The impacts of climate change on meteorological drought in the Central Highlands region of Vietnam

Drought is a long-term natural phenomenon in which water conditions are less sustainable than the local norm. Due to global warming and the El Niño phenomenon in the Pacific Ocean, which increases the destabilization of the atmospheric structure, ocean monsoons are unable to make landfall and form precipitation. This has resulted in frequent drought disasters around the world in recent years, affecting regions such as North America, Australia, Europe (Forzieri et al. 2014; Gudmundsson & Seneviratne 2015), East Africa (Sheffield et al. 2012; Masih et al. 2014) and Asia (Grafton et al. 2014; Hao et al. 2018; Nguyen & Nguyen 2018; Shukla et al. 2019). Drought disasters lead to a lack of drinking water for humans and animals as well as the destruction of crops and land. Additionally, they can cause the water level of rivers to drop, bringing shipping to a standstill (Mkhabela et al. 2010; Al-Kaisi et al. 2013; Zengxin et al. 2015). Understanding the temporal and spatial characteristics of droughts is essential for the planning and management of water resources.

The United States Environment Protection Agency (EPA) has also classified severe drought as a natural disaster. According to Obasi (1994) statistics on natural disasters, the World Meteorological Organization (WMO) found that between 1967 and 1991, drought accounted for only 7.3% of all events of all types of natural disasters but resulted in 38% of fatalities, the highest figure among all natural hazards (Obasi 1994).

Droughts can be quantified by drought indices, such as drought severity, drought duration and drought intensity (Keyantash & Dracup 2002). An increasing number of drought indices have been developed for drought monitoring and quantification (Mahmoudi et al. 2019). The commonly used drought indices include the Palmer drought index (PDSI) (Gobena & Gan 2013), standardized precipitation index (SPI) (Mckee et al. 1993), standardized precipitation evapotranspiration index (SPEI) (Vicente-Serrano et al. 2010), comprehensive drought index (CI) (Murthy et al. 2017), reconnaissance drought index (RDI) (Mohammad et al. 2015) and humid index (HI) (Zhuguo et al. 2004). Hydrological drought occurs as a result of the expansion of meteorological drought within the terrestrial hydrological cycle, and as such, it is impacted by the characteristics of the hydrological cycle (Vidal et al. 2010). Mohamed & Cahit (2022) studied the relationships between climate indices and hydrological drought in the Meriç Basin, Turkey, using the standardized streamflow index (SSFI), which revealed a close relationship between the El Niño Southern Oscillation (ENSO), North Atlantic Oscillation (NAO) and Atlantic Multidecadal Oscillation (AMO) and hydrological drought in the research area. The SPI is likely the most widely used meteorological drought index in the world because it only needs precipitation data, and the results are reliable (Wu et al. 2005). Labedzki et al. used the SPI to verify precipitation condition forecasts and found that the 10-day SPI forecast was reliable regardless of whether the SPI classification forecast or the SPI value forecast was used, and the 20-day forecast should be reserved, accepted and used cautiously (Łabędzki 2017). Sobral et al. reported that the SPI could better reflect the drought situation in Rio de Janeiro, Brazil. The SPI plays an important role in the analysis of drought characteristics because the income from precipitation has a direct impact on the occurrence and evolution of drought (Bruno Serafini et al. 2019). However, it should not be ignored that water expenditure, such as potential evapotranspiration (PET), also has a direct impact on the evolution of drought. Therefore, taking the expenditure of PET into account has become the focus of scholars' research in recent years; thus, several drought indices, such as the SPEI and HI, have been developed. Predicting drought is a significant challenge for hydrologists and climatologists because it occurs at various spatial and temporal scales (Hao et al. 2018). Hydrologists and climatologists have continuously tried to forecast droughts using methods such as statistical, dynamic and hybrid models (Zhang et al. 2023). Khalid et al. (2024) used machine learning algorithms and climate data to assess and forecast meteorological drought in Morocco's Upper Drâa Basin (UDB) and analysed data spanning from 1980 to 2019. The findings emphasize the uneven distribution of drought, different patterns among indices and particular areas displaying noteworthy decreasing trends. Rezaiy & Shabri (2023) presented a new hybrid W-ARIMA model and utilized it to predict the occurrence of drought in Kabul, Afghanistan, and the W-ARIMA model's drought prediction results consistently surpassed those of the traditional ARIMA model, highlighting its ability to improve drought forecast precision.

The SPI is a probability index that was developed to provide a better representation of abnormal wetness and dryness. The SPI is essentially a standardizing transform of the probability of the observed precipitation (Guttman 1999). The standardization of the drought index ensures that the index in question is independent of its geographical consideration in terms of average rainfall (Vu-Thanh et al. 2014). In Asia, especially in Vietnam, many researchers have adopted this index as an efficient and useful tool for drought monitoring at both the regional and national levels (Vu-Thanh et al. 2014; Huynh et al. 2020). The impact of climate change on drought is complex and difficult to assess because it depends on many climate factors, including land and sea surface temperature, water resources and rainfall patterns (Huynh et al. 2020). Various modelling approaches can be used to estimate the future climate; however, the accuracy of such estimates depends mainly on the quality of the hydrogeological and climatic data.

