International Journal of Computer Applications (0975 – 8887)

Volume 51– No.18, August 2012

Application of Artificial Neural Networks in Weather

Forecasting: A Comprehensive Literature Review
Gyanesh Shrivastava Sanjeev Karmakar Manoj Kumar Kowar
Dr. C.V. Raman University, Bhilai Institute of Technology (BIT) Bhilai Institute of Technology (BIT)
Bilaspur, Chhattisgarh, India Bhilai House, Durg, 491001, Bhilai House, Durg, 491001,
Chhattisgarh, India Chhattisgarh, India
Pulak Guhathakurta
India Meteorological Department (IMD)
Sivaji Nagar, Pune, India

ABSTRACT global predictors to explain the variability of the two

neighbouring regions having distinguished rainfall features.
To recognize application of Artificial Neural Networks (ANNs)
in weather forecasting, especially in rainfall forecasting a Karmakar et al., 2008,2009, has found that, the variability of
comprehensive literature review from 1923 to 2012 is done and monsoon rainfall over the Indian district level or very small scale
presented in this paper. And it is found that architectures of geographical region is due to external forcing i.e., global weather
ANN such as BPN, RBFN is best established to be forecast parameter and also the internal variability within the time series
chaotic behavior and have efficient enough to forecast monsoon itself. If external forcing is assumed to be same then also there
rainfall as well as other weather parameter prediction remains variability within the time series that can only be
phenomenon over the smaller geographical region. explained if we are able to predict the internal dynamical
behavior of the weather data time series. The pattern recognition
and prediction in a deterministic approach through ANN
Keywords technique based on back-propagation algorithm is found to be
the most efficient way by which internal dynamics of rainfall
ANN, BPN, Back-propagation, RBF, Rainfall, Forecasting. time series data has been successfully identified. Moreover,
pattern recognition and prediction have a longer lead-time as
1. INTRODUCTION they can be made a year in advance. No other model (except the
neural network model) so far has been able to forecast district
Weather forecasting (especially rainfall) is one of the most level long-range weather parameter so accurately.
important and challenging operational tasks carried out by Artificial Neural Network came into existence in 1986 which is
meteorological services all over the world. It is furthermore a able to get rid of the above two drawbacks and drawing
complicated procedure that includes multiple specialized fields considerable attention of research workers, as it can handle the
of expertise. Researchers in this field have separated weather complex non-linearity problems better than the conventional
forecasting methodologies into two main branches in terms of existing statistical techniques.
numerical modeling and scientific processing of meteorological
data. The most widespread techniques used for rainfall The objective of this study is to expand and evaluate the ANN.
forecasting are the numerical and statistical methods. Even This objective is considered via comprehensive review of
though researches in these fields are being conducted for a long literature (1923-2012). It is found that, detail of discussion
time, successes of these models are rarely visible. There is concerning the architecture of ANN for the same is rarely visible
limited success in forecasting the weather parameters using the in the literature; however various applications of ANN are
numerical model. The accuracy of the models is dependent upon available. The paper has been constructed with the sections.
the initial conditions that are inherently incomplete. These Section II discussed a comprehensive review of world-wide
systems are not able to produce satisfactory results in local and contribution from 1923 to 2012. Methodology to forecast those
short-term cases. The performances, however, are poor for long are unquestionably accepted and no scientific controversy to
range prediction of monsoon rainfall even for the larger spatial forecast rainfall has been discussed in section III. Result of the
scale and particularly, for the Indian region. As an alternative, review has been discussed in section IV and finally a conclusion
statistical methods in which rainfall time series are treated as is discussed in the section V.
stochastic are widely used for long-range predication of rainfall.
IMD has been using statistical models for predicting monsoon 2. LITERATURE REVIEW
rainfall. Statistical models were successful in those years of
normal monsoon rainfall and failed remarkably during the After initial work of Walker [1,2] several attempts by Gowariker
extreme monsoon years like 2002 and 2004. However, it is very et al., [3,4], Thapliyal et al., [5,6] have been made for
difficult to get the same or better skill in predicting district level developing better models for long-range forecasts of summer
monsoon rainfall as that of all-India level monsoon rainfall using monsoon rainfall in India. Performance of the Gowariker et al.,
these statistical models. Two main drawbacks of these statistical [4]; Rajeevan et al., [7,8], Thapliyal et al., [9] regression models
models are: based on different sets of predictors have found to be satisfactory
1. Statistical models are not useful to study the highly and reasonably accurate during last eleven years. These models
nonlinear relationships between rainfall and its predictors, are being extensively used by IMD for long-range forecasts of
even if one considers models like power regression. summer monsoon (June-September) rainfall over India as a
2. There is no ultimate end in finding the best predictors. It whole. Recently, IMD has been trying to forecast for Indian sub
will never be possible to get different sets of regional and regions and issuing long-range forecasts for three broad

