Choosing the appropriate Coordinated Regional Downscaling Experiments South East Asia (CORDEX-SEA) Model for drought future hazard assessment in the Bintan Island

Management for sustainable water resources requires attention to future climate variability to anticipate the hazards that may arise, such as drought and flood. CORDEX-SEA is a downscale result of several gridded climate models that provide historical data and future projections for the Southeast Asia region useful to predict future climate, including the extreme event’s potential for deriving hazards. This research was done to determine which CORDEX-SEA climate model would be most suitable for predicting future drought risks, especially for small islands. This study uses statistical tests (probability density functions, skewness, etc.) in the historical period to determine the climate model that best fits the observed data, using Bintan Island as the case area. The model with historical data that best fits the observations will be considered the best model for predicting future (drought and flood) conditions. The corrected MPI and ensemble model of CORDEX-SEA showed well results in representing the drought index.


Introduction
Hydrometeorological disasters happen more frequently today, which are suspected to be a consequence of climate change, which we envisage to occur more often and increase in intensity in the future.Hydrometeorological disasters may cause substantial losses and are becoming more frequent due to climate change, and we estimate that they will increase in frequency and intensity in the future [1; 2].Drought has reportedly led to significant economic and human losses in recent decades.Unlike other hydrometeorological catastrophes like floods and storms, the effects of a drought take more time to 2 develop.These droughts develop gradually and without warning, and are usually only noticed until there are issues with water shortages, agricultural failures, etc [1; 4; 5].Consequently, identifying, quantification, and minimizing the effects of drought are significant challenges.Therefore, good information on the behavior of future rainfall is beneficial for anticipating potential future disasters as early as possible [6].Future rainfall datasets are now available, resulting from the development various climate change models, both global climate (GCM) and regional climate (RCM).For the Southeast Asia region, CORDEX-SEA (Coordinated Regional Climate Downscaling Experiment for Southeast Asia) is now providing a data series of climate models, both historical  and future projections (up to 2100), compiled through the Regional Climate Downscale Project.Southeast Asia (SEACLID).This project was a collaborative regional project on climate change risk reduction involving many collaborators from Southeast Asian countries, such as Malaysia, Indonesia, , Thailand, the Philippines, Cambodia, Laos, and Vietnam.The Regional Climate Downscale Project downscaled several models, and BMKG provides five models that best represent the climate in Indonesia.However, each model shows different data patterns with advantages and disadvantages that affect the level of uncertainty.To minimize uncertainty, We select the model that is best appropriate to be used in an area by comparing it to historical observation data.We assume the most similar data is the slightest uncertainty.
We use Bintan island data in this study.Bintan island lies in the Malacca Strait, one of the small outermost islands of Indonesian territory.Recent studies [1; 7; 8] have been published, however no indepth research was done on the drought over Bintan Island.Due to its strategic location, the island is experiencing rapid population growth and economic development [7], leading to the risk of limited water resources threatening the island [7].The physical circumstance of this island also strengthens the climatic factors that affect rainfall.Previous research showed the high number of dry days on the Northwest Marine Islands [1; 8; 9].Several previous studies indicated that El-Nino and IOD (+) decreased annual and seasonal rainfall across all of Indonesia, including in western Indonesia near the equator., where the study area is located [7; 10].The reduction in rainfall is greater when IOD (+) and El Nino concurrently occur than when just IOD (+) or El Nino occurrences takes place [4].Small islands will experience significant drought issues due to their limited catchment areas and water storage capacities.The island's water resources' carrying capacity will be significantly reduced by drought.
The standardized Precipitation Index (SPI) and the difference between the model's data series and observation results that results from comparing the probability density function (PDF) curve and the magnitude of the statistical description parameter are used in this study to measure the potential for drought.
Due to the low spatial resolution (0.25 o ) for use on small islands where the area coverage is limited, spatial resolution improvements were made [11] through correction of CORDEX-SEA data against the higher-resolution CHIRPS data ( 0.05 o ).

Field of study
The Bintan Island, shown in Figure 1, stands between 0.8166° and 1.2516° N and between 104.2216 °and 104.6884°E. The Bintan island is strategically located at the intersection of the South China Sea, the Malacca Strait, and the Karimata Strait.According to RI Law No. 44 of 2007, in terms of the development of Special Economic Zone, Bintan island is one of the Republic of Indonesia's economic growth centers.The island cover a total area of about 1,173 km2 that classed into small island according to UN classification.

Climate data series and data preparation
Five different models (CSIRO, CNRM, ENSAMBLE, EC-EARTH, and MPI) provided the downscaled CORDEX-SEA historical data (1981-2004) of RCP 4.5.The Indonesian Agency of Meteorology, Climatology, and Geophysics (BMKG) provided the data.We use daily temporal and 0.05• spatial resolutions CHIRPS rainfall satellite dataset to improve the resolution for CORDEX-SEA.We obtaided CHIRPS data from The Famine Early Warning System Network's (FEWS NEThttps://data.chc.ucsb.edu/products/CHIRPS-2.0/).Before CORDEX spatial downscaling, we adjusted the daily CHIRPS dataset of the same position to Kijang station observation datasets for 1981 -2021 periods.We used Quantile method Bias Correctiin [11; 1] for downscalling and adjustment of the datasets.

