Comparative assessment of drought indices for evaluating drought patterns in Peninsular Malaysia

Drought is a phenomenon where a region does not receive the normal amount of rainfall for a prolonged period. Factors, such as lack of rainfall, overuse of water, overpopulation and human activities, may lead to drought. According to Gil et al. (2013), drought events have brought direct and indirect impacts to the human population, society and the agricultural sectors. The direct impacts include reduced clean water sources, increase in wildfire frequency, destruction of aquatic habitat and reduced crop yield, whereas the indirect impacts comprise reduced income for the agricultural sector, increment in food prices and increased unemployment rate. These impacts are interconnected, thus drought needs to be monitored. With millions of people relying on the agricultural sector as their main source of income, drought may impact agricultural production, posing a big challenge in preventing poverty and hunger in affected regions. In addition, an insufficient supply of crops also leads to price inflation. The Food Price Index by the United Nations Food and Agriculture Organization showed that food prices have been hiked up since 2011 due to insufficient supply of grains, vegetables, oils and butter as farmers around the world face an array of difficulties, which include drought (Coulibaly 2013). Moreover, drought may cause water shortage that leads to health issues, namely malnutrition, increased risk of infectious diseases, mental health issues and disruption of local health services, which are all interconnected and caused by limited availability of food, water supplies and sanitation (Palmer 1968). These issues reflect the importance of drought monitoring, which can help decision-makers or authorities in drought assessment and agricultural planning to improve resilience against drought.

Peninsular Malaysia is a humid tropical country that has two monsoonal seasons, namely, the southwest monsoon (SWM) and the northeast monsoon (NEM). These two monsoonal seasons occur from November to March (NEM) and from June to September (SWM), causing certain areas to experience high rainfall at certain periods and experience drier climate during other periods. Thus, Peninsular Malaysia is vulnerable to drought. Other than that, El Nino events have also been among the culprits of drought occurrence, namely the drought events in Klang Valley in 1997 and 2016 (Shaadan et al. 2015). Apart from the difficulties faced by the population due to water shortages, Peninsular Malaysia's economy took a big hit during those drought events due to the downturn in agricultural sectors (Sanusi et al. 2015). Hence, drought monitoring is essential in Peninsular Malaysia.

To assess drought conditions, drought indices are utilised as numerical representations of drought severity. Drought indices can be classified into a few categories based on their input data: (1) precipitation and potential evapotranspiration (PET)-based indices, such as standardised precipitation evaporation index (SPEI) Vicente-Serrano et al. (2010). and Palmer drought severity index (PDSI) (Palmer 1965); (2) streamflow-based indices such as streamflow drought index (SDI) (Nalbantis & Tsakiris 2008); and (3) precipitation-based indices such as the deciles index (DI) (Gibbs & Maher 1967), reconnaissance drought index (RDI) (Tsakiris & Vangelis 2005) and standard precipitation index (SPI) (Edwards & McKee 1997).

Drought may occur at various temporal scales ranging from one to 24 months (Kam et al. 2014). The effect of different timescales on the performance of the drought indices differs under different geographical conditions (Prajapati et al. 2021). Hence, it is important to consider the timescales of drought indices while assessing the drought of a particular region to reduce operational error from time intervals between droughts .

With various drought indices developed in the past, the comparison between drought indices for specific study areas has been the main subject of recent studies. Liu et al. (2018) compared three drought indices that used the data of different natures (precipitation-based and evapotranspiration-based and precipitation-based respectively), namely the PDSI, SPI and SPEI based on their correlation with each other. Both SPI and SPEI performed similarly in terms of identifying agricultural drought trends and showed higher correlation coefficients compared with the PDSI. The study mentioned above successfully compared multiple drought indices in terms of comparison between indices on multiple timescales but lacked an accurate analysis of the indices based on historical drought events. Salimi et al. (2021) carried out a study by comparing three different types of drought indices, namely, precipitation-based, evapotranspiration-based and streamflow-based indices. Standard streamflow index (SSI) is used as a model index for correlation analysis with the performance of SPI and SPEI in drought assessment. Although this study examined the accuracy of various types of drought indices, it did not consider the impact of temporal variability in evaluating drought.

Of all drought indices, SPI has been used extensively on many occasions due to its ease of calculation which requires only precipitation data (Javed et al. 2021; Yaseen et al. 2021). Nonetheless, it is still important to evaluate the accuracy and temporal characteristics of various drought indices in Peninsular Malaysia as different indices' performance may vary in different geographical conditions. Far too little attention has been paid to comparing precipitation-based drought indices in terms of accuracy and temporal characteristics in tropical regions, especially in Peninsular Malaysia. Thus, this study aimed to assess five precipitation-based drought indices consisting of the SPI, DI, percent of normal precipitation (PNPI), rainfall anomaly index (RAI) and Z-score index (ZSI) at various timescales (one-, six- and 12-month) throughout Peninsular Malaysia from 1984 to 2018. The temporal characteristics of each index were compared by conducting a temporal trend analysis. Additionally, accuracy evaluation was carried out using SPI as the model index by analysing DI, PNPI, RAI and ZSI using a statistical analysis. The finding of the study will provide a better understanding of how each drought index performs on different timescales and determine an alternative precipitation-based drought index other than SPI for drought assessment in Peninsular Malaysia.

