Application of artificial intelligence models for the prediction of standardized precipitation evapotranspiration index (SPEI) at Langat River Basin, Malaysia

doi:10.1016/j.compag.2017.12.002

Computers and Electronics in Agriculture

Volume 144, January 2018, Pages 164-173

https://doi.org/10.1016/j.compag.2017.12.002 Get rights and content

Highlights

•
WAANN outperformed WANFIS in predicting short-term and mid-term tropical SPEIs.
•
Performance measures showed that WAANN predicted SPEI-3 and -6 accurately.
•
Drought predictions can be used for agricultural and irrigation planning and design.

Abstract

Drought forecasting is a vital for mitigating the impact of drought events on the economy, tourism, agriculture and water resource systems. This paper adopts the proposed Wavelet-ARIMA-ANN (WAANN) model and the latest Wavelet-Adaptive Neuro-Fuzzy Inference System (WANFIS) model to predict the Standardized Precipitation Evapotranspiration Index (SPEI) at the Langat River Basin for different time scales (1-month, 3-months and 6-months). Model input data pre-processing with wavelet decomposition for improving the performance of the models was carried out apriori. The historical SPEI from 1976 to 2007 were used in the WAANN and WANFIS models for predicting the SPEI for the test period from 2008 to 2015. The Adjusted Coefficient of Determination (R²_adj), Root-Mean-Square-Error (RMSE), Mean Absolute Error (MAE), Willmott's Index of Agreement (d) and the Nash-Sutcliffe Coefficient of Efficiency (E) were used to assess the models. It was found that the prediction accuracy of the two models improved with time scale length. For the prediction of SPEI-1 (1-month), the errors associated with both models were considered relatively high. Based on the performance measures and graphical plots, the WAANN model is better for the prediction of SPEI-3 and SPEI-6. The WANFIS model had satisfactory prediction of the mid-term drought forecasting for all stations. The WAANN model developed in this study however, gives better accuracy for both, the short-term and mid-term drought forecasting.

Introduction

On the basis of the long-run average precipitation (normal precipitation) for a particular basin, the declining trend of precipitation indicates the initiation of droughts (Jalalkamali et al., 2015). Low relative humidity, temperatures, high wind velocity, rainfall characteristics including intensity, duration of precipitation and the distribution of rainfall during the crop growing seasons are important features of the droughts (Mishra and Singh, 2010). It has been reported in researches that the drought-induced conditions developed seasonally in the event of El Nino (also known as the warm phase of El Nino Southern Oscillation, ENSO). El Nino is induced by the reduction of trade winds and, in tandem with the increase of earth surface temperatures, appears to occur on the average, every 3–4 years (Paz et al., 2007).

Even though Malaysia receives an average of 2800 mm of precipitation annually, the country however is still subjected to prolonged dry spells; especially at the Langat River Basin, where the rapid urbanisation of the Kuala Lumpur City area has resulted in an increased demand for the freshwater supply (Pour et al., 2014). The ENSO profoundly affects the condition of climate in Malaysia and in South-east Asia. Yusof et al. (2012) applied the Kriging method to analyse the upward and downward trends during the occurrence of droughts in Peninsular Malaysia. Their results showed that the major regions of West Malaysia are subjected to an upward trend throughout the dry season, particularly in the eastern and western regions. It is extremely imperative for the water resources department to predict the drought intensity, severity and duration. With prior awareness of the onset of droughts, appropriate actions to mitigate the consequential damages can be considered. Precise and representative drought index series about the onset, extent and the end of the drought event allows the proper drought contingency plans to be established (Subash et al., 2011).