Drought potential analyses.
We use The Standardized Precipitation Index (SPI) to analyze drought potentials of every climate dataseries.SPI 3, SPI 6 and SPI 12 was applied to ground observation dataseries and to corrected and CHIRPS's spatial adjusted CORDEX models.
We calculate SPI for 3, 6, and 12 months in this paper to support in water resource management planning.The 3-month SPI result indicates the humidity associated with agriculture in the short to medium term.The 6-month SPI analysis reflects the seasonal distribution of rainfall associated with river flow anomalies, which manifest later than agricultural droughts.The 12-month SPI associated with water resource availability can indicate a period of potential groundwater scarcity [12].
SPI is a normalization index that assesses the probabilities of observed precipitation in relation to long-term rainfall at a certain station [12][13][14].SPI is determined based on the total cumulative probability of rainfall events occuring at the station for at least 30 years.SPI is calculated based on the gamma distribution's probability density function (PDF) to match the frequency distribution of rainfall amounts at each station.30 years of station rainfall data has been entered into the Gamma distribution.[13].To match the frequency distribution of rainfall amount at each station, SPI is calculated using the probability density function (PDF) of the gamma distribution.A 30-year period's worth of station rainfall data would be examined using the gamma distribution [15; 12].While a higher SPI value shows an abundance of rainfall, a lower SPI score shows insufficient amount of rainfall.SPI value calculation using gamma distribution [ 1 2 ; 1 5 -1 7 ] is as follows: The gamma probability density function is matched to the frequency distribution of the amount of rainfall for each station for the purpose of the SPI calculation.
The equation for optimizing the estimated values of α and β is as follows: n = amount of observation data Since x = 0 has no effect on the gamma function, q = P(x=0) > 0, where P(x=0) is the probability of no rainfall.Consequently, the cumulative probability distribution function is: With: q = number of rain occurrences = 0 (m) / amount of data (n).where q is the probability of a zero, and is estimated by m/n, in which m is the total number of zeros in a time series of precipitation n For 0 <H(x) ≤ 0.5, Z or SPI calculation is : Where t= ඨ 1 For 0.5 <H (x) ≤ 1.0, the SPI value calculation is Where t= ඨ 1 With: c0 = 2.515517, c1 = 0.802853, c2 = 0.010328, d1 = 1.432788, d2 = 0.189269, d3 = 0.001308 The standard random variable Z is then contructed using the cumulative probability of H(x) and it has an average value of 0 and a variation of 1.The SPI value is the result of z.The SPI values represent conditions comparative to the annual rainfall average.To examine drought events at various time scales, the index can be calculated computed for a number of time periods between 1 and 48 months..In this study, we calculated SPI at three, six, and twelve months (Table 1).If the SPI is positive, it indicates that the rainfall is still above average.If the SPI number is negative, it indicates that there will be less rain than usual.The degree of drought or wetness in an area during the current year can be classified using one of the various SPI drought indices value categories.

Similarity analysis
The comparison of probability density functions between the two groups for evaluation is one of the statistical techniques used in distribution similarity testing.A fundamental statistical concept is the probability density function [18; 19; 20].Consider into the probability density function of any random variable X.A bell-shaped probability density function graph is shown.The probability of the outcome of the specified observation is given by the area that lies between any two specified values.To determine the probability values associated with a continuous random variable, we solve the integral of this function.By defining the function, one may naturally describe the distribution of X and use the relation to estimate the probabilities connected to X.
We      The PDF analysis results for SPI 3, 6, and 12 are shown in Figures 5, Figure 6, and Figure 7, also Tabel 2. According to the graph on Figure 5-7, the climate model that tends to be wetter than the observation is EC-Earth, while the drier climate model is CSIRO.Table 2 displays the delta of the PDFs between the SPI index of all the climate model and the SPI Index of observation for SPI 3, 6 and 12.A small delta value indicates that the climate model's SPI index value is similar to the observed SPI index value.The higher the delta value, the less similar the SPI Index value of the climate model is to the SPI Index value of the observation.Tables 3, Table 4, and Table 5 show the total deltas of other statistical parameters and pdf for SPI 3, 6, and 12. Table 3. show the delta of pdf value and statistical deccription between SPI-3 index of all climate model and observation.Table 3. show SPI-3 MPI model is the most similar with SPI-3 observation.Table 4. Show SPI-6 Ensamble model is the most similar with SPI-6 observation.Table 5. Show SPI-12 EC-Earth model is the most similar with SPI-12 observation.The greater the delta value, the greater the SPI Index value indicates the value of the SPI index of observations.The smaller the delta value, on the other hand, indicates that the model SPI index value is more similar to the observed index value.3, and Table 4.

