Streamflow prediction using multi-site rainfall obtained from hydroclimatic teleconnection
Introduction
Basin-scale streamflow prediction is an important step in water resources management for sustainable development. The variation of basin-scale streamflow is influenced by rainfall depth, its distribution pattern, catchment characteristics and the ground water contribution to the streamflow. The rainfall distribution, over the catchment, depends on local meteorology, large scale atmospheric circulation patterns and the geography of the catchment. It may be difficult to predict weekly streamflow accurately, by just considering the rainfall of few previous weeks, because the rainfall in current week also has substantial contribution to streamflow. This is especially true for monsoon season, when the soil is saturated, leading to insignificant infiltration. In the present study, a two-step approach is proposed for weekly streamflow prediction. In first step, current (weekly) multi-gridded rainfall is predicted with Genetic Programming (GP), based on large scale atmospheric circulation patterns, OLR and lag rainfall at grid points. Current step weekly streamflow is then predicted, by using observed gridded rainfall, up to previous weekly time step and GP predicted multi-gridded rainfall, at current weekly time step.
A single step model, for streamflow forecasting, is also developed, by using same inputs and the results are compared with the performance of two step model.
The objectives of this study are summarized as following:
- (1)
To develop GP-based models, for weekly multi-site (multi-gridded) rainfall prediction, based on large scale atmospheric circulation patterns with hydroclimatic teleconnection, lag multi-gridded rainfall and OLR.
- (2)
To develop GP-based weekly basin-scale streamflow prediction model, based on observed gridded rainfall at few previous weekly time steps, GP predicted gridded rainfall at current time step and lagged streamflow at immediate previous weekly time step.
- (3)
To assess the improvement in streamflow predictions due to inclusion of current step (week), GP predicted rainfall in the input set, in addition to the observed gridded rainfall up to the lag-1 weekly time step.
- (4)
To compare the results of the aforesaid two-step model with a single-step model, that uses lag streamflow, ENSO indices, EQUINOO indices, OLR anomaly, historical avg. rainfall and rainfall at the all grid points over last six weeks.
The methodology adopted for streamflow prediction in the two-step model can be visualized in flowchart (Fig. 1a). The first step deals with the multi-gridded rainfall prediction, and the second step deals with the basin-scale streamflow prediction.
Similarly the methodology of a single-step model can be seen in flowchart (Fig. 1b).
The rainfall as well as streamflow-prediction models are developed by using Genetic Programming tool. The methodology is different from the methodologies mentioned in the literature (Eltahir, 1996, Piechota et al., 1997, Chiew et al., 1998, Chandimala and Zubair, 2007).
The novelty of this method lies in the inclusion of predicted current week rainfall in streamflow-prediction model that contributes to current-week streamflow, especially in monsoon season.
Section snippets
Influence of large scale atmospheric circulation patterns over spatio-temporal rainfall distribution
Simultaneous variations in weather and climate over widely separated regions on earth have long been noted in the meteorological literature. Such recurrent patterns are commonly referred as “teleconnections”. Rainfall distribution patterns over the continents are significantly linked with the atmospheric circulation through hydroclimatic teleconnection.
It is also established that the natural variation of rainfall is linked with these large scale atmospheric circulation patterns, through
Data and case study
Daily gridded rainfall data at a spatial resolution of 1° latitude by 1° longitude for the period 1951–2003 is obtained from India Meteorological Department (IMD). Weekly rainfall values are calculated from these daily rainfall values for the grid points, encapsulating the upper Mahanadi River basin upstream of ‘Basantpur’ stream gauging station (Fig. 2). Historical average of weekly rainfall at each grid point is then computed as climatological mean rainfall at a particular grid point for a
Multi-site weekly rainfall prediction methodology
It is mathematically difficult to use climate signals for the prediction of basin-scale hydrologic variables due to the inherent complexity of the climate systems. But such complex systems can be modeled by using the modern Artificial Intelligence (AI) tools, like Artificial Neural Networks (ANN), Genetic Algorithm (GA)-based evolutionary optimizer and Genetic Programming (GP). Applications of ANN may be found in wide range of hydrologic studies, viz., rainfall–runoff modeling (Hsu et al., 1995
Streamflow prediction by rainfall–runoff modeling approach
Genetic Programming is used to translate the gridded rainfall information into streamflow. Auto-correlations are normally found to be significant in streamflow studies. Hence, streamflow in immediate previous time step is also considered as input in the present study. The basin-scale streamflow prediction is performed using data from 1990 to 2003, among which, monsoon rainfall data of years 1990–1998 are used for training purpose. The data from 1999 to 2003 are used for testing purpose.