Drought monitoring and early warning are important phases to manage droughts (Bachmir et al., 2016). Among the approaches for drought forecasting, the use of artificial intelligence (AI) shows outstanding performance and accuracy (Masinde, 2014, Özger et al., 2011, Belayneh et al., 2014). The flexibility and adaptability of AI is useful in predicting the occurrence of droughts that poses varying durations, frequencies and intensities; characteristics that are not easily determined effectively using empirical relations. Artificial Neural Network (ANN), Fuzzy Logic (FL) and Support Vector Regression (SVR) are the examples of AI models that can be used to forecast the time series modelling based on historical data (Djerbouai and Souag-Gamane, 2016). Another popular type of data-driven model is the stochastic method. The older Auto Regressive Integrated Moving Average Model (ARIMA) and Seasonal Auto Regressive Moving Integrated Average Model (SARIMA) are the most widely adopted stochastic approaches, as they are relatively simple yet providing excellent results (Mossad and Alazba, 2015, Bazrafshan et al., 2015).

In order to improve the prediction accuracy in the time series forecasting, different types of hybrid models were developed to minimize the error between the actual and predicted values. For example, both Choubin et al., 2014, Choubin et al., 2016 explored the effectiveness of ANFIS model in forecasting the Standardized Precipitation Index (SPI) for different regions. Apart from that, ARIMA-ANN, Wavelet-ANN (WANN) and WANFIS are also effective forecasting tools among the hybrid models (Belayneh et al., 2014, Shabri, 2014). The wavelet transform had been proven to decompose the incoming signal into different frequencies (Djerbouai and Souag-Gamane, 2016, Belayneh and Adamowski, 2012, Belayneh and Adamowski, 2013, Özger et al., 2011, Özger et al., 2012) and thus, providing a more accurate output for the data driven models. By combining the strength of different models, the hybrid models can give a higher degree of accuracy.

In 2013, the WANFIS model was first used in drought forecasting because the wavelet transform has the ability to improve the accuracy of the models. The performance of the WANFIS model was compared with the WANN, ANN and ANFIS. WANFIS model results were more precise than other models in the study and the wavelet transform was particularly good in improving meteorological drought forecasting. Shabri (2014) also proposed to use the WANFIS model to forecast the drought events in Malaysia. Subsequently, the WANFIS model was widely recognised as an effective tool to forecast the time series of drought index. However, the performance of the WANFIS highly relied on the modelling of the ANFIS model. Seo et al. (2016) had pointed the drawbacks of using the ANFIS modelling that would lead to difficulty in the identification of network parameters and the number and type of membership functions. Moreover, the development of the ANFIS model is computationally expensive and complex.

The aim of this paper is to improve the drought forecasting procedure at the Langat River Basin, using hybrid modelling. An alternative model, the WAANN was proposed in this study since the wavelet transform is able to reduce the complexity of the time series and the ARIMA-ANN model was proven as an appropriate tool in the time series forecasting (Babu and Reddy, 2014, Khashei and Bijari, 2011). The latest hybrid model WANFIS in drought forecasting is used for comparison. The expected results are the SPEI series predicted with both models and the accompanying performance evaluations.

Section snippets

Study area and data acquisition

The catchment area of the Langat River Basin in Selangor, Malaysia is about 2400 km², supplying approximately 65% of water usage in the state of Selangor. Two reservoirs in the Langat River Basin, namely the Semenyih dam (area of 56.6 km²) and the Hulu Langat dam (area of 41.0 km²) supply water to the state. The recorded daily precipitation and temperature for each station in the basin were retrieved from the Department of Irrigation and Drainage (DID) Malaysia and the Malaysian Meteorology

Homogeneity tests

The calibration of the homogeneity test was: firstly, the time series was divided into twelve months (from January until December) and then each month of the dataset was tested for homogeneity. The p-value for all the selected tests should be greater than the 5% significance value to indicate that the time series is homogeneous in a strict sense. The four tests were tested with the null hypothesis where the data were assumed homogeneous. The results were classified into three categories

Conclusions

Two hybrid models, namely the WAANN and WANFIS were selected to forecast the future SPEI at the Langat River Basin, Malaysia. The input data for development of both models was pre-processed with wavelet transform to enhance the performance of the models. Based on the key statistical parameters, the performance of both WAANN and WANFIS models improved due to the reduction of white noise by increasing the filter length, when the time scale increased. For the prediction of SPEI-1, the errors

Acknowledgments

The authors wish to express their sincere gratitude to Universiti Tunku Abdul Rahman (UTAR) for the financial support through the grant of UTARRF Project No. IPSR/RMC/UTARRF/2015-C1/H01.

