Climate Change Impact on Rainfall and Temperature in Muda Irrigation Area Using Multicorrelation Matrix and Downscaling Method

647 © IWA Publishing 2015 Journal of Water and Climate Change | 06.3 | 2015

Climate change impact on rainfall and temperature in Muda irrigation area using multicorrelation matrix and downscaling method Nurul Nadrah Aqilah Tukimat and Sobri Harun

ABSTRACT

Statistical downscaling model was used to generate 30-year climate trend of Kedah – the state which Nurul Nadrah Aqilah Tukimat Department of Water Resources & Environmental, has the largest cultivation area in Malaysia, resulting from climate changes. To obtain a better Universiti Malaysia Pahang, Gambang, predictors set, multicorrelation matrix analysis was added in the climate model as a screening tool to Pahang, Malaysia explain the multiple correlation relationship among 26 predictors and 20 predictands. The Sobri Harun (corresponding author) performance of the predictor set was evaluated statistically in terms of mean absolute error, mean Department of Hydraulic and Hydrology, square error, and standard deviation. The simulation results depict the climatic changing trend in this Universiti Teknologi Malaysia, Skudai, region in terms of temperature, rainfall, and wet and dry length compared to historical data captured Johor, Malaysia W from 1961 to 2008. Annual temperature and rainfall depth are expected to increase 0.2 C per decade E-mail: [email protected] and 0.9% per year, respectively, from the historical record. The months of November and January are expected to receive the highest and lowest rainfall depth, respectively, because of the two monsoon seasons. The wet spell is estimated to be from May to November in the middle of Kedah. The annual dry spell shall be from January to March, and is expected to shorten yearly. Key words | climate projection, multiple correlation, rainfall, statistical downscaling, temperature

INTRODUCTION

Climate refers to the average weather recorded in a year; it is In Malaysia, global warming has a serious impact on the generally affected by geographic factors such as the oceans land and water resources, agricultural activities, and hydro- and the altitude of the region. Changes in climate occur when logical cycle. As claimed by the Department of Irrigation the accumulated heat absorbed by the existing greenhouse and Drainage, Malaysia, more than 70% of water use in gases, such as nitrous oxide, oxygen, methane, water vapor, Malaysia is allocated for irrigation purposes and the remain- carbon dioxide, and tropospheric ozone, increases and acceler- ing 30% is for domestic, industry, and other demands. ates global warming. As reported by the Intergovernmental Unfavorable calamitous events caused by climate changes Panel on Climate Change AR4 (IPCC ), the global average jeopardize the agricultural sectors by, for example, affecting surface temperature for the past 100 years has increased from the cultivation growth, destroying machinery, inducing 0.6 W C (1901–2000) to 0.74 W C (1906–2005). The World Meteor- losses, and affecting agricultural production. Data have ological Organization claimed the year 2010 as the warmest shown that a previous calamity on the irrigation area year where an increase of 1.2–1.4 W C had been recorded particu- during years 1989–2010 destroyed thousands hectares of larly in Africa, parts of Asia, and parts of the Arctic. Since 1961, paddies, rubbers, and vegetables in several states in Malaysia more than 80% of the heat had been added to the climate such as Perlis, Kedah, and Penang which then escalated into system, causing the average temperature of the ocean to food scarcity problems and millions Ringgit losses. increase at 3,000 m depth; the rate of observed sea level rise In terms of annual precipitation, Bates et al. () is estimated to be 0.17 m in the 20th century. have proved that the volume of water resources for our

doi: 10.2166/wcc.2015.015

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 648 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