Climate models are important tools for studying the mechanisms of climate change and attributing climate change. To date, there are many global climate models (GCMs) and regional climate models (RCMs) for studying the mechanisms of climate change and researching the factors of climate change. A series of GCMs can effectively simulate the average properties of a large-scale grid, near-surface temperature properties, high-level gas fields and atmospheric circulation (Wang et al. 2019). The fifth phase of the Coupled Model Intercomparison Project (CMIP5) was developed by the World Climate Research Programmer's (WCRP) Working Group on Coupled Modelling (WGCM) (Taylor et al. 2012). The latest CMIP phase, the CMIP6 dataset (Eyring et al. 2016), provides the underlying scientific ground for the latest IPCC Sixth Assessment Report (AR6) of the Intergovernmental Panel on Climate Change (IPCC 2021). The shared socioeconomic pathway (SSP) scenarios used in AR6 include the SSP1 (sustainability), SSP2 (middle of the road), SSP3 (regional rivalry), SSP4 (inequality) and SSP5 (fossil-fuelled development) pathways (Keywan et al. 2017). CMIP6 is an important tool for predicting future climate change because it has good performance in verifying accuracy and applying research in different regions and has been widely used in recent years in various fields, such as near-surface wind speed, extreme precipitation, floods and drought. Thus, CMIP6 could be applied to the study of climate-related impacts, especially in drought research.

The climate in Vietnam is strongly influenced by monsoons and complex topography. The total rainfall in the dry season accounts for approximately 15–25% of the annual rainfall. There are statistical data on the sectors affected by drought events in Vietnam since the 1980s. Severe drought occurred in connection with the 1997–1998 El Niño episode, which is considered the strongest event of the 20th century. During this period, most regions of Vietnam, especially the central and southern regions, were affected by severe drought. According to a drought assessment by the Vietnamese Ministry of Agriculture and Rural Development, approximately three million people were affected in 1997–1998, and total losses in agricultural production were estimated at approximately 400 million US dollars. Very few studies have been conducted on drought in Vietnam. In a recent study comparing different drought indices for Vietnam, Vu-Thanh et al. (2014) concluded that no particular meteorological drought index best represents drought conditions in Vietnam, and it remains an important challenge to better understand the relationship between the different types of droughts in Vietnam and climate.

Therefore, this study aimed to investigate the effects of climate change on drought events in the Central Highlands region of Vietnam using the SSP2-4.5 and SSP5-8.5 scenarios. In this study, four CMIP6 GCM models under two climate change scenarios, SSP2-4.5 and SSP5-8.5, namely, CNRM-CM6-1-HR, BCC-CSM2-MR, MIROC6 and MRI-ESM2-0, were selected after increasing the horizontal resolution via bias correction and spatial disaggregation (BCSD) for the period 2031–2060 to predict rainfall in the future. The SPI with a difference time scale of 3, 6, 9 and 12 months was used to assess drought events in the Central Highlands. This paper is divided into sections. Section 2 provides an overview of the dataset and methods. Sections 3 and 4 discuss the changes in drought intensity, frequency and duration under the two climate change scenarios SSP2-4.5 and SSP5-8.5 with four GCM models. Section 5 contains the conclusion.

The study area is the Central Highlands of Vietnam, which is situated in the southeastern part of the Indochina Peninsula and is bounded by latitudes 11°N–15°N and longitudes 107°E–109°E. The region has varied complex topography. (1) High mountains ranging from 1,000 to 2,500 m are located in the northeast and southeast of the study region. The area is covered by forest, and most ethnic people live in this area. (2) The plateau topography extends westwards from the DaLat highland (1,000 m) to the border with Cambodia (100 m). (3) Valley topography with a small area distributed along the river basin. The study region contains 1.7 million hectares (ha) of basaltic soil (comprising 90% of the total basaltic area in Vietnam) and 3 million ha of forest (22% of the total forested land). Therefore, the region is suitable for planting perennial plants such as coffee and rubber or annual plants such as cocoa, pepper, sugarcane and cashew. The cultivated area is approximately 800,000 ha with major crops in the region.

To evaluate drought conditions in the Central Highlands of Vietnam, we collected the observed meteorological data from 11 regional stations. The locations of these stations are displayed in Figure 1. The data include observed daily and monthly rainfall from the Vietnam National Hydrometeorological Center covering 1985 to 2014. We followed standardized operational procedures to verify the collected data. Then, we used the three-sigma (five-sigma) rule to identify any potentially inaccurate values in the precipitation data. Each flagged case was carefully reviewed for accuracy.