1
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

homogeneous regions of India, viz., Northwest India, Northeast developed for all the major cities with different sets of related
India and the Peninsula through the updated three individual data but the network architecture will be different.
power regression models based on different sets of predictors
[10]. However, Guhathakurta [11] found these statistical models After comparative study of short term rainfall prediction models
to be successful in those years of normal monsoon rainfall and for real time flood forecasting, E. Toth et al., have found that the
failed remarkably during the extreme monsoon years like 2002 time series analysis technique based on ANN provides
and 2004. Also Rajeevan et al., Thapliyal et al., [5, 6, 8-10] significant improvement in the flood forecasting accuracy in
have found that the statistical models have many inherent comparison to the use of simple rainfall prediction approaches
limitations. Guhathakurta et al., [12] have observed that the [31]
correlations between monsoon rainfall and the predictors can
never be perfect and there is no ultimate end in finding the best In 2001 Luk et al., have developed and compared three types of
predictors. Parthasarathy et al., Hastenrath et al., [13,14] have ANNs suitable for rainfall prediction i.e. multilayer feed forward
found they may suffer epochal changes and there may be cross- neural network, Elman partial recurrent neural network and time
correlations between the parameters. Rajeevan et al., delay neural network [32]. Michaelides et al., have found that
Guhathakurta, Krishnamurthy et al., Sahai et al., [9,15-18] ANN is a suitable tool for the study of the medium and long term
have found that attempts to forecast monsoon rainfall as well as climatic variability. The ANN models trained were capable of
climate parameters through statistical technique over smaller detecting even minor characteristics and differentiating between
areas like a district, or monsoon periods such as a July, monsoon various classes [33]. After a study of RBFNN, Chang et al.,
(June-September), have become unsuccessful as correlations fall 2001, have found that RBFNN is a suitable technique for a
drastically. Guhathakurta [19,20] has observed that the weather rainfall runoff model for three hours ahead floods forecasting
prediction over high-resolution geographical regions is very [34].
complicated. However, since 1986, the ANN technique has been
drawing considerable attention of research workers, as it can Brath et al., 2002, have presented time series analysis technique
handle the complex non-linearity problems better than the for improving the real time flood forecast by a deterministic
conventional existing statistical techniques. lumped rainfall runoff model and they have concluded that apart
from ANNs with adaptive training , all the time series analysis
In the case study of The chaotic time series of Indian monsoon techniques considered allow significant improvements if flood
rainfall, Basu and Andharia,1992,have found that the resulting forecasting accuracy compared with the use of empirical rainfall
forecast formula uses only the rainfall of past seven years as predictors [35]. Using ANN for daily rainfall runoff modeling,
predictors, making a forecast eight months in advance[21]. Rajurkar et al., have found that coupling of ANN with a
multiple-input single-output model predicted the daily runoff
After development of recurrent Sigma- Pi neural network rainfall values with high accuracy both in the training and validation
forecasting system, Chow and Cho, 1997, have concluded that periods [36]. For a rainfall runoff relationship Harun and Irwan,
the neural network based now casting system is capable of 2002, have concluded that the performance of neural network
providing a reliable rainfall now casting in Hong Kong [22]. model is better than HEC-HMS and MLR models for modeling
the rainfall runoff relationship [37]. Iseri et al., 2002, have
Lee et al., 1998, have found that RBF networks produced good developed medium term forecasting of August rainfall in
prediction while the linear models poor prediction [23]. Hsieh, Fukuoka city. In order to identify the sufficient predictors, the
W.H. and B. Tang, 1998, applied various ANN models for partial mutual information was used for the candidate predictors,
prediction and analysis in meteorology data as well as which are Sea Surface Temperature anomalies (SSTa) in the
oceanography data and have found ANN technique is extremely Pacific Ocean and three climate indices. When data with lead
useful [24]. In another research Dawson and Wilby have found times between one and twelve months were used to forecast
that, rainfall runoff modeling, the ability of the ANN to cope August rainfall, it was found that a model with the North Pacific
with missing data and to “learn” from the event currently being index and selected SSTa as inputs performed reasonably well
forecast in real time makes it an appealing alternative to [38].
conventional lumped or semi distributed flood forecasting
models [25]. Guhathakurta et al., have found that performance Silva and Snell et al., 2003, have applied this approach both in
of the hybrid model (model III), has been the best among all terms of predictive accuracy and model encompassing. This
three models developed. RMSE of this hybrid model is 4.93%. technology is currently being widely applied to climate
As this hybrid model is showing good results it is now used by prediction because of its ability to explain the complex behavior
the IMD for experimental long range forecast of summer through time series as well as regression data analysis [39,40].
monsoon rainfall over India as a whole [26]. Richard, 2003, has completed simulation of European climate,
through this technique. In that study, Neural Network was used
Ricardo et al., 1999, have used this technology for simulation of for linear regression analysis [41].
daily temperature for climate change over Portugal [27]. In 2004 Maqsood et al., have found that HFM is relatively less
Wherein, performances of linear models and non-linear ANN are accurate and RBFN is relatively more reliable for the weather
compared using a set of rigorous validation techniques. Finally, forecasting problems and in comparison the ensembles of neural
the non-linear ANN model is initialized with general circulation networks produced the most accurate forecast [42]. After
model output to construct scenarios of daily temperature at the applying soft computing techniques Pasero and Moniaci, 2004,
present day (1970–79) and for a future decade (2090–99). have found that the system is able to forecast the evolution of the
Charles Jones and Pete Peterson, 1999, [28] have completed a parameters in next three hours giving previous indications about
research at the University of California, Santa Barbara, the possibility of rain, ice and fog [43]. Lekkas et al., 2004 have
California, for air surface temperature prediction over the city. used a multilayer back propagation network and found that BPN
Guhathakurta, 1999, [29,30] has implemented this technique for will not always find the correct weight and biases fort the
short-term prediction of surface ozone at Pune city. In this work optimum solution, whereas their results supported the hypothesis
multiple regression data analysis using ANN technique has been that ANNs can produce qualitative forecast. A 7 hour ahead
used. It has been observed that, the parallel model can be forecast in particular proves to be of fairly high precision,