society and ecosystems is strongly correlated to climate et al. (), Guttman et al. (), Lopes (), and Spak change. Unusual warms change the water volume stored et al. (). in reservoirs due to unpredictable rain availability, timing, and water quality. In the United States, Izuka et al. () stated the storage of the Garlinghouse Tunnel had METHODOLOGY decreased around 50% since the 1980s; the storage depends fully on the precipitation and river runoff during Climate model wet periods. When global warming escalates, it increases the water evaporation rate and may easily lead to water Statistical downscaling is analogous to the model output storage loss of more than 20% of the average annual statistic and perfect program approach used for short- runoff. This becomes especially worrying since the valley range numerical weather prediction (Wilby & Dawson width is impounded and has larger water open area ). The model was developed by Robert L. Wilby and (FAO ). This also means that water storage has Christian W. Dawson from the United Kingdom and uses turned into an uncontrollable factor when rainfall patterns a weather generator method to produce a multiple realiz- become unpredictable, and the compounded effects are felt ation of the synthetic daily weather sequence. This by all sectors. software calculates the statistical relationship based on mul- For the agriculture sector, more uncertainties in climate tiple regression techniques between large scale (predictor) make it increasingly difficult to predict and plan the irriga- and local climate (predictand). Table 1 gives the list of pre- tion demand even for the following month. New dictor and predictand combinations used in this study. approaches are therefore needed to gauge changes in rain- These relationships were developed using observed fall directions, duration, trending, and depth. This has weather data and previously captured relationships among been largely addressed using various climate models devel- GCM-derived predictors. This produces the maximum, oped to primarily understand the changes in present/ mean, and minimum temperature, the precipitation and future climatic conditions with due consideration on green- humidity of site-specific daily scenarios for a selected house gases and aerosols emission (Goosse et al. ). Such region, and the range of statistical parameters such as var- models also assist in evaluating regional surface water iance and frequencies of extremes. The SDSM downscaling response and reservoir system capacity to meet local used two types of data, viz. predictand and predictor at the demands (Yano et al. ). grid box of 28X × 33Y. This case study introduces a statistical downscaling The historical rainfall data (1961–2008) recorded at 20 model (SDSM) which has been used to generate the cur- locations were used as the predictands while the National rent/future local climate trend. SDSM models have been Center for Environmental Prediction (NCEP) reanalysis data widely applied for hydrological issues caused by climate and GCM outputs of the Hadley Center General Circulation scenarios because it provides station-scale climate infor- Model (HadCM3) under A2 scenario were used as the predic- mation at general circulation model (GCM) scale between tor to simulate climate trend. The HadCM3 model was a atmospheric circulation pattern (predictors) and local-scale modified version of HadCM2, done to improve the accuracy parameters (predictands) using multiple regression tech- of the climate projection results without the application of niques. It is a popular tool among researchers since it can flux adjustments. It has wider coarse spatial resolutions of portray an easily understandable relationship pattern 2.5W × 3.75W (latitude by longitude) which can be applied in between predictor and predictand. Besides, the model does many climatic regional studies. Samadi et al. () consider not require high computational demand to view the simu- HadCM3 to be the best GCM model, i.e., superior to other lation result because the output is presented in finer models such as CCSNIES, CSRIO (Australian Government), resolutions. To conclude, it is cost-effective and gives satis- and Geophysical Fluid Dynamics Laboratory (GFDL). In factory climate simulation with its capability and reliability this study, the A2 scenario had been chosen to give an upper proven by researchers such as Sharma et al. (), Khan bound on future emissions, and was selected from an

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 649 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Table 1 | List of predictor and predictand combinations

No. Predictor variables Predictor description Predictand variables (station) Predictor description

1 Mlsp Mean sea level pressure GM Gajah Mati 2 p_f Surface airflow strength IBT Ibu Bekalan Tupah 3 p_u Surface zonal velocity KP Kedah Peak 4 p_v Surface meridional velocity KT Keretapi Tokai 5 p_z Surface vorticity Kod Kodiang 6 p_th Surface wind direction KSS Kota Sarang Semut 7 p_zh Surface divergence Pen Pendang 8 p5_f 500 hPa airflow strength SL Sungai Limau 9 p5_u 500 hPa zonal velocity TC Telok Chengai 10 p5_v 500 hPa meridional velocity LTP Ladang Tanjung Pauh 11 p5_z 500 hPa vorticity KN Kuala Nerang 12 p500 500 hPa geopotential height Kg.T Kg. Terabak 13 p5th 500 hPa wind direction Sik Sik 14 p5zh 500 hPa divergence Kg.LB Kg. Lubok Badak 15 p8_f 850 hPa airflow strength Kg.LS Kg. Lubok Segintah 16 p8_u 850 hPa zonal velocity KSS Kuala Sala 17 p8_v 850 hPa meridional velocity LH Ladang Henrietta 18 p8_z 850 hPa vorticity SG SM Gurun 19 p850 850 hPa geopotential height Jit Jitra 20 p8th 850 hPa wind direction AP Ampang Pedu 21 p8zh 850 hPa divergence 22 r500 Relative humidity at 500 hPa 23 r850 Relative humidity at 850 hPa 24 Rhum Near surface relative humidity 25 Shum Surface specific humidity 26 Temp Mean temperature