CMIP6 has been widely used in recent years as an important tool for analysing climate change. Determining the number of models in the CMIP6 GCM is a challenging task. Most previous studies selected the first few top-ranked GCMs to prepare the ensemble. However, deciding on the number of models often depends on human subjectivity. Lutz et al. (2016) selected the top two models and then refined them to a model consistent with their ability to reproduce monsoon rainfall. Ahmed et al. (2019) and Kamal et al. (2020) choose the four highest-rated models for poor reasons. Ahmed et al. (2019) and Nashwan & Shahid (2020) identified the top 50% of GCMs to indicate the performance of the models. In Vietnam, when developing climate change scenarios for Vietnam, the Ministry of Natural Resources and Environment also uses 10 CMIP5 GCM models with 2 scenarios, RCP4.5 and RCP8.5 (MONRE 2020). On the other hand, Nguyen-Duy et al. (2023) provided a ranking for the effectiveness of CMIP6 GCMs for Vietnam. In this study, we selected 4 out of 36 CMIP6 GCM models for drought analysis based on their rankings (Nguyen-Duy et al. 2023). Among the four models, one was highly efficient, one had a low ranking and two had an average ranking for analysis and evaluation. Four CMIP6 GCM models, namely, CNRM-CM6-1-HR, MRI-ESM2-0, BBC-CSM2-MR and MIROC6, were used in this study to analyse drought trends and the frequency and characteristics of extreme drought. The specific information on the four CMIP6 models is shown in Table 1.

The current resolution of the CMIP6 GCM is still too coarse, typically above 100 km, to be effectively utilized in various aspects, such as risk assessment, adaptive management and decision-making processes at the regional or local scale. In addition to the coarse resolution, the biases and uncertainties in GCMs are often amplified from global to regional and local scales, limiting their usefulness and applicability in small-scale research studies (Bauer & Del Genio 2006; Guo & Wang 2016). In this research, the BCSD method, which is referred to as the BCSD method, is used (Wood et al. 2004). Using statistical downscaling to maximize the construction period of the BCSD approach (Reiter et al. 2016; Trinh-Tuan 2019; Tran-Anh et al. 2023), the entire 30 years from 1985 to 2014 were used to generate transfer functions (TFs) and guide spatial disaggregation (SD) for the 2031–2060 future period. The BCSD consists of two major steps: bias correction (BC) and SD. Before BC is carried out, the data from all GCMs and the observations (OBS) are regridded to an intermediate resolution of 1° × 1°. The process consists of two main steps: BC and SD. Before BC, we removed the trend from the temperature data at each grid point. After that, we added the trends to ensure that the original GCM's climatic trends were preserved. The BC method uses the quantile mapping (QM) method to adjust the biases in the GCM data relative to the OBS data at a resolution of 1° × 1°. For each month, for each variable in each grid cell, the CDFs for both the OBS and historical GCM data are generated separately. TFs are developed to map the model CDFs onto the OBS CDFs. The biases in the GCM monthly outputs are adjusted by using TFs to transform the GCM data to the OBS data for the same CDF quantile. It is assumed that those QM TFs will remain stable throughout historical and future periods, and as a result, they are used to adjust the future projected variables. For temperatures, after the QM step, the climatic trends that were previously saved are added back to the QM model data. Subsequently, the climatological mean bias adjustment between the GCM and the OBS temperature at each grid point is applied, resulting in BC temperature data at an intermediate resolution of 1° × 1°. During the SD step, the BC data undergo interpolation to a resolution of 0.1° × 0.1° through a three-step procedure: first, interpolation of the additive (for temperature) and multiplicative (for precipitation) change factors, and the comparison between the BC GCM fields and the OBS climatology is expected to be performed at the intended high resolution of 0.1° × 0.1° through the use of bilinear interpolation; second, generating high-resolution BC GCM data involves incorporating interpolated change factors and 0.1° × 0.1° OBS climatology by addition (for temperature) or multiplication (for precipitation); and finally, temporal disaggregation of the monthly BC fields of the future period to a daily scale by randomly selecting a month from OBS and adjusting its daily values additively and multiplicatively to replicate the future monthly BC data. Note that when downscaling precipitation, it is important to assess the scaling factor and the number of wet days to prevent unrealistic precipitation values during temporal disaggregation. For instance, if the selected month has fewer than three wet days and a scaling factor greater than three, another year with more wet days in that month should be selected.

In this study, the BCSD approach is implemented in two primary stages: Phase 1 – Testing: Utilizing the climatological fields from the OBS and GCMs during the training period of 1985–1999, the TFs are developed to establish the relationship between the simulations and observations. Subsequently, the BCSD is used during the independent period of 2000–2014, and the results are compared with the OBS data to assess the effectiveness of the BCSD in replicating past climate conditions. Phase 2 – Future downscaling: The BCSD approach makes use of the entire 30-year period from 1985 to 2014 to create TFs and provide guidance for SD during the future period from 2031 to 2060.