2
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

especially when an error prediction technique is introduced to uncertainty level [57]. D. Nagesh et al., 2006, have used
the ANN model [44]. Chang Shu and Donald H. Burn, 2004, Artificial Intelligence techniques for forecasting regional rainfall
have found that artificial neural network ensembles generate and they found that this technique shows reasonably good
improved flood estimates and are less sensitive to the choice of accuracy for monthly and seasonal rainfall forecasting [58].
initial parameters when compared with a single artificial neural Guhathakurta, 2006, developed a model for rainfall forecast for
network [45]. In 2004 Nayak et al., have presented the the Kerala sub-division based on the area weighted value of all
application of an adaptive neuro-fuzzy inference system district forecast. The performance was found satisfactory than
(ANFIS) to hydrologic time series modeling, and it was the statistical technique [59].
observed that the ANFIS model preserves the potential of the
ANN approach fully, and eases the model building process [46]. As per the utmost necessities of the hydrologists around the
M.Asce et al., 2004, have applied two ANN-hydrologic globe, Bustami et al., 2007, have studied ANN for precipitation
forecasting models and they found encouraging results indicating and water level prediction; they found that ANN is an effective
that ANN-hydrologic forecasting models can be considered an tool in forecasting both missing precipitation and water level
alternate and practical tool for stream-flow forecast [47]. In 2004 data [60]. Paras et al., 2007, have introduced a pioneering
R.E. Abdel-Aal, studied alternative abductive networks approach, Feature Based Neural Network Model for Weather Forecasting
and concluded that the performance is significantly superior to and the results were very encouraging and it is found that the
naive forecasts based on persistence and climatology [48]. In feature based forecasting model can make predictions with high
coastal areas it is an enormous prediction of tidal level, Tsong- degree of accuracy [61]. Mohsen Hayati, and Zahra Mohebi,
Lin Lee, 2004, has predicted long-term tidal level using back 2007, have used ANN in a new experiment of short term
propagation neural network, as compare to conventional temperature forecasting(STTF) and he found that MLP network
harmonic method, he concluded that back-propagation neural has the minimum forecasting error and can be considered as a
network mode also efficiently predicts the long-term tidal levels good method to model the STTF systems [62]. Using time series
[49]. of draught indices with artificial neural network Morid et al.,
2007, have tested number of different ANN models for both
To estimate the maximum surface temperature and relative Effective Drought Index (EDI) and the Standard Precipitation
humidity a Feed forward multi-layered artificial neural network Index (SPI) with the lead times of 1 to 12 months [63]. As load
model is designed by Chaudhuri, and Chattopadhyay, in 2005, forecasting is an important prediction aspect for industrial
and stated that one hidden-layer neural network is an efficient sectors all over the world, Mohsen Hayati, and Yazdan Shirvany,
forecasting tool by which an estimation of maximum surface 2007, have put in an approach for short term load forecasting
temperature and maximum relative humidity can be obtained (STLF) using Artificial Neural Network, and they concluded that
[50]. A further contribution of Gwo-Fong Lin* and Lu-Hsien MLP network has the minimum forecasting error and can be
Chen, 2005, in neural network is that, two hidden layers is considered as a good method to model the STLF systems [64].
developed to forecast typhoon rainfall, and it has been observed To apply reliable and robust procedures for monthly
that the forecasting model can produce reasonable forecasts [51]. reconstruction of precipitation time series, Lucio et al., 2007,
Using an innovation in the researches Jon Vandegriff et al., have found that ANN can be applied to explore the
2005, have studied Forecasting space weather with ANN and spatiotemporal dependence of meteorological attributes [65]. In
they found that an ANN can be trained to predict the shock another experiment Hartmann et al., 2007, have found that the
arrival with better accuracy than existing methods [52]. Ozgur neural network algorithms are capable of explaining most of the
KISI, 2005, has selected three simple neural network (NN) rainfall variability even it can predict the summer rainfall also
architectures, i.e. ANN, Auto-Regressive Models and sum of [66].
square errors, for comparison of forecasting probabilities and he Aliev et al., 2008, have proposed, fuzzy recurrent neural network
found that NNs were able to produce better results than AR (FRNN) based time series forecasting method for solving
models when given the same data inputs [53]. Exploring the new forecasting problems, in an experiment and they found that The
concept, soft computing models based on Radial Basis Function performance of the proposed method for forecasting fuzzy time
Network for 24-h weather forecasting, Maqsood et al., have series shows its high efficiency and effectiveness for a wide
concluded that the RBFN produces the most accurate forecasts domain of application areas ranging from weather forecasting to
compared to the MLP, ERNN and HFM [54]. planning in economics and business [67]. Chattopadhyay and
In 2006, Somvanshi et al., have proved that ANN model can be Chattopadhyay, 2008, have worked out to find out best hidden
used as an appropriate forecasting tool to predict the rainfall, layer size for three layered neural net in predicting monsoon
which out performs the ARIMA (Autoregressive Integrated rainfall in India, and they have found that eleven-hidden-nodes
Moving Average) model [55]. As in the previous researches we three-layered neural network has more efficacy than asymptotic
have observed that most of the researchers have been used regression in the present forecasting task [68]. Hung et al., 2008,
Artificial Neural Network for various annual predictions like have developed a new ANN model for forecasting rainfall from
rainfall, tide, temperature etc, but in this study Niravesh Srikalra 1 to 6 h ahead at 75 rain gauge stations in the study area as
and Chularat Tanprasert have used Artificial Neural Network forecast point from the data of 3 consecutive years (1997–1999),
for daily rainfall prediction in Chao Phraya River with Online and they observed that the developed ANN model can be used
Data Collection, and they found that it is possible to predict for real-time rainfall forecasting and flood management [69]. In
rainfall on daily basis with acceptably accuracy using Artificial an comparative study between Artificial Intelligence and
Neural Network [56]. A.D. Kumarasiri and D.U.J. Sonnadara, Artificial Neural Network for rainfall runoff modeling, Aytek et
2006, have applied an innovative technique for rainfall al.,2008, have found that genetic programming (GP) formulation
forecasting using Artificial Neural Networks based on feed- performs quite well compared to results obtained by ANNs and
forward back-propagation architecture. Three Neural Network is quite practical for use. It is concluded from the results that
models were developed; a one-day-ahead model for predicting GEP can be proposed as an alternative to ANN models [70].
the rainfall occurrence of the next day, which was able to make Chattopadhyay et al.,2008, have studied the complexities in the
predictions with a 74.25% accuracy, and two long term relationship between rainfall and sea surface temperature (SST)
forecasting models for monthly and yearly rainfall depth anomalies during the winter monsoon using scatter plot matrices
predictions with 58.33% and 76.67% accuracies within a 5% and autocorrelation functions, and they found that the statistical