References (35)

M.I. Ahmad et al.
Log-logistic flood frequency analysis
J. Hydrol.
(1988)
C.N. Babu et al.
A moving-average filter based hybrid ARIMA–ANN model for forecasting time series data
Appl. Soft Comput.
(2014)
A. Belayneh et al.
Long-term SPI drought forecasting in the Awash River Basin in Ethiopia using wavelet neural network and wavelet support vector regression models
J. Hydrol.
(2014)
B. Choubin et al.
Application of several data-driven techniques to predict a standardized precipitation index
Atmosfera
(2016)
M. Khashei et al.
A novel hybridization of artificial neural networks and ARIMA models for time series forecasting
Appl. Soft Comput.
(2011)
O. Kisi et al.
Long-term monthly evapotranspiration modeling by several data-driven methods without climatic data
Comput. Electron. Agric.
(2015)
O. Kisi et al.
Daily pan evaporation modeling using chi-squared automatic interaction detector, neural networks, classification and regression tree
Comput. Electron. Agric.
(2016)
A.K. Mishra et al.
A review of drought concepts
J. Hydrol.
(2010)
J.O. Paz et al.
Development of an ENSO-based irrigation decision support tool for peanut production in the southeastern US
Comput. Electron. Agric.
(2007)
N. Subash et al.
Comparing water-vegetative indices for rice (Oryza sativa L.)–wheat (Triticum aestivum L.) drought assessment
Comput. Electron. Agric.
(2011)

Q.F. Wang et al.

Temporal-spatial characteristics of severe drought events and their impact on agriculture on a global scale

Quatern. Int.

(2014)

S. Bachmir et al.

A quantitative analysis to objectively appraise drought indicators and model drought impacts

Hydrol. Earth Syst. Sci.

(2016)

O. Bazrafshan et al.

Hydrological drought forecasting using ARIMA models (Case Study : Karkheh Basin)

Ecopersia

(2015)

S. Beguería et al.

Standardized precipitation evapotranspiration index (SPEI) revisited: parameter fitting, evapotranspiration models, tools, datasets and drought monitoring

Int. J. Climatol.

(2014)

A. Belayneh et al.

Standard precipitation index drought forecasting using neural networks, wavelet neural networks, and support vector regression

Appl. Comput. Intell. Soft Comput.

(2012)

A. Belayneh et al.

Drought forecasting using new machine learning methods

J. Water Land Dev.

(2013)

B. Choubin et al.

Drought forecasting in a semi-arid watershed using climate signals: a Neuro-fuzzy modeling approach

J. Mount. Sci.

(2014)

Cited by (92)