After applying the statistical downscaling method, the resolution is increased to 0.1° × 0.1° latitude and longitude. The precipitation data of the four CMIP6 GCM models are used as the input data for the calculation of the SPI.

The calculation of the SPI can be divided into 1, 3, 6, 9, 12 and 18 months and other time scales, with the different time scales representing different states of water accumulation. The duration of drought events should not be too short or too long because there are two distinct seasons and obvious seasonal variations in drought events in the Central Highlands region of Vietnam. Therefore, the SPI at 3-month and 6-month scales (SPI3 and SPI6, respectively) is suitable for identifying continuous drought events in the region and precipitation over different seasons. At the same time, the SPI9 and SPI12 were used as annual drought indices indicating water scarcity in a given year. When the 9-month SPI (SPI9) is less than −1.5, there is a significant impact on agriculture and other sectors. The 1-year SPI (SPI12) is linked to the development of flows and levels of water reservoirs and groundwater.

According to different SPI values, the classification of drought was based on the research of McKee et al., as shown in Table 2, and the application of the SPI duration is shown in Table 3 (Mckee et al. 1993).

The SPI was calculated for meteorological drought using monthly precipitation (both station data and gridded observations) for the period 1985–2014 at 11 rainfall stations. The drought parameters and their statistical properties (number of droughts, duration, frequency and deficit/severity) proposed by Yevjevich (1967) at different stations were computed based on the station data and the bilinearly interpolated gridded data to the station point. Severe drought events in the past were identified from drought characteristics. The spatial distribution of these severe drought events was calculated.

The Mann-Kendall (MK) test (Mann 1945) and Theil-Sen (SS) test (Sen 1968) are the most commonly used approaches for trend analysis of hydrometeorological variables. Although the MK test can detect statistically significant trends, it may yield inaccurate results if serial correlations exist in the time series (Kalra et al. 2017). Transformed series such as drought indices can be noticeably affected by serial correlations, which could result in incorrect conclusions based on data characteristics. Consequently, it is necessary to use procedures such as the modified Mann-Kendall test (MMK) (Hamed & Ramachandra Rao 1998) for such analyses. Therefore, in this study, the MK, MMK and SS methods were used to analyse drought trends in the Central Highlands region of Vietnam.

The MK method (Mann 1945) is a nonparametric test recommended by the WMO and has been widely used. It can assess the significance of the time series' change trend or conduct a mutation test. The advantage of this approach is that it does not require the sample to follow a certain distribution, nor is it interfered with by a few abnormal values (Wang et al. 2019). In this study, the original MK test is employed to identify the trend of the SPI series.

When Zs was greater than 1.96, the change trend increased, and when Zs was less than −1.96, the change trend decreased (at the 5% significance level).

The positive (Q_i) values show a rising trend, while the negative Q_i values indicate a decreasing trend in the time-based data.

Run theory, proposed by Yevjevich in 1967, was employed to recognize the characteristics of drought (Yevjevich 1967). A run is composed of a part of the time series of drought variables, where each value is higher or lower than a given truncation level (Huang et al. 2014). In this study, the value of −1.0 was selected as the truncation level. If the SPI is less than −1.0, then the drought is classified as extreme. In this study, drought duration, drought frequency (DF) and drought intensity were chosen to analyse extreme drought events in the Central Highlands region of Vietnam.

Here, m is the number of months in which drought occurred, and M is the total number of months.

Here, I is drought intensity (−), n is the number of drought occurrences in months with SPI < −1.0 and SPI_i is the SPI value below the threshold (−1).

Model calibration and validation are two of the most important steps in this process. The aim was to evaluate the model's performance in simulating climate data at the selected location to determine whether it is suitable for use. In situ observed precipitation data from 1985 to 2014 from 11 meteorological stations in the Central Highlands of Vietnam were used for calibration and validation. The first 15 years of rainfall data (from 1985 to 1999) were used for calibration, and the remaining 15 years (from 2000 to 2014) were used for validation. The performance of the BCSD method during calibration and validation was evaluated using statistical parameters such as correlation coefficients. The correlation coefficient compares the explained variance of the modelled data with the total variance of the observed data; the value is between 0 and 1. A higher value indicates better model performance. In general, the simulation model can be rated as satisfactory if the correlation coefficient is greater than 0.70.

The simulation results of the CNRM-CM6-1-HR model show that rainfall decreased from January to June in the entire Central Highlands region, especially in May, with the largest decrease in rainfall in DakNong of 105.03 mm, followed by Coming to KonTum, with a decrease of 101.8 mm. The amount of rainfall at the BaoLoc, PleiKu, DakTo and DaLat stations decreased from 80 to 92 mm, and that at the other stations decreased from 50 to 68 mm. The amount of precipitation increased significantly at most stations in August.