3
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

assessment revealed the potential of artificial neural network [85]. Tiron, and Gosav, 2010, have estimated rainfall from
over exponential regression [71]. However Mar, and Naing, BARNOVA WSR-98 D Radar using Artificial Neural Network
2008, have tested more over 100 cases by changing the number and the efficiency of ANN in the estimation of the rain rate on
of input and hidden nodes from 1 to 10 nodes, respectively, and the ground in comparison with that supplied by the weather radar
only one output node in an optimum artificial neural network is evaluated [86]. Goyal and Ojha, 2010, have focused their
architecture and they concluded that 3 inputs-10 hiddens-1 working on a concept of using dimensionless variables as input
output architecture model gives the best prediction result for and output to ANN, finally they have concluded that ANN
monthly precipitation prediction [72]. model using dimensionless variables were able to provide a
better representation of rainfall–runoff process in com-parison
Karmakar et.al., 2008, et al., 2008, have developed the ANN with the ANN models using process variables investigated in this
models for Long-Range Meteorological Parameters Pattern study [87]. On the basis of humidity, dew point and pressure in
Recognition over the Smaller Scale Geographical Region and the India, Enireddy et al., 2010, have used the back propagation
performances of these models in pattern recognition and neural network model for predicting the rainfall. In the training
prediction have been found to be extremely good [74]. they have obtained 99.79% of accuracy and 94.28% in testing.
From these results they have concluded that rainfall can
Hocaoglu et al., 2009, have developed adaptive neuro-fuzzy predicted in future using the same method [88]. Haghizadeh et
inference system for missing wind data forecasting [75]. In a al., 2010, have proposed ANN model and Multiple Regression
Case Study on Jarahi Watershed, Karim Solaimani, 2009, has (MR) for prediction of total sediment at basin scale and they
studied Rainfall-runoff Prediction Based on Artificial Neural found that estimated rate of sediment yield by Artificial neural
Network and he found that Artificial Neural Network method is networks is much better fits with the observed data in
more appropriate and efficient to predict the river runoff than comparison to MR model [89]. Subhajini and Raj, 2010, have
classical regression model [76]. KOŠCAK et al., 2009, have put in a Computational Analysis of Optical Neural Network
compared common meteorological forecasting method with Models to Weather Forecasting; in this study they have
ANN and he found the performance of ANN with high accuracy compared Electronic Neural Network (ENN) model and opto-
[77]. Karamouz et al., 2009, have experimented to perform long electronic neural network model. Overall their conclusion was,
lead rainfall forecasting Using Statistical Downscaling and the training of opto-electronic neural network is fast compared to
Artificial Neural Network Modeling; finally they found that the ANN. The accuracy of optoelectronic neural network is as good
SDSM outperforms the ANN model [78]. In a comparative study as ENN [90]. Durdu Omer Faruk, 2010, has experimented with
between ASTAR and ARIMA methods for rainfall forecasting in A hybrid neural network and ARIMA model for water quality
Indonesia Otok, and Suhartono, 2009, have concluded that the time series prediction. He has provided the results that the hybrid
best model is ASTAR model both in sample and out-sample data model provides much better accuracy over the ARIMA and
[79]. neural network models for water quality predictions [91]. To
identify and forecast the intensity of wind power and wind
It can be well-known that neural network can applied for most of speed, Soman et al., 2010, have applied ANN and hybrid
the prediction aspects, Nekoukar et al., 2010, have used radial techniques over different time-scales, they found the accuracy in
basis function neural network for financial time-series prediction associated with wind power and speed, based on
forecasting, and the result of their experiment shows the numeric weather prediction (NWP) [92].
feasibility and effectiveness [80]. Weerasinghe et al., 2010, have
tested the performance of neural network, in an experiment, for Forecasting daily rainfall at Mashhad Synoptic Station, Khalili et
forecasting daily precipitation using multiple sources, A cluster al., 2011, have applied Artificial Neural Networks model and
of ten neighboring weather stations having 30 years of daily they found that the black box model is capable of predicting the
precipitation data (1970 – 1999) was used in training and testing rainfall [93]. Pan et al., 2011, have experimented with feed
the models. Twenty years of daily precipitation data were used to forward neural network to predict Typhoon Rainfall. FNN is
train the networks while ten years of daily precipitation data applied to estimate the residuals from the linear model to the
were used to test the effectiveness of the models. They found differences between simulated rainfalls by a typhoon rainfall
that the models were able to predict the occurrence of daily climatology model (TRCM) and observations and their results
precipitation with an accuracy of 79±3% and Fuzzy were satisfactory [94]. Joshi and Patel, 2011, have put in a
classification produced a higher accuracy in predicting 'trace' review report on Rainfall-Runoff modeling using ANN, in the
precipitation than other categories [81]. Luenam et al, , 2010 same study they have reviewed three neural network methods,
have presented a Neuro-Fuzzy approach for daily rainfall Feed Forward Back Propagation (FFBP), Radial Basis Function
prediction, and their experimental results show that overall (RBF) and Generalized Regression Neural Network (GRNN)
classification accuracy of the neuro-fuzzy classifier is and they have seen that GRNN flow estimation performances
satisfactory [82]. Wu et al., 2010 have attempted to seek a were close to those of the FFBP, RBF and MLR [95]. El-Shafie
relatively optimal data-driven model for Rainfall time series et al., 2011, have developed two rainfall prediction models i.e.
forecasting using Modular Artificial Neural Networks, they Artificial Neural Network model and Multi regression model
found that the normal mode indicate, MANN performs the best (MLR). An analysis of two statistical models developed for
among all four models, but the advantage of MANN over ANN rainfall forecast on yearly and monthly basis in Alexandria,
is not significant in monthly rainfall series forecasting [83]. To Egypt shows that an ANN has a better performance than an
predict the intensity of rainfall using artificial neural network MLR model [96]. Rainfall forecasting in a mountainous region
Nastos et al., 2010 have developed prognostic models and they is a big task in itself Mekanik et al., 2011, have tried to do it
have proved that the results of the developed and applied ANN using ANN modeling a feed forward ANN rainfall model was
models showed a fairly reliable forecast of the rain intensity for developed to investigate its potentials in forecasting rainfall. A
the next four months [84]. Patil and Ghatol, 2010, have used monthly feed forward multi layer perceptron ANN rainfall
various ANN topologies such as radial basis functions and forecasting model was developed for a station in the west
multilayer perceptron with Levenberg Marquardt and mountainous region of Iran [97]. The temperature has a great
momentum learning rules for predicting rainfall using local effect in forecasting rainfall Amanpreet Kaur, and Harpreet
parameters and they found the topologies fit for the same task Singh, 2011, have tested Artificial Neural Network in forecasting

4
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

minimum temperature, they have used multi layer perceptron

architecture to model the forecasting system and back At the end of literature survey from 1923 – 2012, In 2012,
propagation algorithm is used to train the network. They found Sawaitul et al., have presented an approach for classification and
that minimum temperature can be predicted with reasonable prediction of future weather using back propagation algorithm,
accuracy using ANN model [98]. El-Shafie et al., 2011, have and discussed different models which were used in the past for
proposed an idea of using adaptive neuro-fuzzy inference system weather forecasting, finally the study concludes that the new
based model for rainfall forecasting on monthly basis and they technology of wireless medium can be used for weather
found that the ANFIS model showed higher rainfall forecasting forecasting process. It also concludes that the Back Propagation
accuracy and low error compared to the ANN model [99]. Algorithm can also be applied on the weather forecasting data.
Tripathy et al., 2011, have experimented with ANN and Particle Neural Networks are capable of modeling a weather forecast
Swarm Optimization (PSO) Technique for weather forecasting system [110].
and their experimental results indicate that the proposed
approach is useful for weather forecasting [100]. As we have 3. METHODOLOGY
observed that many of the scientists have used ANN and various
ANN models for forecasting Rainfall, Temperature, Wind and Two main architectures of ANN have been found by the above
Flood etc., El-Shafie et al., 2011, have compared and studied literature review and those are sufficiently suitable to predict
Dynamic Vs Static neural network models for rainfall chaotic behavior of monsoon rainfall are discussed in the
forecasting, they have developed AI based forecasting following subsections-
architectures using Multi-Layer Perceptron Neural Networks
(MLPNN), RBFNN and Adaptive Neuron-Fuzzy Inference
Systems (ANFIS), finally they concluded that the dynamic 3.1. Radial Basis Function Network (RBFN)
neural network namely IDNN could be suitable for modeling the
temporal dimension of the rainfall pattern, thus, provides better Lee et al., 1998, have studied that the spatial interpolation
forecasting accuracy [101]. Geetha and Selvaraj, 2011, have comparison 97 is concerned with predicting the daily rainfall at
predicted Rainfall in Chennai using back propagation neural 367 locations based on the daily rainfall at nearby 100 locations
network model, by their research the mean monthly rainfall is in Switzerland. The whole region is divided into four sub-areas
predicted using ANN model. The model can perform well both and each is modeled with a different method. Predictions in two
in training and independent periods [102]. In various researches larger areas were made by RBF networks based on the location
ANNs have been extensively used for simulation of rainfall- information only. The two smaller areas were assumed to be
runoff and other hydrological processes. Reshma et al., 2011, implemented by the Orographic Effect which dictates that
have applied Artificial Neural Network for determination of precipitation is proportional to elevation. Thus, predictions in
Distributed Rainfall- Runoff Model Parameters and they found these two areas were made using a simple linear regression
that the ANN technique can be successfully employed for the model based on the elevation information only. They have
purpose of estimation of model parameters of distributed concluded that the RBF networks produced good prediction
rainfall-runoff model [103]. Mohan Raju et al., 2011, have while the linear models poor prediction [23].
developed Artificial Neural-Network-Based Models for the After a study of RBFN, Chang et al., 2001, have found
Simulation of Spring Discharge, as their training and testing that RBFN is a suitable technique for a rainfall runoff model for
results revealed that the models were predicting the weekly three hours ahead floods forecasting [34]. He has proposed radial
spring discharge satisfactorily [104]. Predicting groundwater basis function (RBF) neural network (NN) to develop a rainfall–
level is somehow a difficult task now a day because it varies runoff model for three-hour-ahead flood forecasting. The
place to place and round the globe, Mayilvaganan, and Naidu, rainfall–runoff relation can be considered as a linear
2011, have attempted to forecast groundwater level of a combination of some nonlinear RBFs. Rainfall and runoff events
watershed using ANN and Fuzzy Logic. A three-layer feed- of the Lanyoung River collected during typhoons are used to
forward ANN was developed using the sigmoid function and the train, validate, and test the network. The results show that the
back propagation algorithm. Now it has been observed that RBF NN can be considered as a suitable technique for predicting
ANNs perform significantly better than Fuzzy Logic [105]. El- flood flow.
shafie et al., 2011, have tried to use neural network and
regression technique for rainfall-runoff prediction finally they Maqsood et al., 2004 have proposed ensemble model
concluded that the results showed that the feed forward back performance is contrasted with multi-layered perceptron network
propagation ANN can describe the behaviour of rainfall-runoff (MLPN), Elman recurrent neural network (ERNN), RBFN,
relation more accurately than the classical regression model Hopfield model (HFM) predictive models and regression
[106]. In another research integrated artificial neural network- techniques. The data of temperature, wind speed and relative
fuzzy logic-wavelet model is employed to predict Long term humidity are used to train and test the different models. With
rainfall by Afshin et al., 2011. The results of the integrated each model, 24-h-ahead forecasts are made for the winter,
model showed superior results when compared to the two year spring, summer and fall seasons. Moreover, the performance and
forecasts to predict the six-month and annual periods. As a result reliability of the seven models are then evaluated by a number of
of the root mean squared error, predicting the two-year and statistical measures. Among the direct approaches employed,
annual periods is 6.22 and 7.11, respectively. However, the empirical results indicate that HFM is relatively less accurate
predicted six months shows 13.15. [107]. Siou et al., 2011, have and RBFN is relatively more reliable for the weather forecasting
experimented with Complexity selection of a neural network problem. In comparison, the ensemble of neural networks
model for flood forecasting, these models yield very good produced the most accurate forecasts. [42].
results, and the forecasted discharge values at the Lez spring are
acceptable up to a 1-day forecasting horizon [108]. Saima et al., Exploring the new concept, soft computing models based on
2011, have reviewed on the various forecasting methods and RBFN for 24-h weather forecasting, Maqsood et al., have
hybrid models, they have put in a contradictory conclusion that applied a soft computing model based on a RBFN for 24-h
there is no such model exists that can forecast accurately in all weather forecasting of southern Saskatchewan, Canada. The
situations. This is because the distinct nature of the model [109]. model is trained and tested using hourly weather data of
temperature, wind speed and relative humidity in 2001. The

