengal and Arabia Sea, resulting in EW conditions over the Potwar region of Pakistan.

Fig. 7
figure 7

Left panels (a–c): Composite anomaly of vertically integrated (surface to 300 hPa) moisture transport (vector, units: kg m−1 s−1) and vertically integrated moisture flux convergence/divergence (shaded, units: kg m−2 s−1) for EW cases during Kharif, Rabi, and annual time-scales, respectively. Right panels (d–f): same as left panels except for ED cases

Full size image

The VIMT and moisture flux convergence/divergence composite anomalies for ED cases on seasonal and annual time scales are shown in Fig. 7d–f. The positive moisture flux anomalies (divergence; > 1.5 kg m−2 s−1) are present during the ED cases, which prevents the moisture transport form the Bay of Bengal and Arabian Sea to the PPP.

3.3.3 Precipitable water and vertical velocity anomalies

The composite anomalies of precipitable water for EW and ED Kharif, Rabi, and annual cases are presented in Fig. 8a–c and Fig. 8d–f, respectively. The precipitable water (or precipitable water vapor) is the depth of water vapor in a vertical column of atmosphere over unit cross-sectional area, if all the water vapor in that column were condensed as precipitation. It is a measure of atmospheric humidity which is an investigating tool of rainfall occurrence (Ferraro et al. 1998; Bhatia 2000). It is observed from Fig. 8a that a positive precipitable water anomaly (about 3 mm) with larger spatial extent is present over Pakistan and its adjoining areas during the EW-Kharif season. The presence of this positive anomaly can be considered as one of the main parameters of moisture availability and precipitation over the region. This positive anomaly is extending from Bay of Bengal to Pakistan, which is associated with the propagation of northward moving depressions from the Bay of Bengal during the Kharif (summer monsoon-dominated time) season. In Rabi season (Fig. 8b), the east-west elongated positive precipitable water vapor anomaly mainly extends from northwest Pakistan towards Tibetan Plateau, with center over Pakistan. The position and spatial structure of this positive anomaly indicates its association with the eastward moving westerlies during the Rabi (affected by the western disturbances) season. Therefore, the water vapors transported from the Arabian Sea accumulate over the study region and cause precipitation during the season. The results of EW annual case show that almost the entire Indo-Pakistan subcontinent is dominated by the strong positive precipitable water anomaly, with more than 2 mm/month is observed over the eastern Indian Peninsula (partially over the eastern Ghats), Arabian Sea, and Pakistan (Fig. 8c). We may conclude that the higher precipitable water, associated with southwest summer monsoon (western disturbances), transported from the Bay of Bengal (Arabian Sea) directly contributes to Kharif (Rabi) precipitation amount. During ED Kharif, Rabi, and annual cases, the negative anomaly (− 2 mm) of precipitable water is dominated over most of the Indo-Pakistan subcontinent and its surrounding areas, contributing decisively in decreasing the precipitation amount over the region (Fig. 8d–f).

Fig. 8
figure 8

Left panels (a–c): Precipitable water composite anomalies (units: mm) for EW cases during Kharif, Rabi, and annual time-scales, respectively. Right panels (d–f): same as left panels except for ED cases

Full size image

The composite anomalies of vertical velocity (omega) are presented at 500 hPa pressure level for EW (Fig. 9a–c) and ED (Fig. 9d–f) cases during Kharif, Rabi, and annual time-scales. A strong positive (negative) anomaly over northern (southern) half of the country is clearly visible in Fig. 9a. A positive anomaly is extended in northeast over Tibetan plateau and some parts of India and in the northwest over Afghanistan and Gulf region. On the other hand, a strong positive anomaly is still encompassing northern parts of the country extending towards south-western parts of the country (Fig. 9d), although its north-eastern and northwest extensions are no more there in this case. Important dominant factor is absence of strong negative anomaly in dry periods. It is now visible only in eastern part of southern half of the country with a weak intensity. In EW-Rabi case, a well-marked strong negative anomaly is clearly dominant over the country (Fig. 9b). On the other hand, a pronounced positive anomaly is distributed over the whole country during the ED-Rabi case (Fig. 9e). On annual time scale, almost same situation is prevailing as observed in EW and ED cases for both Kharif and Rabi seasons. Dominance of a strong negative anomaly over the country is a major feature of Fig. 9c, and it is absent in Fig. 9f, which shows a stronger positive anomaly over northern parts of the country. In dry periods, absence of channels of negative anomalies from Bay of Bengol and Arabian Sea in Kharif and Rabi seasons is highlighting the role of strong monsoon components and strong western disturbances, respectively, in dynamics of droughts over the region. In other words, a weak monsoon supports drought in Kharif and less precipitation in winter season supports drought in Rabi season. It is also noticeable that the role of Tibetan plateau during monsoon season is very crucial in determining precipitation variability over Pakistan (Wang et al. 2019).