Three
Streamflow prediction by single-step method
The problem of weekly streamflow forecasting is solved with single-step method, for comparison, by directly relating large-scale circulation and local meteorological information with streamflow.
The revised model uses
- (i)
Lagged streamflow in immediate previous week
- (ii)
ENSO indices at weekly time steps: (t-10)–(t-1), i.e., 10 values
- (iii)
EQUINOO indices at weekly time steps: (t-7)–(t-1), i.e., 7 values
- (iv)
OLR anomaly over river basin during time step (t-1)
- (v)
Weekly historical avg. rainfall during particular week
Results and Discussion
First, the performance of rainfall–runoff models, developed by using weekly observed rainfall during weeks, (t-6)–(t-1) are evaluated in terms of correlation coefficient. The correlation coefficients between observed streamflow and GP-predicted streamflow are calculated. They are 0.715 during training and 0.741 during testing. (r2 = 0.514 during training and r2 = 0.555 during testing). The plots of Observed and GP-Predicted Streamflow, using this approach, during training and testing are presented
Concluding remarks
The information of large scale atmospheric circulation patterns viz. El Niño Southern Oscillation (ENSO) and Equatorial Indian Ocean Oscillation (EQUINOO), with support of lagged rainfall information at every individual grid point and basin-scale OLR anomaly are successfully used for prediction of weekly gridded monsoon rainfall, in Mahanadi catchment with reasonable accuracy.
Results of this study show that the inclusion of GP-predicted rainfall, at current time step (t), as input, for weekly
References (56)
- et al.
Predictability of stream flow and rainfall based on ENSO for water resources management in Sri Lanka
Journal of Hydrology
(2007) - et al.
Combining a fuzzy optimal model with a genetic algorithm to solve multi-objective rainfall–runoff model calibration
Journal of Hydrology
(2002) - et al.
El Niño/Southern Oscillation and Australian rainfall, streamflow and drought: links and potential for forecasting
Journal of Hydrology
(1998) - et al.
Evolutionary self-organising modelling of a municipal wastewater treatment plant
Water Research
(2003) - et al.
Short-term streamflow forecasting with global climate change implications – a comparative study between genetic programming and neural network models
Journal of Hydrology
(2008) - et al.
Fuzzy conceptual rainfall–runoff models
Journal of Hydrology
(2001) - et al.
Western US streamflow and atmospheric circulation patterns during El Niño-Southern Oscillation
Journal of Hydrology
(1997) - et al.
Comparison of short term rainfall prediction models for real time flood forecasting
Journal of Hydrology
(2000) - et al.
Modeling rainfall–runoff using genetic programming
Mathematical and Computer Modeling
(2001) - et al.
A non-linear combination of the forecast of rainfall-runoff models by first order Takagi-Sugeno fuzzy system
Journal of Hydrology
(2001)
Impact of Indian Ocean dipole on the relationship between the Indian monsoon rainfall and ENSO
Geophysics Research Letters
Individual and combined effect of ENSO and Indian Ocean dipole on the Indian summer monsoon
Journal of Climate
Effects of El Niño Southern Oscillation and Pacific Interdecadal Oscillation on water supply in the Columbia River basin
Journal of Water Resources Planning and Management
Tropical Pacific ENSO Models: ENSO as a mode of couples system
Prediction of PCDDs/PCDFs emissions from municipal incinerators by genetic programming and neural network modeling
Waste Management & Research
Calibration of flow and water quality modeling using genetic algorithms
Lecture Notes in Artificial Intelligence
Hydro-metrological variability in the greater Ganges–Brahmaputra–Meghna basins
International Journal of Climatology
A recurrent neural networks approach using indices of low-frequency climatic variability to forecast regional annual runoff
Hydrological Processes
An artificial neural network approach to rainfall–runoff modeling
Hydrological Sciences Journal
Prediction and modelling of the rainfall–runoff transformation of a typical urban basin using ANN and GP
Applied Artificial Intelligence
The El Niño Southern Oscillation and long-range forecasting of flows in Ganges
International Journal of Climatology
The relat0069onship between US streamflow and La Niña events
Water Resources Research
El Niño and the natural variability in the flow of the Nile River
Water Resources Research
Discipulus Owner’s Manual
Droughts of the Indian summer monsoon: role of clouds over the Indian Ocean
Current Science
Extremes of the Indian summer monsoon rainfall. ENSO and equatorial Indian Ocean Oscillation
Geographical Research Letter
Cited by (41)
Streamflow prediction with large climate indices using several hybrid multilayer perceptrons and copula Bayesian model averaging
2021, Ecological IndicatorsCitation Excerpt :The profound relationship between large climate signals and streamflow is a complex issue. The atmospheric data and oceanic oscillatory phases affect the streamflow patterns in the different regions of the world (Kalra et al., 2013a; Kashid et al., 2010; Wang et al., 2020). The diverse atmospheric circulation and sea surface temperature indices change the streamflow pattern with different lag times (Wang et al., 2020).