Prediction of agricultural drought index in a hot and dry climate using advanced hybrid machine learning
2024, Ain Shams Engineering Journal
Drought monitoring and forecasting are essential for efficient water resources management. The present research aims to provide a reliable prediction of the effective Reconnaissance Drought Index (eRDI) based on seven evaporation stations in the southern Baluchestan sub-basin of Iran. To achieve this purpose, artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and support vector regression (SVR) machine learning methods are used and combined with the marine predator optimization algorithm (MPA) to enhance efficiency. Drought monitoring and forecasting have been performed on time scales of 1-, 3-, and 6-months intervals. The results demonstrated the superiority of the ANFIS-MPA algorithm over the SVR-MPA and ANN-MPA approaches. In addition, as the time scale increased, the accuracy of all models improved. The best results were for the eRDI 6-month at Kajdar Sarbaz station by ANFIS-MPA (MAE = 0.33, NSE = 0.83, R² = 0.99), SVR-MPA (MAE = 0.36, NSE = 0.78, R² = 0.85) and ANN-MPA (MAE = 0.37, NSE = 0.72, R² = 0.83).
Deep learning versus hybrid regularized extreme learning machine for multi-month drought forecasting: A comparative study and trend analysis in tropical region
2024, Heliyon
Drought is a hazardous natural disaster that can negatively affect the environment, water resources, agriculture, and the economy. Precise drought forecasting and trend assessment are essential for water management to reduce the detrimental effects of drought. However, some existing drought modeling techniques have limitations that hinder precise forecasting, necessitating the exploration of suitable approaches. This study examines two forecasting models, Long Short-Term Memory (LSTM) and a hybrid model integrating regularized extreme learning machine and Snake algorithm, to forecast hydrological droughts for one to six months in advance. Using the Multivariate Standardized Streamflow Index (MSSI) computed from 58 years of streamflow data for two drier Malaysian stations, the models forecast droughts and were compared to classical models such as gradient boosting regression and K-nearest model for validation purposes. The RELM-SO model outperformed other models for forecasting one month ahead at station S1, with lower root mean square error (RMSE = 0.1453), mean absolute error (MAE = 0.1164), and a higher Nash-Sutcliffe efficiency index (NSE = 0.9012) and Willmott index (WI = 0.9966). Similarly, at station S2, the hybrid model had lower (RMSE = 0.1211 and MAE = 0.0909), and higher (NSE = 0.8941 and WI = 0.9960), indicating improved accuracy compared to comparable models. Due to significant autocorrelation in the drought data, traditional statistical metrics may be inadequate for selecting the optimal model. Therefore, this study introduced a novel parameter to evaluate the model's effectiveness in accurately capturing the turning points in the data. Accordingly, the hybrid model significantly improved forecast accuracy from 19.32 % to 21.52 % when compared with LSTM. Besides, the reliability analysis showed that the hybrid model was the most accurate for providing long-term forecasts. Additionally, innovative trend analysis, an effective method, was used to analyze hydrological drought trends. The study revealed that October, November, and December experienced higher occurrences of drought than other months. This research advances accurate drought forecasting and trend assessment, providing valuable insights for water management and decision-making in drought-prone regions.
Standardized precipitation evapotranspiration index (SPEI) estimated using variant long short-term memory network at four climatic zones of China
2023, Computers and Electronics in Agriculture
Although the accurate prediction of the Standardized Precipitation Evapotranspiration Index (SPEI) is considered meaningful in reducing drought losses, its wide applications are limited to substantial meteorological data requirements. Considering Long Short-Term Memory network (LSTM) has proved its potential in estimating drought index, the concern is justified regarding the performance of its variants for estimating SPEI using limited meteorological input at the national level. Therefore, this study established the SPEI models using empirical methods, SVM, RNN, LSTM, BiLSTM, and CNN-LSTM, respectively, to cope with different data-missing scenarios. Based on a comprehensive comparison among different methods for multiscale SPEI estimation at four climatic zones of China, the results showed that BiLSTM was the most recommended model for estimating SPEI at 3-month timescale, with R², NSE, and RMSE ranging 0.916–0.997, 0.907–0.997, and 0.143–0.353, respectively. Whereas CNN-LSTM was more suitable for other timescales, with R², NSE, and RMSE being 0.904–0.999, 0.858–0.989, and 0.145–0.365 for estimating SPEI at 6-month timescale, respectively, and 0.858–0.998, 0.795–0.991, and 0.081–0.568 for estimating SPEI at 12-month timescale. Generally, the accuracy performance of SPEI methods can be ranked from best to worst as LSTM-type models, SVM, RNN, and empirical methods. The exception was that H-S exceeded RNN3 in TCZ and MPZ by 4.1–30.0 % for R², 4.1–65.0 % for NSE, and 2.3–19.6 % for RMSE, respectively. Moreover, this study found that the accuracy performance of machine learning models for SPEI estimation got worse with the number of independent variables decreased. Overall, the variants of LSTM exerted excellent performance for multiscale SPEI estimation, which provided the most accurate prediction of meteorological, agroecological, and hydrological droughts throughout China.
A new deep learning method for meteorological drought estimation based-on standard precipitation evapotranspiration index
2023, Engineering Applications of Artificial Intelligence