impacts-and-adaptation point of view – if a system could adapt that could forecast different climate trends at different to large climate change, it would have no problem with smaller locations. To screen, analyse, and select the predictor set climate change and a lower end scenario. The only setback is that could represent all rainfall stations would be a tedious that low emissions scenario will have less information when task. Besides, the rainfall pattern in Malaysia is non-uniform such is the case (NARCCAP ). and very sensitive to the GCM parameters that may inﬂuence Fundamentally, the screening process in the SDSM the rainfall depth. As such, though the MLR concept was model uses the multiple linear regression (MLR) concept to adopted, the focus was on multiple-predictand to multiple- express the predictor–predictand relationship, but it can predictor relationships, presented in a multiple correlation only analyse multiple predictors with a single predictand in matrix or M-CM analysis. The purpose was to measure and a single simulation run. With this, it poses a challenge in clearly state the empirical relationship between 26 predictors choosing the predictors that will correlate well with all cli- and 20 predictands. Figure 1 illustrates the schematic dia- mate stations (predictand) in a study region. In this study, gram of SDSM model that was adopted in this study (see there were 20 rainfall stations surrounding the study area Table 1 for full list of predictors and predictands).

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 650 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Figure 1 | Schematic diagram of SDSM analysis (Wilby & Dawson 2007).

Predictor selection in SDSM model – multicorrelation the correlation matrix rxy is: X matrix method 1 Cov(xy) ¼ (x y xy) (1) N i,j i,j The M-CM method is a multiple regression method famous for defining the input variables between historical seasonal ’ Cov(xy) average rainfall occurrence probabilities and GCM s simu- rxy ¼ qffiffiffiffiffiffiffiffiffi (2) 2 2 lated seasonal mean rainfall depth. It can measure the sxsy linear relationship between multiple dependent variables

and multiple independent variables, and therefore assist in ::: rx1,1y1,1 rx1,2y1,2 rx1,jy1,j gauging the interaction of data from two sets of variables rx2,1y2,1

and their inter-relationship. M-CM was applied in this (3) . study to screen the potential of multiple predictands and .

predictors and the associations between them. The concept rxi,1yi,1 rxi,jyi,j was to show the physical interpretation of the connection

between local surface climate (predictand) and possible pre- where xi,j and yi,j refer to the predictands and predictors data

dictors in multi-site. The generated correlation value shows at i and j raw; xy is the mean value of both variables, and sx

the percentage of variance that can be explained in the form and sy refer to their standard deviation (StD). Basically, the of multi-dependent variables using the multi-independent capability among variables was interpreted as values variables. It also gives the criterion variables (product inno- between 1 and 1 to show positive/negative association. vation variables) among the relationships. Thus, the The positive value illustrates how strong the predictand– predictor selection was based on the inter-variable corre- predictor relationship is in the transformation and vice lation value showed in m × m matrix form. The formula for versa.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 651 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Performance evaluations cultivate paddy – the staple food of Malaysians. Covering 97,000 ha, it is the largest double paddy cultivation area in To evaluate the calibration and validation performance, the Malaysia. mean absolute error (MAE), mean square error (MSE), and Geographically, the area lies between 5W 450–6W 300N lati- StD were computed between observed and simulated cli- tude and 100W 100–100W 300E longitude. The topography of the mate variables over the 100 simulations. The function of area is almost flat with a slope ranging from 1 in 5,000 to MAE is to measure the accuracy of continuous variables 1 in 10,000. The climate of the area, like other parts of through the average of errors between the two datasets. It Malaysia, can be classified into four seasons viz. south- is a relatively simple and common mathematic measure- west monsoon (May–September), north-east monsoon ment that has been widely employed in forecasts. In this (November–March), and two inter-monsoon seasons. case, the average error represents the general disparity of December–February and June–July are warmer seasons two datasets, measured through quadratic scoring rules. while April–May and September–November are humid sea- MSE measures the average squared difference between the sons. The type of soil in the study area is heavy clay in estimated and observed value (accuracy), and the variability nature. The mean temperature varies between 27 and of the actual data (precision). The difference between MAE 32 W C. The average rainfall depth is 199 mm/month and and MSE is that each error in MSE is presented in square 2,390 mm/year. The relative humidity in this area fluctuates value; a larger error has a greater influence on the total between 54 and 94%. The 20 rainfall stations that had been square error and vice versa (Willmott & Matsuura ). identified were chosen from the quality of available rainfall MSE provides partial information to choosing better estima- records and their location in the Muda Irrigation tors. On the other hand, StD measures the spread and Scheme area. To support irrigated water demand efficiently, distribution of a data estimation. It can show the range of two important reservoirs were built by the government to variation from the observed mean value for the entire data- store and supply water for the paddy cultivation area set. Smaller values of StD indicate that the majority of (96,558 ha), namely the Pedu Reservoir and the Muda Reser- estimation data are very close to the mean of the estimation voir. These systems are jointly called the Pedu-Muda dataset. The formulae of these mathematical statistics are reservoir (see Figure 2). presented in Table 2.