Peninsular Malaysia is located between longitude 100° E and 119°E and latitude 1°N and 7°N. It is surrounded by sea and located near the equator. Its climate is categorised as equatorial, as it experiences a warm and humid climate throughout the year. In addition, Peninsular Malaysia also experiences monsoon seasons which lead to drought or flood, namely the NEM, SWM, the first inter-monsoonal cycle and the second inter-monsoonal cycle.

All five indices were computed using precipitation data of one-month (short-term), six-month (mid-term) and 12-month (long-term) timescales to analyse the trend and accuracy of the indices.

For the computation of DI, monthly precipitation values were ranked from the highest to the lowest to form a cumulative frequency distribution. The cumulative frequency is split into ten parts of deciles which are then grouped into five classes with two deciles per class. Deciles 1 and 2 are classified as much below normal if rainfall values fall into the lowest 20%. Deciles 3 and 4 are classified as below normal with between 20% and 40% rainfall values while deciles 5 and 6 indicate near-normal rainfall of between 40% and 60%. Deciles 7 and 8 are between 60% and 80% above normal and deciles 9 and 10 are the wettest 20%.

SPI is a drought index widely used in humid and sub-humid regions to characterise drought events due to its accuracy and simplicity (Mondol et al. 2017). In this study, SPI was used as the standard model to evaluate the performances of the other drought indices and indicate which drought index is most suitable to be applied as an alternative index to SPI (Khanmohammadi et al. 2022). Three statistical tests, such as mean absolute percentage error (MAPE), root mean square error (RMSE) and mean absolute error (MAE), were applied in this study to evaluate the accuracy of drought indices.

This section is divided into two subsections which include temporal analysis and performance evaluation of drought indices. Temporal analysis is divided into two sub-subsections: average moving range (MAE), and MK and Sen's slope estimator.

The temporal evaluation of all five drought indices was done to help us understand the temporal variation of drought events in Peninsular Malaysia. For this case, AMR was used to quantify the degree of fluctuations at the different prescribed timescales, which are one-, six- and 12-month. The AMR values represented the mean values between two moving averages of monthly drought index values in a series. Higher values of AMR indicated a higher sensitivity of drought indices. Table 2 shows the AMR of the SPI, PNPI, DI, RAI and ZSI on one-, six- and 12-month timescales, respectively.

Based on Table 2, a comparison of each drought index on different timescales revealed that all drought indices experienced the highest sensitivity on a one-month timescale, with SPI-1 ranging from 1.04 to 1.18, PNPI-1 ranging from 45.79 to 71.90, DI-1 ranging from 2.93 to 3.46, RAI-1 ranging from 3.29 to 3.74 and ZSI-1 ranging from 0.00 to 1.19. From this it was concluded that short-term precipitation data (one-month) are more responsive to the fluctuation of dry and wet conditions in short-term periods. Similarly, all drought indices showcased lower fluctuation when timescales increased as all indices have the lowest AMR value when deploying 12-month rainfall values, clearly showcasing long-term drought features. The findings are consistent with Fung et al. (2020), who revealed drought indices on one-month timescales had the highest fluctuation, indicating that a series on smaller timescales has lower stability.

Trends of dry spell events over the 25 years (1984–2018) were analysed in this study. The MK test and Sen's slope estimator were conducted to assess temporal analysis for five different drought indices at 11 rainfall stations over the study period. Z-value indicates the presence of an upward (positive) or downward (negative) trend. The significance level is set to be 95%, and if the P-value is lower than 0.05, the trend is then statistically present in the time-series data. Sen's slope estimator indicates the magnitude of the increasing or decreasing trend of every drought index on each timescale. The trend analysis results are presented in Table 3.

As shown in Table 3, most stations (Ipoh, Bayan Lepas, Kuala Krai, P.Ter Haiwan Tanah Merah, Pos Blau and Subang) showed an increasing trend for SPI on one-, six- and 12-month timescales. A significant increasing trend can be observed in both SPI-6 and SPI-12 in Bayan Lepas, Ipoh and Subang stations. P.Ter Haiwan Tanah Merah and Pos Blau exhibited a significant increasing trend for SPI on all timescales except for SPI-1. Of all meteorological stations, SPI-12 at Subang station exhibited the highest trend magnitude with a Sen's slope of 0.0047. Kuala Betis station was the only station that exhibited negative trends in both SPI-6 and SPI-12. No significant trend was found in any drought indices for stations Senai, Kota Bahru and Batu Pahat.