Fig. 9
figure 9

Left panels (a-c): Omega (vertical velocity) composite anomalies (units: ×10−2 Pa S−1) at 500 hPa for EW cases during Kharif, Rabi, and annual time-scales, respectively. Right panels (d–f): same as left panels except for ED cases

Full size image

Although some earlier studies explain the role of large-scale dynamical and thermo-dynamical characteristics of different circulation features over south Asia/Pakistan (see, for instance, Ahmed et al. 2019a, b; Latif et al. 2017; Syed et al. 2010), none of the studies explicitly investigated the role of dynamical mechanisms associated with drought and wetness conditions during two major cropping seasons (Kharif and Rabi) over the PPP. In this study, we have identified a clear role of Tibetan Plateau during Kharif (summer monsoon dominated) season, which plays an important role in transporting moisture from the Bay of Bengal and causes wet conditions over the region. Latif and Syed (2016) and Syed et al. (2010) have explained the role of the Tibetan Plateau; however, their results show that although weather-producing low pressure systems come from the Bay of Bengal, the moisture is mainly transported from the Arabian Sea during summer (July–September) and rainy seasons. Our results in relation to the identification of positive height anomaly (anti-cyclone) over Tibetan Plateau and associated moisture transport from the Bay of Bengal during Kharif season is a new contribution to the existing literature. Since, this study explains a physical linkage between large-scale dynamics and drought/wetness conditions, the same association could be studied in the future using regional or global climate models to see why the trends of drought severity are increasing in the future. This will be helpful in understanding the large-scale dynamics of future projected droughts.

3.4 Analysis of future drought characteristics

3.4.1 Model selection and evaluation

In this section, future drought conditions are studied by using CORDEX south Asia RCM (RegCM4.4 and RCA4) simulations. We first selected the best performing model datasets using RCM outputs driven by different GCMs as LBCs (Table 1). The selection of model is based on a statistical comparison of precipitation between the model outputs for the historical run (1981–2005) and PMD station datasets (observations) over the PPP (Fig. 10). It can be clearly seen from the figure that the RegCM4.4 model, driven by CCMA-CanESM2 GCM as boundary conditions, best correlates with the observations. This model has a relatively high correlation and smaller RMSE, and the standard deviation (STDV) is closer to the observations. Although RCM4.4 (run with CNRM-CM5 GCM as LBCs) has mean closer to the station data, the RMSE and STDV do not approximate well with observations. In addition, the mean of RegCM4.4 (GFDL-ESM2M) is noted within the range of half of STDV, but the spread of maximum and minimum values is very large. It is also noted that RCA4, which run with CCMA-CanESM2, IPSL-CM5A-MR, and CSIRO-MK3.6.0 LBCs, do not perform well with the observations over the Potwar region of Pakistan.