RESULTS AND DISCUSSION STUDY AREA Temperature simulation result The study area is the Malaysian Muda Irrigation Scheme under the Muda Agricultural Development Authority (MADA, The simulation of temperature data (predictand) referred to Kedah state), which is made up of important areas that the meteorological station at Alor Setar. It was assumed that the recorded temperature at Alor Setar meteorological station could represent the temperature trend at Kedah Table 2 | List of statistical test equations state. Contrary to the rainfall trend, the temperature differ-

Name Formula Description ence between districts in Kedah is very small and almost XN MAE 1 distributed uniformly across all surrounding states. The pat- (X X ) Average error between two variables N esti obsi i¼1 tern of temperature is also consistent even at different 1 XN MSE (X X )2 Average squared error between two locations. Because of the very small difference in tempera- N esti obsi i¼1 variables ture, the temperature reading at Alor Setar station was vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi StD u u XN Standard deviation taken to represent the temperature trend at Kedah. t 1 2 (Xi Xi) fi N In the SDSM screening results, ve predictors were i¼1 selected to generate the temperature trend at the study

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 652 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Figure 2 | Geographic map of Kedah state.

site: surface airﬂow strength (p_f); 500 hPa geopotential using predictor sets from NCEP for three conditions – maxi- height (p500); relative humidity at 500 hPa (r500); relative mum, mean, and minimum temperature. humidity at 850 hPa (r850); and mean temperature at 2 m The line graphs showed that the selected predictors (temp). The rationale was that the temperature record had were close to the monthly observed data. However, the esti- better interconnection with these ﬁve atmospheric charac- mated minimum temperature during the validation process teristics since it produced higher correlation values within was slightly lower than observed record during January to the range of 0.3–0.5 compared to the remaining 21 available September. Table 3 summarizes the performances of cali- predictors. Figure 3 shows the potential of selected predic- bration and validation results. The error value was within tors associated with the local temperature station in the range of 0.1 W C (October and December) to 0.9 W C calibrated (1972–1986) and validated (1987–2001) process (January). As for the max and mean temperature, the error

Figure 3 | Calibrated and validated results of temperature at station Alor Setar.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 653 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Table 3 | Performance of calibrated and validated results for temperature at station Alor Setar

Maximum Mean Minimum

Calibration Validation Calibration Validation Calibration Validation

r 0.9 1.0 0.9 0.9 0.9 0.9 MAE 0.5 0.1 0.3 0.4 0.2 0.2 MSE 0.4 0 0.1 0.2 0.1 0

range was only within the range of 0.1–0.4 W C. These results results are consistently similar to the peninsular Malaysia produced very small values in MAE and MSE in the whole climatic report by Meteorological Department Malaysia, analysis, ranging from 0.0 to 0.5 W C. The correlation values where the warmest season will be in December–January– had been estimated higher, recorded at 0.9 and 1.0 for cali- February (DJF) at the end of the century and the tempera- brated and validated analyses, respectively. This shows that ture increment will be 1.1–3.0 W C. the calibrated and validated values were in good agreement with historical record. Therefore, the projection analysis Rainfall simulation result results produced by the SDSM model are reliable and accep- table in this stage. The analysis of rainfall trend was more complex than the To generate the future temperature trend in this area, the temperature analysis. The reason is Malaysia has a non-uni- GCM output type HadCM3-A2 was used with the constant form distribution of rainfall which produces different water predictors set. These predictors represent the changes of quantity at different locations in the same day. Thus, the 20 the atmospheric pattern response to the greenhouse gases rainfall stations selected surrounding Kedah state became effect based on the level of regional development. The temp- significant in recognizing the rainfall distribution pattern erature simulation and projection trend are showed in of Kedah. While selecting the associated predictors for Figure 4. The future temperature shows that the monthly these rainfall stations, M-CM analysis was added into the temperature reading is estimated to increase in all tempera- SDSM model to screen the relationships among the 26 ture conditions; minimum (þ1.8%), mean (þ3.6%), and predictors and 20 predictands already presented in a corre- maximum (þ4.5%). The annual mean temperature in the lation matrix form. The purpose was to view the physical future (2040–2069) is expected to increase to 28.3 W C with interpretation of the connection between local surface cli- þ0.2 W C from the current temperature reading in year mate (predictand) and multiple sites in order to identify 2010. A higher temperature is predicted in February and better predictors for climate projection. March; this may be affected by the interchange of north- Table 4 shows that five predictors have been selected to east monsoon to the south-west monsoon. Therefore, it develop the statistical relationship between the local and can be concluded that the average temperature of the regional scale of climate association, i.e., 500 hPa zonal vel- study area will continue to rise by 0.2 W C per decade. The ocity (p5_u); airflow strength (p_f); 500 hPa relative