Based on the results obtained from temporal analysis, all drought indices had similar outcomes and indicated that six out of 11 stations (Bayan Lepas, Ipoh, Kuala Krai, P.Ter Haiwan Tanah Merah, Pos Blau and Subang) exhibit increasing trends on one-, six- and 12-month timescales. This indicated an increasing trend of wet conditions in those regions. Other than that, drought indices in Kuala Betis exhibited both increasing and decreasing trends on different timescales. When drought indices were computed on a one-month timescale, SPI, ZSI, RAI and DI exhibited an increasing trend while all five indices exhibited a decreasing trend when computed on six- and 12-month timescales for Kuala Betis station. This suggested that drought indices on a one-month scale are more sensitive to short-term variations in rainfall. When these indices are calculated on longer scales such as six- and 12-month, the short-term variation may be averaged out, revealing a more accurate representation of the longer-term patterns. Other than that, Pos Gob is the only rainfall station to exhibit a decreasing trend, with a Z-value and Sen's of −0.0767 and −0.0013, respectively, for ZSI-6. Although only ZSI-6 exhibited a significant decreasing trend, other indices especially on six- and 12-month scales also showed marginally significant trends. While these indices did not meet the 0.05 threshold, it is possible that with more data or a larger sample size, these trends could become statistically significant. This suggested that the Pos Gob region is more likely to experience drought, which is supported by the occurrence of multiple severe droughts in the past in that region (Tan et al. 2017).

RMSE, MAE and MAPE were applied to determine the highest accuracy drought index by comparing them with SPI. The drought index that obtained the lowest-measure values among the other drought indices would be chosen as the most accurate drought index.

Table 4 shows the performances of PNPI, DI, RAI and ZSI on one-, six- and 12-month timescales. Statistical tests used in this study evaluate drought indices based on error, for example, if a drought index obtains a lower score on any of the statistical tests (RMSE, MAE and MAPE), it indicates that the index has higher accuracy and vice versa. Although all drought indices use precipitation data as their only parameter, ZSI obtained the lowest RMSE, MAE and MAPE values for all stations on every timescale, which indicated that it acquired the highest accuracy among the other drought indices. Like SPI, ZSI indicates the desertion of moisture conditions by computing a standard deviation above or below the mean precipitation value of a data series. However, the indices should not be confused with each other as precipitation data do not require to be fitted into a probability distribution in the computation of ZSI. With the similarity in the performance of ZSI and SPI and the addition of ease in calculation, ZSI is regarded as the best alternative drought index (Dogan et al. 2012). In comparison, PNPI showcased weaker accuracy than SPI, obtaining the highest RMSE, MAE and MAPE values. This revealed that PNPI was not suitable to identify droughts. The DI obtained the second lowest RMSE, MAE and MAPE values followed by the RAI. This finding has been supported by Myronidis et al. (2018) and Shahabfar & Eitzinger (2013), who revealed that the Z-score is a better tool for monitoring droughts.

This study focuses on drought assessment by developing five drought indices (SPI, PNPI, DI, RAI and ZSI) on timescales of one-, six- and 12-month from 1994 to 2018 throughout Peninsular Malaysia. Comparisons between all drought indices were made based on their behaviour via temporal analysis. AMR shows that all drought indices fluctuate more by short-term rainfall (one-month). In comparison, the sensitivity of drought index fluctuation decreases when long-term rainfall is used, as the smallest AMR values are detected for drought indices on a 12-month timescale. Other than that, the MK test and Sen's slope estimator were used to investigate the trend magnitude of drought indices. The results of MK showed that most drought indices performed similarly in terms of trends on different timescales. Stations such as Senai, Kota Bahru and Batu Pahat did not exhibit any significant trend for all drought indices.

Moreover, the results from the accuracy evaluation showed that ZSI is the best alternative drought index due to its similar performance to SPI. The results from MAE, MAPE and RMSE indicated that ZSI has the highest accuracy on all timescales (one-, six- and 12-month). Other than that, the calculation of ZSI is less complex than the other indices and is even more suitable to be applied in regions with limited access to precipitation data (Jain et al. 2015). Combined with the results obtained from temporal analysis, SPI and ZSI can be applied interchangeably for drought assessment in Peninsular Malaysia.

In addition, it is recommended to compare drought indices that take into account PET's impact, such as the standardised precipitation evapotranspiration index (SPEI) in future studies since the drought indices evaluated in this study only use precipitation data. To top that off, a drought index such as the SDI is recommended to be compared for a better understanding of drought characteristics in Peninsular Malaysia. Since several stations in this study are located in Kelantan State, it is recommended to include more stations in other locations throughout Peninsular Malaysia for drought assessment.

The authors would like to thank the Malaysian Meteorological Department (MMD) for the provision of the meteorological data.

All authors equally contributed to the preparation of this manuscript. All authors read and approved the final manuscript.

The study was funded by the Malaysian Ministry of Higher Education through the Fundamental Research Grant Scheme (FRGS/1/2021/TK0/UCSI/03/3).

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

The authors declare there is no conflict.