A synthesis of drought prediction research over India
2021, Water SecuritySuspended sediment yield modeling in Mahanadi River, India by multi-objective optimization hybridizing artificial intelligence algorithms
2021, International Journal of Sediment ResearchCitation Excerpt :Numerous ANN and mathematical approach have been successfully applied to estimate and forecasting of streamflow, rainfall and runoff and flood in Mahanadi river basin, India (Kar et al., 2010; Meher & Jha, 2013; Pramanik & Panda, 2009). Kashid et al. (2010) applied GA to estimating the streamflow in Mahanadi river, basin, India. Kant et al. (2013) applied a genetic algorithm-based multi-objective optimization of neural network evolutionary algorithm for flood forecasting at Naraj gauge station in Mahanadi river.
An enhanced extreme learning machine model for river flow forecasting: State-of-the-art, practical applications in water resource engineering area and future research direction
2019, Journal of HydrologyCitation Excerpt :Consequently, AI have recently been given the credit for being a reasonably accurate and non-linear tool within the scope of hydrological science (Nourani et al., 2014; Yaseen et al., 2015). Some popular examples of AI models include artificial neural network (Jain and Kumar, 2007; Kagoda et al., 2010; Yonaba et al., 2010), support vector machine (Behzad et al., 2009; He et al., 2014; Noori et al., 2011), adaptive neuro-fuzzy system (Greco, 2012; Pramanik and Panda, 2009; Sharma et al., 2015), evolutionary computing (Kashid et al., 2010; Ni et al., 2010), and the hybrid wavelet-AI model (Danandeh Mehr et al., 2013; Makwana and Tiwari, 2014; Prasad et al., 2017; Sahay and Srivastava, 2014; (Deo et al., 2017; Prasad et al., 2017, 2018b)). However, all of these applied models still have a notable degree of shortcoming about their generalization and implementation as an expert system, including its time-consuming characteristic and human controlled (i.e., non-automated) modeling process.
Genetic programming in water resources engineering: A state-of-the-art review
2018, Journal of HydrologyCitation Excerpt :In the last decade, there was an attempt that GP was combined with the importance analysis method to provide an accurate intelligent methodology identifying the global climate inputs and local meteorological variables required for modeling seasonal runoff time series. Kashid et al. (2010) and Maity and Kashid (2011) demonstrated such applications in modelling weekly runoff time series for Mahanadi River in India using the following data: (i) the large-scale El Niño-Southern Oscillation (ENSO) and equatorial Indian Ocean Oscillation (EQUINOO) indices, (ii) the local meteorological variables including outgoing longwave radiation, total perceptible water, temperature anomaly, and pressure anomaly; and (iii) streamflow from previous time steps. The rainfall information was not used in the study in order to check the possibility of the proposed model to apply on the catchments with no rainfall measurements.
A binary genetic programing model for teleconnection identification between global sea surface temperature and local maximum monthly rainfall events
2017, Journal of HydrologyCitation Excerpt :To cope with the problem of nonlinearity, data-driven techniques such as artificial neural networks (ANN), extreme learning machine (ELM), association rules (AR), and genetic programing (GP) are suggested in recent studies (e.g., Mwale and Gan 2005; Mwale and Gan 2010; Kashid and Maity 2012; Danandeh Mehr et al., 2014; Chang et al., 2017b). For example, linear GP was used by Kashid et al. (2010) to predict gridded multi-site weekly rainfall from ENSO indices, Equatorial Indian Ocean Oscillation (EQUINOO) indices, Outgoing Longwave Radiation (OLR), and lag rainfall at grid points, over the Mahanadi catchment, India. The authors demonstrated that the model outputs can be used satisfactorily to predict streamflow in the catchment.