Figure 4 | Simulated and projected (years 2040–2069) temperature trend at Alor Setar meteorological station.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 654 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Table 4 | Correlation values between rainfall station and ﬁve climate variables

p8_v r500 p_f Shum p5_u p8_v r500 p_f Shum p5_u

IBT 0.13 0.13 0.15 0.17 0.17 TC 0.17 0.13 0.20 0.17 0.18 KT 0.01 0.09 0.02 0.12 0.08 KLB 0.01 0.12 0.02 0.13 0.10 Kg.T 0.15 0.12 0.17 0.16 0.17 KP 0.04 0.05 0.05 0.03 0.02 LTP 0.13 0.12 0.15 0.17 0.18 Jit 0.09 0.06 0.12 0.12 0.10 AP 0.17 0.14 0.18 0.15 0.17 Kg.LS 0.02 0.07 0.03 0.05 0.05 SL 0.18 0.12 0.19 0.17 0.20 KN 0.10 0.12 0.11 0.17 0.14 GM 0.10 0.11 0.13 0.17 0.15 KS 0.12 0.07 0.16 0.13 0.10 KOD 0.14 0.10 0.16 0.16 0.16 LH 0.05 0.04 0.08 0.10 0.07 KSS 0.16 0.13 0.18 0.17 0.19 SIK 0.03 0.11 0.06 0.12 0.11 PEN 0.12 0.11 0.15 0.18 0.18 SG 0.08 0.04 0.07 0.10 0.02

humidity (r500); 850 hPa meridional velocity (8_v); and In general, the rainfall distribution is non-uniform specific humidity (shum). The combination of these five because each location receives different rainfall depth, but selected predictors successfully modeled the relationships within the range of 5–30 mm/day or 1,645–3,335 mm/year. with the local stations, with a correlation range of between To evaluate the predictor set’s performance, Table 5 shows 0.08 and þ0.20, and was better than the remaining 21 pre- the MAE, MSE, and StD for each rainfall station between dictors. Figure 5 shows the calibrated and validated results the historical data and simulated results. As expected in cor- for 20 rainfall stations in this region. Results showed that relation results, stations Kg.LS, SIK, Kg.LB, and SG had most stations had been calibrated and validated close to higher errors in the simulated results of more than observed values except at a few rainfall stations, namely 2.0 mm/day in MAE and 7.0 mm/day in MSE. The error IBT, KT, SL, SIK, SG, and Kg.LS. These were attributed to was around ±20%, and might have been caused by the the underestimated correlation of the predictor–predictand frail association with p_5u and 8_v due to the smaller coeffi- relationship. These areas are expected to have poor vali- cient values produced at these rainfall stations. The dation during February, March, April, and December; this attributes of U and V in the atmospheric parameters rep- will influence the pattern of climatic trend in future years. resent the direction of wind speed that would directly Even though the simulated rainfall did not correlate well influence the advection moisture at the region (Jinqiang & with the observed rainfall, the simulated patterns were Simona ). Report by the Meteorological Department able to consistently preserve the historical pattern through- Malaysia (MDM, Malaysia) has stated that the rainfall out the year. Thus, it is concluded that the selected amount in Malaysia is influenced by the strength of the predictors can accurately simulate the rainfall with local wind flow coming from western Pacific toward the South predictands. China Sea during the winter monsoon. Therefore, these pre- Since these predictors correlated well with 13 other dictors sets were still used to develop the future climate rainfall stations, for example, the simulation of GM, trend at this region. KSS, Kod, LTP, Pen, KS, Kg.T, KN, KS, and AP were Based on the StD results, the simulated value spread very close to observed values with minor errors, they was to the mean of historical data. The biggest value was were retained. And besides, atmospheric characters 2.2 mm/day, which was obtained from station SIK, and such as specific humidity, airflow indices, zonal velocity, the lowest value was 0.4 mm/day, recorded at station AP and relative humidity have always been used as precipi- and Kod. The results illustrated that the rainfall stations esti- tation predictors (Crane & Hewitson ; Goodess & mated correlated well with the observed data, except for Palutikof ; Wilby & Wigley ; Willems & Vrac stations Kg.LS, SIK, Kg.LB, and SG. It would appear that ). the predictor selection based on M-CM analysis was