In this study, we studied drought parameter estimation based on the standard precipitation evapotranspiration index (SPEI), for four stations located at different places in Türkiye. We proposed a new hybrid deep structure of convolutional neural network (CNN) and long short-term memory (LSTM) methods and compared it with several other existing methods. Our extensive experimental results obtained in different cases of the station, time scale, and estimator show that better estimation results are often obtained with 12-month time scale for all methods. For instance using SPEI-9 values for Şanlıurfa station, the CNN method achieves 0.57 RMSE, 0.24 MADE, 0.11 MAE, and 99.38% R², whereas the proposed adaptive CNN-LSTM obtains values of 0.42, 0.18, 0.04, and 99.69% for the same metrics, respectively. Also, using SPEI-6 values for Mardin station, the bidirectional-LSTM method achieves 0.72 RMSE, 0.35 MADE, 0.22 MAE, and 98.94% R², while the proposed adaptive CNN-LSTM obtains values of 0.32, 0.12, 0.02, and 99.53%, respectively In this means, the proposed method is superior to existing methods in all cases in terms of almost all performance metrics. It is also robust to changes in given input time series data coming from different stations. We also investigated the percentage occurrence of drought categories at different time scales. The results of this analysis also show that the highest percentages are seen in the wet category for all stations. The percentages occurrence of drought wet category are 48.3, 50.07, 50.14, and 50.79 at SPEI-3, SPEI-6, SPEI-9, and SPEI-12 time scales, respectively.
Development of wavelet-based Kalman Online Sequential Extreme Learning Machine optimized with Boruta-Random Forest for drought index forecasting
2023, Engineering Applications of Artificial Intelligence
Citation Excerpt :
A regional investigation over Iran was conducted and the reported predictive model confirmed its potential in predicting the drought index. SPEI over tropical region of Malaysia was forecasted using coupled wavelet transformation with ANN model (Soh et al., 2018). The data pre-processing of wavelet revealed a satisfactory pre-modeling step for drought index forecasting.
Drought is a stochastic and recurring hydrological natural hazard that occurs due to a shortage of precipitation over a period of time. Drought forecasting in water resources systems has an important role in reducing devastating ecological and social impacts. However, due to the fluctuating nature of drought indicator time series, their forecasting on a short time scale without advanced pre-processing is extremely challenging, and little research has been presented in this field. In this study, a new complementary machine learning (ML) approach, Kalman filter regression-based Online Sequential Extreme Learning Machine and (KOSELM) coupled with the Boundary Corrected Maximal Overlap Discrete Wavelet Transform (BC-MODWT-KOSELM), was implemented for forecasting one-month/three-month ahead Standardized Precipitation Evapotranspiration Index (i.e., SPEI 12 and SPEI 24). Here, the Kalman filter regression was employed to optimize the hyper-parameters of the Online Sequential Extreme Learning Machine (OSELM). Precipitation and potential evapotranspiration data from Bandar Abbas (warm semi-humid climate) and Rasmar (humid climate) synoptic stations, Iran were used in the analysis for a period of (1987–2019). In order to validate the BC-MODWT-KOSELM model, it was compared with two other well-known ML approaches, classical Extreme Learning Machine (ELM) and Generalized Regression Neural Network (GRNN) in both standalone and BC-MODWT-based complementary frameworks (i.e., BC-MODWT-ELM and BC-MODWT-GRNN). First, the original time series of the benchmark drought index was decomposed into the wavelet and scaling coefficients based on three mother wavelets and different decomposition levels. Afterward, the most significant lags were extracted using the Boruta-random forest (B-RF) feature selection. The BC-MODWT-KOSELM was identified as the superior forecasting model for SPEI 12 (t ＋ 1) with R $=$ 0.9511 and RMSE $=$ 0.3318; for SPEI 12 (t ＋ 3) with R $=$ 0.8171 and RMSE $=$ 0.6168; for SPEI 24 (t ＋ 1) with R $=$ 0.9710 and RMSE $=$ 0.2156; for SPEI 24 (t ＋ 3) with R $=$ 0.9014 and RMSE $=$ 0.3937 in the Bandar Abbas station. Besides, the BC-MODWT-KOSELM outperformed the comparative counterpart model for SPEI 12 (t ＋ 1 with R $=$ 0.9401 and RMSE $=$ 0.3371; SPEI 12 (t ＋ 3) with R $=$ 0.8266 and RMSE $=$ 0.5723; for SPEI 24 (t ＋ 1) with R $=$ 0.9640 and RMSE $=$ 0.2974; for SPEI 24 (t ＋ 3) with R $=$ 0.9063 and RMSE $=$ 0.4874 in the Ramsar station.
Application of WNN-PSO model in drought prediction at crop growth stages: A case study of spring maize in semi-arid regions of northern China
2022, Computers and Electronics in Agriculture
Citation Excerpt :
However, the current research on drought prediction of crop growth season is mostly on the scale of season and month, without taking into account crop growth characteristics in different stages. The results are not very meaningful for the guidance of spring maize production (Hao et al., 2018; Soh et al., 2018). It is necessary to use an index to delimit the severity of drought.
Drought prediction of regional crops during the growth stages can get drought information in advance and prepare for the response, so as to effectively guide the water-saving irrigation and lessen yield losses of spring maize. This study used the daily meteorological data in the research area during 1965–2019 to calculate the Crop Water Deficit Index (CWDI) and used the Pearson correlation coefficient (PCC) method to select the relevant factors for the CWDI. Then, CWDI was predicted using Support Vector Machine (SVM), Back Propagation Neural Network (BPNN), Wavelet Neural Network (WNN), Support Vector Machine-Particle Swarm Optimize (SVM-PSO), Back Propagation Neural Network-Particle Swarm Optimize (BPNN-PSO), and Back Wavelet Neural Network-Particle Swarm Optimize (WNN-PSO) models. By comparing d, R², Root Mean Square Error (RMSE), and Mean Relative Error (MRE), the best model was selected and used to predict drought in the next five years. The compared results showed that WNN-PSO models performed better than all the other models. When the input variables were CWDI and 3 main relevant meteorological factors relative humidity, maximum temperature and precipitation at sowing-seedling stage, the MRE (1.58%∼6.65%), the MAE (1.2922 ∼ 3.5866) as well as RMSE (0.0174 ∼ 0.0481) were the smallest and the d(0.8608 ∼ 0.915) and R² (0.8402 ∼ 0.9853) was the largest. The model R² increased by 10.6%, 32.8% and 125.9% compared with WNN, BPNN-PSO and SVM-PSO. It is proved that WNN-PSO is suitable for predicting CWDI of spring maize in drought-affected areas.