According to the MIROC6 model, the rainfall at the BuonMaThuot station sharply decreased, with the largest decrease of −163 mm occurring in October. From March to June, the total amount of rainfall decreased by approximately 400 mm, while the amount of rainfall at the other stations decreased. The trend did not change significantly. In November, the amount of rainfall increased at all stations in the region, with the largest increase occurring at the MDrack station and the smallest increase occurring at the DakTo station (260 and 12 mm, respectively). According to the MRI-ESM2-0 model, rainfall decreases sharply during the dry season, particularly from March to June, with values ranging from 35 mm to 86 mm throughout the Central Highlands region.

For the SSP5-8.5 scenario, the results of the BCC-CSM2-MR model show a sharp decrease in rainfall in October and November, with values between 36 and 85 mm across the region, while the CNRM-CM6-1-HR model shows that rainfall decreased from January to June, with rainfall at the DaLat, DakNong, PleiKu and LamDong stations. The MRI-ESM2-0 model also showed a decrease in rainfall in the first months of the year, especially in May; rainfall decreased throughout the region, most significantly in PleiKu and DakNong, KonTum and DakTo, with a decrease in rainfall that fluctuated by approximately 90 mm. At other stations, rainfall fell between 45 and 73 mm. According to the MIROC6 model, the amount of rainfall at the BuonMaThuot station decreased sharply during most months of the year. In April, June and October, the amount of rainfall decreased by 115.67, 184.47 and 126.10 mm, respectively; at the AnKhe, PleiKu and DaLat stations, the amount of rainfall decreased in August and fluctuated around 50 mm.

Thus, rainfall in the Central Highlands region will be subject to strong fluctuations in the period 2031–2060 compared with the base period 1985–2014 in both the SSP2-4.5 and SSP5-8.5 scenarios when using GCM models. In particular, the CNRM-CM6-1-HR model and the MRI-ESM2-0 model show a significant decrease in rainfall during the dry season in both the SSP2-4.5 and SSP5-8.5 scenarios. The MIROC6 model shows a decrease in rainfall at the BuonMaThuot station, while the BCC-CSM2-MR model shows a slight decrease in precipitation in most months of the year. The results show that under the influence of climate change, rainfall in the Central Highlands tends to decrease during the dry season, which will have an impact on agricultural development (especially on industrial cultivation in the region).

The SPI3 calculations indicate that 1998 and 2005 experienced severe to extremely severe drought conditions, with SPI values less than −1.5 (SPI3 was −1.52 in 1998 and −2.67 in 2005). This finding aligns with Nguyen's research in 2005, which revealed that the droughts in 1998 and 2005 followed the El Niño phenomena in 1997 and 2004 in Vietnam. The simulation results for SPI6, SPI9, and SPI12 also confirm severe to extremely severe droughts in 1998 and 2005. Specifically, the SPI6 values were −1.69 and −2.30 in 1998 and 2005, respectively. The SPI9 indices were −1.78 and −1.86 in 1998 and 2005, respectively, while the SPI12 indices were −1.50 and −1.67 in 1998 and 2005, respectively.

Drought characteristics were also assessed by drought deficit (DD) and DF (Table 5). The DD, which represents the extent of drought during the study period, is calculated as the sum of the SPI values (negative and without units) for all drought periods. DF describes the probability of a drought occurring during the study period, which is the division of the total drought duration (in months) by the total number of months in the study period.

In Table 5, we observe that the number of drought periods is consistent for SPI3 and SPI12, both of which have three drought periods. The SPI6 index shows the lowest number of drought periods with two periods, while the SPI9 index has four periods. Regarding the deficit level, SPI3 had the smallest deficit at −33.74, followed by SPI9 at −30.88 and SPI12 at −17.39.

Figure 8 displays the spatial distribution of drought events in May 1998 over the Central Highlands region of Vietnam. The calculation results of the SPI show that the SPI is less than −0.5 across the entire research area. When considering the calculation periods of 3, 6, 9 and 12 months, severe drought with an SPI drought index less than −2 was observed in Kon Tum and parts of Gia Lai in all four calculation periods. In Lam Dong, severe drought was observed on 3- and 6-month scales. The results of the SPI at the 6- and 9-month scales showed that the area of severe drought, with an SPI between −2 and −1.5, expanded further into DakLak Province. The spatial distribution map of drought calculated using the SPI corresponds to the actual drought of 1998, particularly in the provinces of Kon Tum and Gia Lai.

Table 6 shows the Z and p values of the MK and MMK methods and the Sen's slope (SS) values computed by applying nonparametric tests to the SPI values at time scales of 3, 6, 9 and 12 months.