5
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

performance of the RBFN is compared with those of multi- a large number of observed tidal data using the spectral method
layered perceptron (MLP) network, Elman recurrent neural based on two years of data. In addition, the parameters in the
network (ERNN) and Hopfield model (HFM) to examine their harmonic tide level model are estimated by the method of least
applicability for weather analysis. Reliabilities of the models are squares for a long duration (more than one month). The BPN
then evaluated by a number of statistical measures. The results method only requires a part of observed tidal data to determine
indicate that the RBFN produces the most accurate forecasts the total number of constituent tides and the harmonic
compared to the MLP, ERNN and HFM. They have concluded parameters. The results of the case study show that the major
that the RBFN produces the most accurate forecasts compared to constituents can be obtained by using two-month measured data.
the MLP, ERNN and HFM [54]. Comparing the BPN model results and the harmonic data, BPN
Patil and Ghatol, 2010, have revealed that the flapping of for predicting one-year tidal level can be satisfactorily achieved
butterfly wings causing a tropical cyclone. Numerous researcher with 15 days of observed data. [49].
have introduces number of models for rainfall forecasting. No
two model forecast the same situation in exactly same way. On On the basis of humidity, dew point and pressure in India,
the same time no single model is reliable for rainfall forecasting. Enireddy et al., 2010, have used the back propagation neural
Climate and rainfall are highly non-linear phenomena in nature network model for predicting the rainfall. The back propagation
exhibiting what is known as the "butterfly effect". They have neural network model is used for predicting the rainfall based on
used various ANN topologies such as radial basis functions and humidity, dew point and pressure in the country INDIA. Two-
multilayer perceptron with Levenberg Marquardt and Third of the data was used for training and One-third for testing.
momentum learning rules for predicting rainfall using local The number of training and testing patterns is 250 training and
parameters and they found the topologies fit for the same task 120 testing. In the training they have obtained 99.79% of
[85]. accuracy and 94.28% in testing. For rainfall prediction, Artificial
Neural Network was applied and the rainfall was predicted in
Joshi and Patel, 2011, have put in a review report on Rainfall- India. According to the results back propagation neural network
Runoff modeling using ANN, in the same study they have were acceptably accurate and can be used for predicting the
reviewed three neural network methods, Feed Forward Back rainfall. So by using this method for prediction we can find the
Propagation (FFBP), Radial Basis Function (RBF) and amount of rainfall in the region by using the attributes like
Generalized Regression Neural Network (GRNN) and they have humidity, dew point and pressure. From these results they have
seen that GRNN flow estimation performances were close to concluded that rainfall can predicted in future using the same
those of the FFBP, RBF and MLR. The theoretical basis of the method [88].
RBF approach lies in the field of interpolation of multivariate
functions. The solution of exact interpolating RBF mapping ANN approach is extensively used in the water resources
passes through every data point. Different number of hidden literature. In most of the studies of ANN either the conventional
layer neurons & spread constants were tried in study [95]. BPN or the RBFN was employed to train the neural networks.
Joshi and Patel, 2011, have put in a review report on Rainfall-
As we have observed that many of the scientists have used ANN Runoff modeling using ANN, in the same study they have
and various ANN models for forecasting Rainfall, Temperature, reviewed three neural network methods, BPN, RBFN and
Wind and Flood etc., El-Shafie et al., 2011, have compared and GRNN and they have seen that GRNN flow estimation
studied Dynamic Vs Static neural network models for rainfall performances were close to those of the FFBP, RBF and MLR.
forecasting, they have developed AI based forecasting An important result of the study is the positive contribution of
architectures using Multi-Layer Perceptron Neural Networks the utilization of initial statistical analysis results in determining
(MLPNN), RBFNN and Adaptive Neuron-Fuzzy Inference the ANN input layer node number [95].
Systems (ANFIS), finally they concluded that the dynamic
neural network namely IDNN could be suitable for modeling the The temperature has a great effect in forecasting rainfall Kaur
temporal dimension of the rainfall pattern, thus, provides better and Singh, 2011, have tested ANN in forecasting minimum
forecasting accuracy [101]. temperature, they have used multi layer perceptron architecture
to model the forecasting system and back propagation algorithm
3.2. Back-Propagation Network (BPN) is used to train the network. Neural networks have been trained
to perform complex functions in various fields of application
Lekkas et al., 2004 have used a multilayer back propagation including forecasting, pattern recognition, identification,
network and found that BPN will not always find the correct classification and control systems. From this work it can be
weight and biases for the optimum solution, whereas their results concluded that a feed-forward ANN model using back-
supported the hypothesis that ANNs can produce qualitative propagation algorithm is developed to identify the minimum
forecast. A 7 hour ahead forecast in particular proves to be of temperature. The results show that an appropriate accuracy can
fairly high precision, especially when an error prediction be achieved using this network. The neural network approach for
technique is introduced to the ANN model. In real-time weather forecasting is capable of yielding good results and can
applications, like flow regulation and flood forecasting where the be considered as an alternative to traditional meteorological
precision and modeling speed are crucial, black box models and approaches. Further, this approach is also able to determine the
signal processing techniques need to be implemented. As is values of other parameters like maximum relative humidity,
presented by this case study, error prediction manages to correct minimum relative humidity, maximum temperature etc in a
the model output and produce an excellent forecast. [44]. particular year. They found that minimum temperature can be
predicted with reasonable accuracy using ANN model [98].
In coastal areas it is an enormous prediction of tidal level,
Tsong-Lin Lee, 2004, has predicted long-term tidal level using Since the mechanisms of rainfall are still not understood well,
BPN network, as compare to conventional harmonic method, he ANNs are a good choice worth trying to analyze the relationship
concluded that back-propagation neural network mode also between meteorological parameters and rainfall. A real world
efficiently predicts the long-term tidal levels. The main case study was observed in Chennai for 32 years of monthly
constituents of tide in the conventional harmonic method require mean data with meteorological parameters such as wind speed,