Fig. 10
figure 10

Statistical comparison of precipitation (units: mm/month) between RCM simulations and station datasets for the baseline period 1981–2005. Black horizontal thick line shows mean precipitation using station data, whereas the dashed horizontal lines show mean plus/minus half standard deviation (STDV). Black circles with error bar show mean and mean plus/minus half STDV of RCMs driven by different GCMs as LBCs. Up and down triangles show maximum and minimum values, whereas asterisk represents the RMSE

Full size image

The performance of selected RCM model, i.e., RegCM4.4 with CCMA-CanESM2 GCM as LBCs, with so-called perfect boundary conditions (ERA-Interim) and GPCC precipitation is further evaluated for the present (1981–2005) climate over the PPP (Fig. 11). Figure 11 (top row) shows the spatial pattern of observed (GPCC) precipitation climatology for Kharif, Rabi, and annual time-scales, respectively, over the region. The Core Monsoon Region of Pakistan (CMRP, Latif et al. 2017) receives the highest amount of precipitation during the Kharif season (Fig. 11a), as the region is influenced by the penetration of weather-producing low pressure systems from the Bay of Bengal (Latif and Syed 2016; Faisal and Sadiq 2012). In the Rabi season, the CMRP and its adjoining eastern regions again received the highest amount of precipitation, caused by the eastward moving western disturbances (Fig. 11d). The annual time-scale shows the contribution of precipitation of both Kharif and Rabi seasons (Fig. 11g). It is observed from Fig. 11 (middle row) that the spatial patterns of seasonal and annual mean precipitation in RegCM4.4 (driven by Era-Interim LBCs) is consistent with the observations (GPCC). However, the model simulates slightly more precipitation (particularly in Rabi season) over the north, north-eastern parts of Pakistan, and along the foothills of the western Himalayas. The RegCM4.4 simulations with global model CCCMA-CanESM2 as LBCs are quite consistent with the model’s Era-Interim run (Fig. 11, bottom row), but a wet bias is noted along the coastal belt of Pakistan during the Kharif season. This indicates that the selected model RegCM4.4 (CCCMA-CanESM2) shows a reasonable skill in capturing the spatial pattern of seasonal and annual mean precipitation, but with slight overestimates (wet bias) over the region where summer and winter precipitation-producing weather systems prevail.

Fig. 11
figure 11

Left panels (a): Precipitation climatology (units: mm/month) for Kharif season using GPCC data for the period from 1981 to 2005. (b,c): Same as (a) except for RegCM4-4 with Era-Interim and CCCMA-CanESM2 as LBCs, respectively. Middle and right panels (d–f) and (g–i): same as left panels except for Rabi and annual basis, respectively

Full size image

3.4.2 Changes in future drought with RegCM4.4 projections

In order to analyze future drought frequency, severity, and trends over the PPP on seasonal (Kharif and Rabi) and annual basis using SPI and RDI drought indices, RegCM4.4 model simulation with CCCMA-CanESM2M GCM LBCs for two RCP scenarios, i.e., RCP4.5 and RCP8.5, are used. The analysis was performed for two time periods, i.e., reference period (1981–2005) and future projections (2006–2100).

The SPI and RDI results using RegCM4.4 (CCCMA-CanESM2M) simulations are presented in Fig. 12 (top panel) and (bottom panel), respectively. SPI results for the historical period show a negative slope of trend line for Kharif, Rabi, and annual time scales (Fig. 12a–c, top panel), which are consistent with the baseline observation (Fig. 4). The future SPI monthly time series show a positive slope of trend line for Kharif, Rabi, and annual time-scales under both RCP4.5 and RCP8.5 scenarios; however, Rabi witnesses a negative trend under RCP4.5 scenario. The results of the slope (negative) of trend line of RDI for the historical period (Fig. 12a–c, bottom panel) are again consistent with the baseline observations. The future RDI time series shows positive trend for Kharif and Rabi seasons under both RCP scenarios, whereas the annual showed almost no trend in the time series of the future.