Table 6 presents the results of the MK and MMK tests for a 3-month scale (SPI3), showing that the trend of the SPI3 values is increasing at all 11 stations in the Central Highlands region of Vietnam. The trend of the SPI3 at the AnKhe, BaoLoc, DakTo and MDrack stations is statistically significant at the 5% level. However, the other stations – Ayunpa, BuonHo, BuonMaThuot, DakNong, DaLat, and PleiKu – show a tendency for the SPI3 values to increase, but this increase is not statistically significant at the 5% level. At the KonTum station, there is a statistically significant increase in the SPI3 value. The increase is significant at the 5% level with the MMK method and at the 10% level with the MK method. This indicates that for the AnKhe, BaoLoc, DakTo, KonTum and MDrack stations, there was a statistically significant decrease in the magnitude of drought, while at the remaining stations, the decreasing trend of drought was not statistically significant. For the 6-, 9- and 12-month drought indices (SPI6, SPI9 and SPI12, respectively), there is a decreasing trend at the Ayunpa station, but the results are not statistically significant at the 5% level. All the other 10 meteorological stations show an increasing trend. At the AnKhe, BaoLoc, DakNong, DakTo, KonTum and MDrack stations, the increasing trend in SPI6 is statistically significant at the 5% level. At the BuonMaThuot and PleiKu stations, the SPI6 results are statistically significant at the 5% level according to the MMK method. The SPI9 values at the AnKhe, BaoLoc, BuonMaThuot, DakNong, DakTo, KonTum and MDrack stations tend to increase and are statistically significant at the 5% level according to both the MK and MMK methods. The SPI12 values at the Ayunpa, DakTo, DaLat, and PleiKu stations are not significantly different when using the MK and MMK methods.

At the PleiKu station, except for SPI6, SPI3, SPI9 and SPI12 show an increasing trend (indicating a decrease in drought levels), but these results do not demonstrate statistical significance for the period between 1985 and 2014. When evaluating the SS values estimated from the SPI at 3, 6, 9 and 12 months for all 11 meteorological stations in the Central Highland region of Vietnam, it is evident that only the Ayunpa station has a negative SS value when calculated with the SPI9 and SPI12. The other stations display positive SS values with all SPI values. This indicates a downward trend in the SPI9 and SPI12 at the Ayunpa station (suggesting an increasing drought intensity), while the SPI values at the remaining stations demonstrate an upward trend (indicating a decreasing drought intensity). Therefore, the drought trend in the Central Highlands region during this period did not clearly increase or decrease, as the results were not statistically significant, except for those at the AnKhe and BaoLoc stations or at the DakTo and MDrack stations.

Figure 9 shows the drought events using the SPI3 and the 6-, 9- and 12-month drought indices calculated for the four CMIP6 GCMs under the SSP2-4.5 scenario. According to the results from the BCC-CSM2-MR scenario, severe to extreme droughts will occur between 2038 and 2039 and between 2046 and 2050. The smallest values of SPI3 and SPI6 are −2.04 and −1.76, respectively, in 2036, while the smallest indices of SPI9 and SPI12 are −1.63 and −1.86, respectively, in 2039. The MIROC6 model for the smallest SPI3 index was −2.61 in the 2033 drought period, while the minimum values of SPI6, SPI9 and SPI12 were −2.31, −163 and −1.95, respectively, in the 2048–2049 period. The CNRM-CM6-1-HR model shows that severe droughts will occur at all SPI time scales in the period 2033–2035. The smallest SPI3 index is −2.01 in 2033, and the smallest values of the SPI6, SPI9 and SPI12 indices are −2.01, −1.46 and −1.46, respectively, in 2033. The MRI-CSM2-0 model produces the smallest SPI3 and SPI6 index values of −1.92 and −1.99, respectively, in 2038, while the smallest SPI9 and SPI12 index values are −1.32 and −1.46, respectively, in 2053.

Figure 10 shows the calculation results of the SPI3, SPI6, SPI9 and SPI12 indices for the four CMIP6 GCMs under the SSP5-8.5 scenario. All four models indicate severe to extreme drought occurrence during the 2031–2060 periods. Specifically, the BCC-CSM2-MR model exhibits the smallest SPI3 value of −1.17 in 2046, the smallest SPI6 value of −1.68 in 2049, the smallest SPI9 value of −1.83 in 2046–2047, and the smallest SPI12 value of −1.76 in 2038. Next, the CNRM-CM6-1-HRM shows extreme drought conditions during 2034–2035, with the smallest SPI3, SPI6, SPI9 and SPI12 values of −2.47, −2.77, −2.36 and −2.25, respectively, observed in 2034. Similarly, the MIROC6 model demonstrated the most severe drought during the 2050–2051 period, with SPI3, SPI6, SPI9 and SPI12 values of −1.51, −1.56, −1.99 and −1.76, respectively. Finally, the MRI-ESM2-0 model results reveal that the smallest SPI6, SPI9 and SPI12 values occur between 2041 and 2043, with values of −1.66, −1.89 and −2.49, respectively, indicating a range from severe to extreme drought. The smallest SPI3 of −2.31 was observed in 2058.