6
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

mean temperature, relative humidity, aerosol values (RSPM) in runoff etc, and found significant results using the same
the area were used to develop the ANN model. In order to architectures. Hence in literature review of moreover 25 years of
forecast rainfall in Chennai City, BPNs, a data driven technique researches. Most of the scientists have concluded that BPNN and
based on the working principle of biological neurons are applied RBFN are the appropriate method to predict weather
in this study. Geetha and Selvaraj, 2011, have predicted Rainfall phenomenon. In the development of Recurrent Sigma-pi Neural
in Chennai using back propagation neural network model, by Network Rainfall Forecasting System in Hong Kong by Chow
their research the mean monthly rainfall is predicted using ANN and Cho, 1997, it has been observed that A Recurrent Sigma-Pi
model. They have observed that the model can perform well both neural network was selected as the network architecture
in training and independent periods [102]. providing strong dynamical properties for the modeling of this
type of non-linear time series. The overall forecasting relative
Predicting groundwater level is somehow a difficult task now a percentage error is about 3-5%, the results indicate that the
day because it varies place to place and round the globe, neural network-based now casting system is capable of
Mayilvaganan, and Naidu, 2011, have attempted to forecast providing a reliable rainfall now casting in Hong [22]. Lee et al.,
groundwater level of a watershed using ANN and Fuzzy Logic. 1998, have applied the "divide-and-conquer" approach to divide
A three-layer feed-forward ANN was developed using the the whole region into four areas and the problem was solved
sigmoid function and the back propagation algorithm. Now it has separately based on the following assumptions: first, rainfall has
been observed that ANNs perform significantly better than a different pattern in different areas; second, rainfall pattern
Fuzzy Logic [105]. within smaller areas is continuous and smooth; third, orographic
effect exists. Predictions of the two large areas using RBF
El-shafie et al., 2011, have tried to use neural network and networks were reasonably good. But, for the two small areas,
regression technique for rainfall-runoff prediction finally they predictions were not good since the orographic effect was not
concluded that the results showed that the BPN can describe the apparent [23]. In a comparative study of various ANN
behaviour of rainfall-runoff relation more accurately than the architectures such as, MLP, ERNN, HFM and RBFN, performed
classical regression model [106]. by Maqsood et al., 2004 have strongly favoured that the
performance of RBFN is more promising and accurate than other
Sawaitul et al., 2011, have presented an approach for three methods [42]. For prediction of tidal level in coastal areas
classification and prediction of future weather using back Tsong-Lin Lee, 2004, has tried BPN technique and found to be
propagation algorithm, and discussed different models which an appropriate solution as compare to other conventional
were used in the past for weather forecasting, finally the study harmonic methods [49]. An innovative application of ANN for
concludes that the new technology of wireless medium can be weather forecasting in space is applied by Jon Vandegriff et al.,
used for weather forecasting process. It also concludes that the 2005, and they have found that an average of the absolute error
BPN Algorithm can also be applied on the weather forecasting using this approach is about 14.5 for both training and test data
data. Neural Networks are capable of modeling a weather sets. The network consistently beats this simple forecast, and so
forecast system [110]. ANN provides better accuracy than existing methods for
forecasting shock arrivals in the space weather [52]. Ozgur KISI,
3.3. Other Methods 2005, has selected three simple neural network architectures, i.e.
ANN, Auto-Regressive Models and sum of square errors, for
After development of recurrent Sigma- Pi neural network rainfall comparison of forecasting probabilities and he found that NNs
forecasting system, Chow and Cho, 1997, have concluded that were able to produce better results than AR models when given
the neural network based now casting system is capable of the same data inputs. The potential of ANN models for
providing a reliable rainfall now casting in Hong Kong [22]. simulating the hydrologic behavior of stream flow has been
Aliev et al., 2008, have proposed, fuzzy recurrent neural network presented in this study. The greatest difficulty lay in determining
(FRNN) based time series forecasting method for solving the appropriate model inputs for such a problem. The results
forecasting problems, in an experiment and they found that the obtained with ANNs for 1-day ahead forecasts are better than
performance of the proposed method for forecasting fuzzy time those reached in the AR models and confirm the ability of this
series shows its high efficiency and effectiveness for a wide approach to provide a useful tool in solving a specific problem in
domain of application areas ranging from weather forecasting to hydrology, that of stream flow forecasting. The results suggest
planning in economics and business [67]. In 2001 Luk et al., that the ANN approach may provide a superior alternative to the
have developed and compared three types of ANNs suitable for AR models for developing input {output simulations and
rainfall prediction i.e. multilayer feed forward neural network, forecasting models in situations that do not require modeling of
Elman partial recurrent neural network and time delay neural the internal structure of the watershed [53]. In 2006, Somvanshi
network [32]. For a rainfall runoff relationship Harun and Irwan, et al., have proved that ANN model can be used as an
2002, have concluded that the performance of neural network appropriate forecasting tool to predict the rainfall, which out
model is better than HEC-HMS and MLR models for modeling performs the ARIMA model. Complexity of the nature of annual
the rainfall runoff relationship [37]. In 2004 Maqsood et al., have rainfall record has been studied using the ANN and ARIMA
found that HFM is relatively less accurate and RBFN is techniques. An annual rainfall data spanning over a period of
relatively more reliable for the weather forecasting problems and 1901- 2003 of Hyderabad region was used to develop and test
in comparison the ensembles of neural networks produced the the models. Autocorrelation and partial autocorrelation
most accurate forecast [42]. coefficient for various lags (in year) of rainfall data was used to
find out number of past observations as inputs to neural network.
4. RESULTS The present analysis uses four past observations as inputs to
neural network model. The study reveals that ANN model can be
After review of a wide range of ANN architectures for rainfall used as an appropriate forecasting tool to predict the rainfall,
forecasting, it has been observed that most of the researchers which out performs the ARIMA model. Further refinement of
have used BPN and RBFN techniques for forecasting various the model using the data separately from the different zones of
weather phenomenon e.g. rainfall, temperature, flood, rainfall- the country may be useful for the long-term prediction [55].
Using ANNs and Fuzzy Logic, Mayilvaganan, and Naidu, 2011,