Fig. 12
figure 12

Top panel (a–c): Time series comparison of SPI for Kharif, Rabi, and annual time-scales, respectively, based on RegCm4.4 (CCCMA-CanESM2M) outputs, under RCP4.5 and RCP8.5 scenarios. Bottom panel (a–c): Same as top panel except for RDI. The values in parenthesis show slope of trend line of SPI and RDI. The trend in time series is checked at 5% significance level and is marked with asterisk

Full size image

The results of SPI on 3-, 6-, 9-, and 12-month aggregation time-scales are presented in Fig. 13. It is observed from Fig. 13 that negative slope of trend line for all time-scales (SPI 3, 6, 9, and 12) for both Rabi and annual bases is well captured by the ReghCM4.4. However, results of SPI 3, 6, and 12 for Kharif season are not consistent with the baseline observation. The future RCPs show almost positive (towards wetness) slope of trend line for both seasonally and annually. The slope of trend line is statistically significant (at 5% significance level) for longer time-scale for Kharif season (SPI 12) as well as annually (SPI 9 and SPI 12 under RCP 4.5 scenario) (Fig. 13). The drought years are further calculated on seasonal and annual basis for all time-scales (Table 4). SPI 3 and SPI 6 analyses show maximum number of moderate level droughts during Kharif season under both RCP scenarios. In Rabi season, severe (3) and extreme (2) level droughts are identified under RCP4.5 and RCP8.5 scenarios, respectively. Based on SPI 9, it is noted that the maximum number of moderate level drought (12 out of 16 under RCP 4.5 and 6 out of 11 under RCP8.5) is observed during Rabi season, compared with the Kharif season. The results of SPI 12 show similar behavior during the Rabi season.

Fig. 13
figure 13

Time series comparison of SPI 3-, 6-, 9-, and 12-month aggregation time-scales for Kharif (a), Rabi (b), and annual (c) bases, using RegCM4.4 driven by CCCMA-CanESM2M LBCs under RCP4.5 and RCP8.5 scenarios. The values in parenthesis show slope of trend line of SPI. The trend in time series is checked at 5% significance level and is marked with asterisk

Full size image
Table 4 Details of extreme, severe, and moderate classes of drought based on SPI 3-, 6-, 9-, and 12-month aggregation time-scales for RegCM4.4, driven by CCMA-CanESM2 LBCs, under RCP 4.5 and RCP 8.5 scenario
Full size table

It is clear from Table 4 that the Kharif season is more vulnerable to short time-scale, whereas the Rabi is more vulnerable to long time-scale. It is interesting to note that the total number of droughts observed for annual cases (under both RCP scenarios) is less than that of seasonal basis, which clearly provides evidence that the Potwar region is strongly affected by the seasonal droughts as compared with the annual droughts. Moreover, the total number of drought and its intensity are less for shorter time-scales (SPI 3 and SPI 6) and more for longer time-scales (SPI 9 and SPI 12). Adnan et al. (2015) conducted a similar study over the Sindh province of Pakistan and found that the frequency of intense droughts may not be observed on shorter time-scales (e.g., SPI 3 or less). Our results also suggest that the drought events under RCP8.5 are less compared with RCP 8.5 on both seasonal and annual bases. In general, no significant trends of droughts are observed in the future under both RCP scenarios for the two cropping seasons. However, a positive slope of trend line of both SPI and RDI indicates an increasing trend towards wetness over the region, which may be attributed to the increasing trends of rainfall and moisture transport from the Arabian Sea (Latif et al. 2018).

4 Summary and conclusions

Drought is one of the most frequently occurring natural hazards, which has serious impacts on the agricultural sector. There have been many studies conducted in the past, but drought assessment and its related large-scale atmospheric dynamics as well as the future projections using RCMs for two major cropping seasons, i.e., Kharif and Rabi, over the Potwar Plateau of Pakistan (PPP), have not yet been studied over Pakistan. The agricultural activities of this region are highly dependent on rain water.

In this study, past and future droughts over the PPP are investigated along with related large-scale dynamics on seasonal and annual time-scales. The analyses are performed using station, observed, reanalysis, and RCM datasets, for the past (1981–2010) and future (2011–2100) time periods. For the identification of EW and ED cases of precipitation, the percentile rank approach and drought indices (SPI and RDI) are used. Future projections of droughts are investigated using RegCM4.4 and RCA4 model outputs under RCP4.5 and RCP8.5 scenarios. Our results show that, among six observational/reanalysis datasets, GPCC (for precipitation) and APHRODITE (for temperature) approximate well with the station data during both Kharif and Rabi seasons. It is also observed that the Era-interim reanalysis does not perform well for both precipitation and temperature cases.