View all citing articles on Scopus

View full text

Original papersApplication of artificial intelligence models for the prediction of standardized precipitation evapotranspiration index (SPEI) at Langat River Basin, Malaysia

Highlights

Abstract

Introduction

Section snippets

Study area and data acquisition

Homogeneity tests

Conclusions

Acknowledgments

J. Hydrol.

Appl. Soft Comput.

J. Hydrol.

Atmosfera

Appl. Soft Comput.

Comput. Electron. Agric.

Comput. Electron. Agric.

J. Hydrol.

Comput. Electron. Agric.

Comput. Electron. Agric.

Quatern. Int.

A quantitative analysis to objectively appraise drought indicators and model drought impacts

Hydrol. Earth Syst. Sci.

Hydrological drought forecasting using ARIMA models (Case Study : Karkheh Basin)

Ecopersia

Standardized precipitation evapotranspiration index (SPEI) revisited: parameter fitting, evapotranspiration models, tools, datasets and drought monitoring

Int. J. Climatol.

Standard precipitation index drought forecasting using neural networks, wavelet neural networks, and support vector regression

Appl. Comput. Intell. Soft Comput.

Drought forecasting using new machine learning methods

J. Water Land Dev.

Drought forecasting in a semi-arid watershed using climate signals: a Neuro-fuzzy modeling approach

J. Mount. Sci.

Original papers
Application of artificial intelligence models for the prediction of standardized precipitation evapotranspiration index (SPEI) at Langat River Basin, Malaysia