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 655 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Figure 5 | Validation result between observed and simulated data for 20 rainfall stations.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 656 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Table 5 | Statistical performance comparison MAE, MSE, and StD (mm/day)

Rainfall station Kg.LB AP GM IBT Jit Kg.LS Kg.T Kod KP KSS

MAE 2.4 0.6 0.9 1.9 0.9 2.0 1.4 0.5 1.9 0.8 MSE 6.7 0.8 1.1 5.2 3.5 10.8 2.2 0.5 5.2 1.2 StD 1.7 0.4 0.7 1.3 0.7 1.4 1.0 0.4 1.3 0.6 Rainfall station KT KN KS LH LTP Pen SIK SL SG TC MAE 1.8 1.0 1.1 1.4 0.6 0.8 3.1 1.6 2.1 1.2 MSE 5.2 1.7 1.2 3.4 0.6 1.1 10.9 3.7 7.3 1.9 StD 1.2 0.7 0.7 1.0 0.5 0.6 2.2 1.1 1.5 0.8

satisfactory. The error produced was also low and within reasonable range at most rainfall stations. The accuracy of the simulated data also proved that the selection of predictors at each location was important for the calibration process. These predictors shall be used to develop climate trend for future years. Figure 6 shows the projection of annual rainfall distribution from year 2040 to 2069 using the GCMs model which represented the physical atmosphere using numerical data. The future trending was generated at every rainfall station using the predictor set provided by the HadCM3- A2 scenario. As can be seen in the map (Figure 6), the maximum annual rainfall is estimated to reach 4680 mm/year at Kuala Muda and west of Sik district. The rainfall depth is estimated to spread from Kuala Muda to the nearby areas such as Pendang, Sik, several parts in Kota Setar, Padang Terap, and Baling. Most of the districts are anticipated to experience higher rainfall amounts from the previous interval year, except at Yan and west of Pendang. The minimum rainfall amount is predicted to be about 1,900 mm/year, an 11.0% increase from the historical minimum rainfall. The average annual rainfall for this interval year is predicted to be 2,944 mm/year, a 22% rise compared to the historical year. The monthly rainfall of this year is predicted to be consistent with the historical year, but with high monthly rainfall depth. November is expected to receive the highest Figure 6 | Annual rainfall trend at Kedah state during years 2040–2069. depth compared to September (this is the peak month in historical data). The main reason for this transition of trend will Rainfall projection at the Pedu-Muda reservoir area be the wind during the north-east monsoon in east Malaysia. The minimum rainfall is predicted to fall in January The projection of climate for the purpose of better reservoir (16 mm/month) at station Kodiang and the maximum rain- management was based on stations 61 and 66, which rep- fall to be in November (892 mm/month) at station Ibu resent the Pedu reservoir and Muda reservoir areas, Bekalan Tupah. respectively. Generally, the monthly rainfall trend at these

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 657 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

regions is predicted to be different because of the reservoir Future wet length is estimated to be longer than the his- locations, even when both reservoirs are only 7 km apart. torical record at several months. Figure 8 illustrates the wet Figure 7(a) and 7(b) shows the average monthly rainfall length distribution in the region on average from year 2040 trend during years 2040–2069 at stations 61 and 66, respect- to 2069. The frequency of rainy days is quite similar to the ively. The projection result in station 61 shows that the historical pattern, whereas the rainy season comes from future rainfall depth will drop, except in August and Novem- May to November and the peak time is from September to ber. The highest and lowest concentration will be in August October (mostly 30 days). Then, it is expected to decrease and January, respectively; this is different from the historical from January to March with only 5–20 days of rain. The rain- record where the highest amount is in October. The average fall stations at Kodiang and Kedah Peak had the least wet annual rainfall during this year is predicted to achieve days in a month while the rainfall stations at Ibu Bekalan 2,565 mm/year, 8% more than the historical record. Tupah, Keretapi Tokai, Kuala Nerang, and Pendang had Meanwhile, the rainfall projection results at station 66 the most rainy days. shows a different pattern from that of station 61. In the Exposure to calamitous events increases with prolonged future, the rainfall depth will increase in February, April, wet length, gauged in terms of rainfall volume per hour. The May, and November while decrease slightly in other Meteorological Department in Malaysia categorizes rainfall months. The highest rainfall depth will be in November depth into four groups, namely light (1–10 mm), moderate and not October as indicated in the historical data. The (11–30 mm), heavy (31–60 mm), and very heavy (more future rainfall trend will affect the expected increment in than 60 mm). Flash flood is expected to occur if the convec- contaminant into the earth systems due to development. tive rainfall is more than 60 mm in 2–4 hours. Cheang et al. However, January will still receive the least rainfall depth; () stated that most rainfall in Malaysia occurs in short the estimated average annual rainfall is 2,382 mm/year, spells. However, other supported indices are required in 5% from previous record (1997–2008). measuring the potential of flood or drought at the study area, such as stream flow, groundwater indices, and at-site Wet and dry length indices (Henny et al. ). Figure 9 shows the average maximum dry spell length A wet day is defined as a day with at least 1 mm of rainfall predicted at the 20 locations from year 2040 to 2069. Gener- depth meanwhile a dry day receives less than 0.1 mm of ally, the dry spell length is estimated to reduce annually at rainfall depth. The depth and pattern of rainfall in Malaysia most locations. This pattern is similar to the historical are influenced by the south-west monsoon (May–Septem- record, which showed that December to March had longer ber), the north-east monsoon (November–March), and two dry length than other months. The longest dry length is inter-monsoon seasons. December–February and June–July expected at station Kodiang with 116 days recorded in a are the warmer seasons while April–May and September– year and February being the hottest month in years 2057, November are the humid seasons. 2060, 2062, 2065, and 2066. A similar trend is also expected