7
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

have tried to predict ground water level and they have concluded
that ANN performs better than Fuzzy Logic [105]. It has been 5000
also proved by the contribution of Karmakar et al., 2008, 2009,
2010, that BPN in deterministic as well as parametric forecast 0 Actual

TMRF
are more efficient technique over the statistical model for

(in
Predicted Training period (1952-1991)

19
52
19
58
19
64
forecasting long-range monsoon rainfall over the high resolution
geographical region such as district or sub-division level [111-
114]. He has successfully obtained global minima up to the level
of 10-04 during the training period. And also has obtained more
than 80% accuracy in prediction during the independent period. Fig. 3. Performance of BPN in deterministic forecast during the
To predict long-range TMRF (in mm.) over the smaller scale training period (1951-1991)
geographical region “district”, eight parameters BPN model has
been developed [47]. Eight parameters BPN model is depicted in
the following Fig. 1 Wherein, eight input parameters which are
physically connected with TMRF over this region are used to
input in the BPN. Three neurons in hidden layer, one neuron in
output layer, a total of 8.3+3=27 trainable weights including 3
biases in hidden layer and 1 bias in output layer have been used
to train the model and observe TMRF as model output. It is
observed that three neurons can be utilize in hidden layer
therefore three neurons in hidden layer is chosen. The detail Fig. 4. Absolute deviation between actual TMRF and predicted
discussion of selection of neurons in hidden layer is given in TMRF during training period (1951-1991)
following subsection. The neuron output is obtained as f (xj).
Where f is a transfer function typically the sigmoid function. 3000
The skeleton of the model is shown in Fig. 2 and Fig. 3. 2000
1000
TMRF (in
0 Actual
Predicted

Independent period (1992-2004)

Fig. 5. Performance of BPN in deterministic forecast during the
independent/testing period (1992-2004)

Fig. 1. 8-Parameters BPN Model

The initial random trainable weights including biases on hidden

and output layer as shown in Fig. 1 is selected randomly by
using “Random” class of Java is optimized by exercising 20,
00000 epochs. The optimized weights are shown in Table 1. The
trained network presented excellent learning curve started with
local minima MSE = 0.0010583407253476 to global minima Fig. 6. Absolute deviation between actual TMRF and predicted
MSE = 6.555600546440187E-04 that is exceptionally close to TMFR during independent period/testing period
the optimum value of 0 as shown in the Fig. 2.
During an intense study of applications of various architectures
1.0000000000000000 of ANN, it has been found that the BPN and RBFN are the
0.18447
0.1000000000000000 methods which have been used by most of the researchers and
1 12091806107061010100100020000000000
the result of their experiment found to be satisfactory without
0.0100000000000000 3160
0.00182 any scientific controversy. In general it has been observed that
0.0010000000000000 0.001050.001049.98675 out of
8340702.055051000.908071506345 75 various forecasting techniques such as statistical and
M

0.0001000000000000 05304673.7556565860
numerical modeling, over the meteorological data, ANN is
8
3476 32548655874869835E-04
30546440 proved to be an appropriate technique undoubtedly for
3988 7343E-04 forecasting various weather phenomenons.
Epochs
5. CONCLUSION
Fig. 2 Minimizing error (MSE) during training process BPN
model in parametric forecast has been developed and evaluated. This study concentrates on capabilities of ANN in prediction of
The performance of the model during the training period and several weather phenomenon such as rainfall, temperature, flood
independent/testing period is given in the and tidal level etc. finally it has been concluded that the major
Table. 1 and Fig. 1 respectively. architectures i.e. BPN, RBFN, MLP are sufficiently suitable to
predict weather phenomenon. In the comparative study among
various ANN techniques, BPN and RBFN are found as
appropriate solutions for prediction of long-range weather
forecasting. The study of BPN and RBFN for long range