Discussion
Southeast Asian countries will experience an increase in drought or a shorter rainy season as more extended a longer dry season in the future [23].The climate experience on the small island of Bintan Island in Indonesia's western marine islands will be the same.Additionally, the island of Bintan's rock basement is mainly composed by impervious granite rocks, which limit the storage of water resources.
Planning for the water resources management in the future can help Bintan Island avoid future water scarcity disasters., the drought probability forecasting is essential.With the above shortages, future water resource management planning on Bintan Island needs to predict the potential for future droughts to reduce disasters caused by drought in the future.
The CORDEX-SEA Regional Climate currently provides models, among which five models: MPI, CNRM, CSIRP, EC-Earth, and ENSEMBLE, could be used to predict the future rainfall in the region.However, we must assess the use of the current regional climate model firstly because the uncertainty of climate models is high [1; 22; 23].No current climate models perform well [21; 24; 25], so determining the best model for a given area is challenging.Several previous studies have assessed climate models and determined that the CNRM model best fits the Johor region of Malaysia [1].While research [15] indicates that the ENSEMBLE model is the best model for Southeast Asia.This study on Bintan Island found that the rainfall model having the highest similarity to observation rainfall data was MPI during the historical period [22].According to previous research, the climate model's uncertainty is high.Because there is no consistent model for each region, an assessment remains required before using this climate model in a specific area.
We conducted research from 1981 to 2005 using five models (CSIRO, CNRM, ENSAMBLE, EC-EARTH, and MPI).We analyzed the five climate models during the historical period, using both uncorrected and corrected models.Our objective is to find out whether the climate model should be corrected or not before analysis.The results showed that using the MPI and ENSEMBLE corrected models to predict future drought potential on Bintan Island is highly recommended (Table 6.).We assessed the potential for drought using SPI 3, 6, and 12 months and obtained that the MPI and ENSEMBLE corrected models have SPI values closest to the observation value.As a result, using the MPI and ENSEMBLE corrected the CORDEX-SEA climate model to understand the potential for a drought on Bintan Island in the future, prevails highly recommended.

Conclusion a)
The CORDEX-SEA Climate Model needs bias correction to observational data before its utilization as a future predictor.b) MPI and ENSEMBLE are the most suitable CORDEX-SEA climate models for predicting future drought at Bintan Island and are recommended for small island.

Figure 1 .
Figure 1.The field of study, Bintan Island.
3, and Figure 4 show the comparison of SPI 3, SPI 6, and SPI 12 of the uncorrected and corrected CORDEX_SEA climate models of the historical period with the SPI of the Kijang observation of the years 2000 to 2005.Compared to the observation data, the SPI value of the biasuncorrected CORDEX-SEA climate models has a reasonably significant bias.The bias value between the climate model and the observation data decreased after climate model correction.

Figure 2 .
Figure 2. SPI 3 of uncorrected and corrected model of CORDEX_SEA at history period.

Figure 3 .
Figure 3. SPI 6 of uncorrected and corrected model of CORDEX_SEA at history period.

Figure 4 .
Figure 4. SPI 12 of uncorrected and corrected model of CORDEX_SEA at history period.

Figure 7 .
Figure 7. Probability Density Function Calculation of SPI 12 of CORDEX_SEA at history period.
test the similarity of every CORDEX model to Ground Observations by the comparison of probability density functions[1; 22].Beside, we compare also statistical descriptions such as mean, maximum, minimum, standard deviation, kurtosis, skewness, and range, of every CORDEX models to Ground observation.Statistical descriptions are used to evaluate a set of SPI index values for climate models, which are then compared to a set of SPI index values for observations.The goal is to observe or test the data group's similarity.The two groups of climate model SPI index values and observational SPI index values are similar if they have a small delta of descriptive statistical parameters.The lower the delta value, the more similar the two groups.

Table 6
shows the total Delta values of statistical parameters and PDF for all models and SPI 3, 6, and 12 values as the final output of assessing the best climate model for the future drought potential predicting of Bintan Island for water resources management supporting.Table 6 also shows that the MPI and Ensemble Climate models possess SPI values that best match the SPI values of the observation result on Bintan Island.The Delta of Descriptive Statistic of SPI 3, 6 and 12 of CORDEX-SEA Model at history period shown in Tabel 2, Table

indeks between Climate Model with Observation Ensamble MPI CSIRO Ecearth CNRM Uncorrected Corrected Uncorrected Corrected Uncorrected Corrected Uncorrected Corrected Uncorrected Corrected
The Delta of descriptive statistic of SPI 3. Delta probability density function and descriptif statistic values for SPI 3, 6 and 12. Delta