Figure 7 | Annual rainfall trend at (a) station 61 (Pedu reservoir), and (b) station 66 (Muda reservoir).

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 658 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Figure 8 | Wet length distribution at Kedah for years 2040–2069.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 659 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

as estimated by Tukimat et al. (). Meanwhile, increment in rainfall depth is expected to decrease water demand. Uncertainties in climate change due to global warming are expected to impact upcoming atmospheric transition. With this, the management of water resource systems will be more important than ever to meet all demands as well as brace for calamitous events. The Muda Irrigation Scheme is the largest paddy cultivation zone and the main suppliers of rice, the staple food of Malaysians. The cultivation area is fully dependent on the Pedu-Muda reservoir for irrigation purposes during the two cultivation seasons in a year. Therefore, the current rule-curve operation used to predict rainfall has to be amended to cater for climate change issues. A better projection of future climatic variability Figure 9 | Max monthly dry length at 20 locations during years 2040–2069. shall enrich the updated hydrological information, which will become useful to operators in preparing or reinventing at station Jitra, where more than 90 days of dry spell is their strategies. expected in a year. Station LTP is expected to only have 10 days of dry spell a year. Based on the average simulated result, no dry day is expected from May to November at this ACKNOWLEDGEMENTS location. This research is supported by the Ministry of Higher Education (MOHE), Jabatan Pengairan dan Saliran (JPS), Universiti CONCLUSION Malaysia Pahang (UMP), and Universiti Teknologi Malaysia (UTM). This study had used the SDSM model to generate climate patterns in terms of temperature, rainfall, and wet and dry length for the next 30 years (2040–2069). Results showed REFERENCES that future climate pattern will still be interrelated to the historical record, but with greater magnitudes. The average Bates, B., Kundzewicz, Z. W., Wu, S. & Palutikof, J.  Climate Change and Water. IPCC Technical Paper VI, IPCC, Geneva, future annual temperature and rainfall are expected to

W Switzerland. increase by 0.2 C per decade and 0.9% per year, respect- Cheang, B. K., Tan, H. V. & Yong, P. W.  Some Aspects of ively. The months of January to March will be the hottest Wet and Dry Spells in Malaysia, 1951-1983. Part II: Synoptic and August to November will be the wettest months condition responsible for the wet and dry spell. Malaysian Meteorological Department, Research Publication 15,1–10. throughout a year with an average of 11 days and 30 days, Crane, R. G. & Hewitson, B. C.  Doubled CO2 precipitation respectively. The month of November will receive the high- changes for the susquehanna basin: downscaling from the est rainfall depth; this prediction differs slightly from GENESIS general circulation model. Int. J. Climatol. 18,66–67.  historical pattern due to expected changes in the north- FAO Rainfall-runoff analysis. (W. Critchley & K. Siegert, eds) A Manual for the Design and Construction of Water east monsoon trend. Harvesting Schemes for Plant Production, Rome, pp. 1–50. Increment in water depth at the Muda Irrigation Goodess, C. M. & Palutikof, J. P.  Development of daily rainfall Scheme area is expected to change the historical reservoir scenarios for southeast Spain using a circulation type approach to downscaling. Int. J. Climatol. 18, 1051–1081. management policies in quantifying the water supply for cul- Goosse, H., Barriat, P. Y., Lefebvre, W., Loutre, M. F. & Zunz, V. tivation purposes. A rise in temperature in the crop land is  Introduction to Climate Dynamics and Climate expected to escalate evapotranspiration by 0.04% annually, Modeling. Online textbook http://www.climate.be/textbook.