By employing percentile score, SPI, and RDI criteria, we found that SPI-based cases of precipitation deficit and surplus are almost equal in Kharif and Rabi seasons; however, surplus cases on annual time-scale are slightly higher than deficit. The RDI results are consistent with the SPI. The percentile, SPI, and RDI results show the years 1986, 1987, 2002, and 2009 as driest, whereas the wet cases are noted during 1987, 1994, and 1997. The SPI and RDI results correlate well with percentile score method. SPI shows a positive slope of trend line for both seasonal and annual bases using GPCC but a negative trend line slope (positive) in Kharif (Rabi) season for the station data. A positive slope of RDI is observed for Kharif and negative for Rabi and annual cases. SPI is further calculated at 3-, 6-, 9-, and 12-month aggregation time-scales. The results show a negative slope on both seasonal and annual time-scales, indicating decreasing trends (towards drought). Due to the high rainfall variability over the PPP and its geographical location (receives highest amount of rainfall both in summer and winter seasons), drought period usually remains for a shorter time period and less intense as compared with other parts of the country.

The composites of geopotential height and wind anomalies show that, during EW-Kharif (EW-Rabi) season, Tibetan anti-cyclone (westerly flow) plays a crucial role in moisture incursion from the Bay of Bengal (Arabian Sea) for wet conditions over the study region. The results are the same for ED cases, but with a reverse sign of anomalies. The composite anomalies of integrated moisture transport and moisture flux convergence/divergence show that the presence of moisture convergence anomalies over most of the Indo-Pakistan subcontinent during seasonal and annual time-scales is responsible for transporting moisture from Bay of Bengal and Arabia Sea, resulting in EW conditions over the study region. Our results in relation to the identification of positive height anomaly (anti-cyclone) over the Tibetan Plateau and associated moisture transport from the Bay of Bengal during Kharif season are a new contribution to the existing literature. The precipitable water composite anomalies show higher precipitable water amount, associated with summer monsoon (western disturbances), transported from the Bay of Bengal (Arabian Sea), which contributes directly to Kharif (Rabi) rainfall amount over the region. The vertical velocity results indicate the important role of negative (positive) anomalies during EW (ED) cases.

Future drought conditions are further investigated using RegCM4.4 and RCA4 outputs, driven by different GCMs as LBCs, from CORDEX south Asia domain. The RegCM4.4 model run with CCMA-CanESM2 LBCs is found to approximate well with the observations. The performance of selected RegCM4.4 is further evaluated with ERA-Interim LBCs and GPCC precipitation and found that the RegCM4.4 model shows a reasonable skill in capturing spatial pattern of seasonal and annual mean precipitation. Based on the selected RegCM4.4 model outputs, the SPI and RDI results for the historical period show negative (towards drought) slope of trend line, which are consistent with the baseline observations. The future SPI and RDI monthly time series show a positive (towards wetness) trend for Kharif, Rabi, and annual time-scales under both RCP scenarios. The negative trends for SPI 3-, 6-, 9-, and 12-month time-scales are well captured by the ReghCM4.4 model for both Kharif and annual cases; however, the results of Kharif season are not consistent with the baseline observations. The projected changes in droughts using SPI and RDI under both RCPs show weak positive trend for both seasonally and annually, which could be linked to an increasing trend of rainfall and moisture transport from the Arabian Sea. Given this study explains a physical linkage between large-scale dynamics and drought/wetness conditions, the same association could be studied in the future using regional or global climate models to see why the trends of drought severity are increasing in the future. This will be helpful in understanding the large-scale dynamics of future projected droughts.

This study will provide assistance for the long-term forecast of drought and to develop an early warning system of drought occurrence. It will be useful in developing contingency strategies for the rain water-dependent region that will ultimately be used for guiding water resource management practices, increasing agricultural yield, and improving ecological protection and socio-economic growth. The study will also be helpful in enhancing the competence of humanitarian response in order to handle extreme weather conditions.