To assess the frequency of drought occurrences ranging from moderate to severe and extreme, this study considered the variability in the number of drought episodes, the frequency of drought events, and the magnitude of deficiency, as depicted in Table 7.

Table 7 indicates that, in the period 2031–2060, using four CMIP6 GCM models under the climate change scenarios SSP2-4.5 and SSP5-8.5, there is a projected increase in the number, frequency, and extent of droughts at 6, 9 and 12 months compared with the period 1985–2014. According to the results of the MIROC6 and MRI-ESM2 models under the SSP5-8.5 scenario, the drought indices SPI9 and SPI12 clearly demonstrate an increase in the number of droughts. Specifically, the number of drought episodes increased from 4 and 3 during 1985–2014 to 7, 6, 5 and 5, respectively. The frequencies of drought occurrence are 6.39 and 3.89% in the 1985–2014 periods and are projected to increase to 14.44 and 16.94% in the SSP2-4.5 scenario and to 8.33 and 12.78% in the SSP5-8.5 scenario, respectively.

Table 7 shows that the deficit of drought with SPI9 and SPI12 under the SSP5-8.5 scenario of the four GCMs tended to increase compared with the DD computed under SSP2-4.5 and DD during 1985–2014. However, the DD of SPI3 decreases during the period of 2031–2060 under both climate change scenarios compared with that in 1985–2014.

Figure 11 shows the frequency of drought events based on the SSP2-4.5 scenario of the four CMIP6 GCM models. The frequency of moderate to severe drought (−2.0 < SPI < −1.0) ranged from 4.5 to 10%. The BCC-CSM2-MR model exhibits the lowest frequency variation between the SPI3, SPI6, SPI9 and SPI12 indices. Specifically, the frequency of the SPI3 index is 4.5%, while those of the SPI6, SPI9 and SPI12 indices are 5.83, 5 and 5.83%, respectively. On the other hand, the MIROC6 model shows the largest increase in the frequency of drought events, ranging from 5.28% for the SPI3 index to 10% for the SPI12 index. The frequencies for the SPI6 and SPI9 indices are 7.78 and 8.33%, respectively. For the CNRM-CM6-1-HR model, the frequencies of the SPI3, SPI6, SPI9 and SPI12 indices are 5.83, 7.78, 8.61 and 5.28%, respectively. Finally, in the MRI-ESM2-0 model, the frequency of SPI3 is 6.39%, and that of SPI12 is 9.44%.

In the SSP5-8.5 scenario, the frequency of the SPI3 index ranges from 4.72% in the CNRM-CM6-1-HR model to 6.1% in the MRI-ESM2-0 model, as well as in the BCC-CSM2-MR model. The occurrence frequency of SPI3 in the MIROC6 model is 5.83%. For the SPI6 index, the frequency fluctuates more widely, with the MIROC6 model at 5.56% and the CNRM-CM6-1-HR model at the highest occurrence frequency of 10.28%. With respect to the SPI9 index, the frequency significantly increased to 14.44% in the MIROC6 model, 13.06% in the BCC-CSM2-MR model and 8.61% in both the CNRM-CM6-1-HR and MRI-ESM2-0 models. For the SPI12 index, the MIROC6 model showed the highest frequency at 17%, followed by the BCC-CSM2-MR model at 14.8%.

When comparing the frequency of occurrence between the period 2031–2060 under the two scenarios SSP2-4.5 and SSP5-8.5 with the baseline period 1985–2014, it is evident that the frequency of occurrence increases significantly over time scales of 6, 9 and 12 months. This suggests that moderate to severe drought levels (−2.0 < SPI < −1.0) in the Central Highlands of Vietnam are forecasted to occur more frequently in the future.

According to Figure 13, it is clear that the average frequency distribution of drought occurrence in the period 2031–2060 is greater in the SSP5-8.5 scenario than in the SSP2-4.5 scenario across the entire Central Highlands region. This trend is consistent across all four models: BCC-CSM2-MR, CNRM-CM6-1-HR, MIROC6 and MRI-ESM2-0, which are represented in darker red. Specifically, for the SPI9 and SPI12 indices of the MIROC6 model, the frequency distribution of average drought occurrence is depicted in dark red under the SSP5-8.5 scenario, with frequency values ranging from 14 to 23%. In contrast, under the SSP2-4.5 scenario, the frequency distributions of drought occurrence for SPI9 and SPI12 are depicted in yellow, with the average frequency of drought occurrence ranging from 13 to 17%.