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020 660 N. N. A. Tukimat & S. Harun | Climate projection using multicorrelation matrix and downscaling method Journal of Water and Climate Change | 06.3 | 2015

Gutmann, E. D., Rasmussen, R. M., Liu, C., Ikeda, K., Gochis, Samadi, S. Z., Gummeneny, S. & Tajiki, M.  Comparison of D. J., Clark, M. P., Dudhia, J. & Thompson, G.  general circulation models: methodology for selecting the Comparison of statistical and dynamical downscaling of best GCM in Kermanshah Synoptic Station, Iran. Int. J. Glob. winter precipitation over complex terrain. J. Clim. 25, Warming 4, 347–365. 262–281. Sharma, M., Coulibaly, P. & Dibike, Y.  Assessing the need for Henny, A. J., Zbigniew, W., Kundzewicz, L. M., Tallaksen, H. H., downscaling RCM data for hydrologic impact study. J. Fendekova, M. & Christel, P.  Indices for Different Types Hydrol. Eng. 16 (6), 534–539. of Droughts and Floods at Different Scales. WATCH Water Spak, S., Holloway, T., Lynn, B. & Goldberg, R.  A comparison and Global Change. Technical Report No. 11, European of statistical and dynamical downscaling for surface Commission. temperature in North America. J. Geophys. Res. 112, D08101. IPCC.  Climate change 2007. In: Contribution of Working Tukimat, N. N. A., Harun, S. & Shahid, S.  Comparison of Group II to the Fourth Assessment Report of the different methods in estimating potential evapotranspiration Intergovernmental Panel on Climate Change, (M. L. Ottmar, at Muda Irrigation Scheme of Malaysia. JARTS 113,77–85. E. Ramon, P.-M. Youba, S. Kristin, M. Patrick, K. Susanne, Z. Wilby, R. L. & Dawson, C. W.  SDSM 4.2 – A Decision Timm, E. Patrick, H. Gerrit, S. Steffen & v.S. Christoph, eds). Support Tool for the Assessment of Regional Climate Change Cambridge University Press, Cambridge, UK, pp. 81–131. Impacts. Lancaster University, Science Department, Izuka, S. K., Oki, D. S. & Chen, C. H.  Effects of Irrigation and Environment Agency of England and Wales Department of Rainfall Reduction on Ground-Water Recharge in the Lihue Computer Science, Loughborough University, UK. Basin, Kauai, Hawaii. Scientiﬁc Investigations Report 2005– Wilby, R. L. & Wigley, T. M. L.  Precipitation predictors for 5146, Country of Kauai Department of Water, Hawaii, USA. downscaling: observed and general circulation model Jinqiang, C. & Simona, B.  Orographic Effects of the Tibetan relationships. Int. J. Climatol. 20, 641–661. Plateau on the East Asian Summer Monsoon, California Willems, P. & Vrac, M.  Statistical precipitation downscaling Institute of Technology, Pasadena, California. for small-scale hydrological impact investigation of climate Khan, M. S., Coulibaly, P. & Dibike, Y.  Uncertainty change. J. Hydrol. 402, 193–205. analysis of statistical downscaling methods. J. Hydrol. 319, Willmott, C. J. & Matsuura, K.  Advantages of the mean 357–382. absolute error (MAE) over the root mean square error Lopes, P.  Assessment of statistical downscaling methods – (RMSE) in assessing average model performance. Clim. Res. application and comparison of two statistical methods to a 30,79–82. single site in Lisbon. Earth Environ. Sci. 6, 022015. Yano, T., Aydin, M. & Haraguchi, T.  Impact of climate NARCCAP.  The A2 Emissions Scenario. North American change on irrigation demand and crop growth in a Regional Climate Change Assessment Program (NARCCAP), Mediterranean environment of Turkey. Sensors 7, http://www.narccap.ucar.edu/about/emissions.html. 2297–2315.

First received 5 February 2014; accepted in revised form 7 February 2015. Available online 25 March 2015

Downloaded from https://iwaponline.com/jwcc/article-pdf/6/3/647/374571/jwc0060647.pdf by guest on 28 May 2020