The results of this study have important implications when considering drought development and assessing drought in the past and the future of the Vietnam Central Highlands, a region with a large area of ​​industrial crops (coffee, pepper, etc.) (Tran 2023) whose economy depends mainly on agriculture. On the other hand, droughts often occur in the region, causing difficulties for the region's socioeconomic development. Recently, there have also been several studies on precipitation and meteorological drought using the SPI in the region or surrounding area (Tue et al. 2015; Le et al. 2019; Tran 2023). The current research focuses only on assessing existing drought conditions and does not provide thorough predictions for future regional drought trends. The drought assessments in the past of this study are also consistent with previous research results. For example, previous studies have also accurately assessed moderate to severe drought developments from 1985 to 2014, the greatest frequency and severity of droughts occurred in 1998, 2004 and early 2005, the time period when the El Niño phenomenon took place.

Finally, to better evaluate the results of this study, important comments regarding the choice of the SPI term index are needed. In particular, according to research by Hayes et al. (1999), the SPI has several advantages, such as being easy to evaluate because it depends only on precipitation. The SPI is a standardized index and therefore ensures that the frequency of extreme phenomena at any location and at any time scale is consistent. The time scale in term assessment with the SPI is useful in term analysis. However, there are also some limitations when using the SPI term index due to the assumption that it is possible to find a suitable theoretical probability distribution to model the raw rainfall data because different distributions can create different results. Furthermore, the length of the rain data can significantly impact the SPI. In addition, when forecasting drought in the future, using rain data from four global rain models, the CMIP6 GCM with the BCSD method, to increase the resolution to a 10 × 10 km may lead to inappropriate conclusions about future droughts. Therefore, further research using other GCM models and drought indices are necessary to ensure accurate and reliable drought predictions for the Central Highlands of Vietnam.

In this research, we assessed the drought situation in Vietnam's Central Highlands from 1985 to 2014 by utilizing the SPI across different time scales (3, 6, 9 and 12 months). By applying the runs theory to examine drought trends, we found that the average frequencies of drought occurrence for the SPI3, SPI6, SPI9 and SPI12 indices were 6.94, 6.11, 6.39 and 3.89%, respectively. The current number of drought episodes for each index is 3, 2, 4 and 3, respectively. When using the MK, MMK and SS methods to evaluate the drought trend during the period of 1985–2014, the results did not demonstrate a clear increase or decrease, suggesting that the drought trend in the Central Highlands region over this period was not statistically significant. Notably, with the exception of the AnKhe, BaoLoc, DakTo and MDrack stations, the SPI indices indicated an increasing trend (decreasing drought level), although the change was not substantial.

For the period 2031–2060, four CMIP6 GCM models – BCC-CSM2-MR, CNRM-CM6-1-HR, MIROC6 and MRI-ESM2-0 – were used in this study, and the BCSD method was used to increase the resolution of the model. The results of the model tests and calibration show that the correlation coefficient between the simulation and the actual measurements is high, ranging from 0.73 to 0.99. The model's average monthly precipitation data can be used to calculate future drought indices.

The average monthly rainfall in the Central Highlands region from 2031 to 2060 is projected to decrease in both the SSP2-4.5 and SSP5-8.5 scenarios compared with the baseline period of 1985–2014. This projection is based on the CNRM-CM6-1-HR and MRI-ESM2-0 models. Additionally, two other models, BCC-CSM2-MR and MIROC6, indicate that there will be localized decreases in rainfall.

The results of the future simulations indicate that the intensity of drought will not change significantly. However, the drought deficits of SPI9 and SPI12 under the SSP5-8.5 scenario appear to increase during the period 2031–2060 compared with those under the SSP2-4.5 scenario, and the DD during 1985–2014.

The analysis of drought patterns from 2031 to 2060 under two climate change scenarios, SSP2-4.5 and SSP5-8.5, revealed that in the SSP5-8.5 scenario, the frequency of average current droughts, as projected by the BCC-CSM2-MR, CNRM-CM6-1-HR and MIROC6 models, is expected to increase from 2 to 4% for 3- and 6-month periods and from 4 to 6% for 9- and 12-month periods. The frequency of severe droughts is projected to decrease from 2 to 4% under the SPI6, SPI9 and SPI12 scenarios compared with that under the SSP2-4.5 scenario. Furthermore, compared with the base period of 1985–2014, the frequency of droughts is expected to increase significantly under both scenarios. Consequently, it is anticipated that moderate to severe droughts (−2.0 <SPI < −1.0) will occur more frequently in the Central Highlands of Vietnam in the future.

I would like to thank the Ministry of Agricultural Rural and Development (MARD), the General Statistics Office Vietnam and the Vietnam National Hydrometeorological Center for providing the information and database for this research. Additionally, I would like to thank the reviewers for carefully reviewing the manuscript and providing valuable comments to help improve this paper.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.