8
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

meteorological parameters pattern recognition over smaller scale [13]. Parthasarathy, B., Rupa Kumar, K., and Munot, A. A.,
geographical region shows a good performance and reasonable 1991, “Evidence of Secular Variations in Indian Summer
prediction accuracy. However, some shortcoming also has been Monsoon Rainfall Circulation Relationships”. J. Climate, 4,
found. These models are exceptionally slow during the learning 927–938.
process. Sometime training process of these models may
introduce a temporal nervousness to the user. [14]. Hastenrath, S., and Greischar, L., 1993, “Changing
Predictability of Indian Monsoon Rainfall Anomalies”.
6. ACKNOWLEDGMENTS Proc. Indian Acad. Sci. (Earth Planet. Sci.), 102, 35–47.
[15]. Guhathakurta P., (2006), “Long-Range Monsoon
We are thankful to Bhilai Institute of Technology (BIT), Durg, Rainfall Prediction of 2005 for the Districts and Sub-
Chhattisgarh, India for providing sufficient paraphernalia, Division Kerala with Artificial Neural Network”, Current
facility as well as encouragement. Science, 90(6), pp-773-779.
[16]. Krishnamurthy, V., and Kinter, J. L., 2002, “The
7. REFERENCES Indian Monsoon and its Relation to Global Climate
Variability”. Global Climate – Current Research and
[1]. Walker, G. T., 1923, “Correlation in Seasonal Variations of
Uncertainties in the Climate System (eds Rodo, X. and
Weather, III. A Preliminary Study of World Weather”.
Comin, F. A.), 186–236.
Mem. India Meteorol. Dep., XXIV, 75–131.
[17].Krishnamurthy, V., and Kirtman, B. P., 2003, “Variability
[2]. Walker, G. T., 1924, “Correlation in Seasonal Variations of
of the Indian Ocean: Relation to Monsoon and ENSO". Q.
Weather, IV. A Further Study of World Weather”. Mem.
J. R.” Meteorol. Soc., 129, 1623–1646
India Meteorol. Dep., XXIV, 275– 332.
[18]. Sahai, A. K., Grimm, A. M., Satyan. V. and Pant, G. B.,
[3]. Gowariker, V., Thapliyal, V., Sarker, R. P., Mandal, G. S.
2002, “Prospects of Prediction of Indian Summer Monsoon
and Sikka, D. R., 1989, “Parametric and Power Regression
Rainfall using Global SST Anomalies”. IITM Research
Models: New Approach to Long Range Forecasting of
Report No. RR-093.
Monsoon Rainfall in India”. Mausam, 40, 115– 122.
[19]. Guhathakurta, P., 1999, “Long Range Forecasting
[4]. Gowariker, V., Thapliyal, V., Kulshrestha, S. M., Mandal,
Indian Summer Monsoon Rainfall By Principle Component
G. S., Sen Roy, N., and Sikka, D. R., 1991, “A Power
Neural Network Model”, Meteor. & Atomos. Phys., 71,
Regression Model for Long Range Forecast of Southwest
255-266.
Monsoon Rainfall over India”. Mausam, 42, 125–130.
[20]. Guhathakurta, P., 1998, “A Hybrid Neural Network
[5]. Thapliyal, V., and Kulshrestha, S. M., 1992, “Recent
Model for Long Range Prediction of All India Summer
Models for Long Range Forecasting of Southwest Monsoon
Monsoon Rainfall”, Proceedings of WMO international
Rainfall over India”. Mausam, 43, 239–248.
workshop on dynamical extended range forecasting,
[6]. Thapliyal, V., 1997, “Preliminary and Final Long Range Toulouse, France, November 17-21, 1997, PWPR No. 11,
Forecasts for Seasonal Monsoon Rainfall over India”. J. WMO/TD. 881, pp 157-161.
Arid Environ., 36, 385–403.
[21]. Basu Sujit & Andharia H I, 1992, “The Chaotic time
[7]. Rajeevan, M., Guhathakurta, P. and Thapliyal, V., 2000, series of Indian Monsoon rainfall and its prediction”, Proc.
“New Models for Long Range Forecasting of Monsoon Indian Acad. Sci., Vol. 101, No. 1, pp 27-34.
Rainfall over Northwest and Peninsular India”. Meteorol.
[22]. Chow T. W. S. and Cho S. Y., 1997, “Development of a
Atmos. Phys., 73, 211–225.
Recurrent Sigma-Pi Neural Network Rainfall Forecasting
[8]. Rajeevan M., 2001, “Prediction of Indian Summer System in Hong Kong”, Springer-Verlag, pp. 66-75.
Monsoon: Status, Problems and Prospects”. Current
[23]. Lee Sunyoung, Cho Sungzoon, Wong Patrick M., 1998,
Science, 81, 1451–1457.
“Rainfall Prediction Using Artificial Neural Networks”,
[9]. Thapliyal, V., and Rajeevan, M., 2003, “Updated Journal of Geographic Information and Decision Analysis,
Operational Models for Long-Range Forecasts of Indian vol. 2, no. 2, pp. 233 – 242.
Summer Monsoon Rainfall”. Mausam, 54, 495–504.
[24]. Hsieh, W.H., and Tang, B., 1998, “Applying Neural
[10]. Rajeevan, M., Pai, D.S., Dikshit, S.K., and Kelkar, R. Network Models to Prediction and Analysis in Meteorology
R., 2004, “IMD’s New Operational Models for Long-Range and Oceanography”. Bull. Amer. Met. Soc., 79, 855-1870.
Forecast of Southwest Monsoon Rainfall over India and
[25]. Dawson Christian W., Wilby Robert, “An artificial
Their Verification for 2003”, Current Science, 86 (3), 422-
neural network approach to rainfall runoff modeling”, 1998,
431.
Hydrological Sciences—Journal, 43(1), pp. 47-66.
[11]. Guhathakurta, P., 2000, “New Models for Long
[26]. Guhathakurta1 P., Rajeevan2 M., and Thapliyal2 V., 1999,
Range Forecasts of Summer Monsoon Rainfall over North
“Long Range Forecasting Indian Summer Monsoon
West and Peninsular India”, Meteor. & Atomos. Phys.,73
Rainfall by a Hybrid Principal Component Neural Network
(3), 211-255.
Model”, Meteorol. Atmos. Phys., 71, pp. 255-266.
[12]. Guhathakurta, P., Rajeevan, M., and Thapliyal, V.,
[27].Ricardo, M., Trigo, Jean, P., Palutikof, 1999, “Simulation
1999, “Long Range Forecasting Indian Summer Monsoon
of Daily Temperatures for Climate Change Scenarios over
Rainfall by Hybrid Principal Component Neural Network
Portugal: A Neural Network Model Approach”, University
Model”, Meteorology and Atmospheric Physics, Springer-
of East Anglia, Norwich, NR4 7TJ, United Kingdom,
Verlag, Austria.
climate research (Clim., Res.), 13, 45–59, 1999.

9
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

[28]. Jones, C., Peterson, P., 1999, “A New Method for [43]. Pasero Eros, Moniaci Walter, 2004, “Artificial Neural
Deriving Ocean Surface Specific Humidity and Air Networks for Meteorological Nowcast”, IEEE international
Temperature: An Artificial Neural Network Approach”, J. Conference,
Applied Meteorology, American Meteorological Society,
38, 1229-1245. [44]. Lekkas D.F.,* Onof1 C., Lee1 M. J., Baltas2 E.A.,
2004, “Application of artificial neural networks for flood
[29].Guhathakurta, P., 1999, “A Short Term Prediction Model forecasting”, Global Nest, Vol 6, No 3, pp 205-211.
for Surface Ozone At Pune: Neural Network Approach”,
Vayu mandal, Special issue on Asian monsoon and [45]. Shu Chang and Burn Donald H., 2004, “Artificial
pollution over the monsoon environment, 29(1-4), 355-358. neural network ensembles and their application in pooled
flood frequency analysis”, Water Resources Research, vol.
[30]. Guhathakurta, P., 1999, “A Neural Network Model for 40.
Short Term Prediction of Surface Ozone at Pune”, Mausam,
50(1), 91-98. [46]. Nayaka P.C.,*, Sudheerb K.P.,1Ranganc, D.M.,2,
Ramasastrid K.S.,3, 2004, “A neuro-fuzzy computing
[31]. Toth E. *, Brath A., Montanari A., 2000, “Comparison technique for modeling hydrological time series”, Journal of
of short-term rainfall prediction models for real-time flood Hydrology, Elsevier, 291, pp. 52–66.
forecasting”, Journal of Hydrology, Elsevier, 239, pp. 132-
147. [47]. Wu Jy S., P.E., ASCE1 M.; Han2 Jun; Annambhotla3
Shastri; and Scott Bryant4, 2004, “Artificial Neural
[32]. Luk Kin C, Ball J. E. and Sharma A., 2001, “An Networks for Forecasting Watershed Runoff and Stream
Application of Artificial Neural Networks for Rainfall Flows”, Journal Of Hydrologic Engineering, 216.
Forecasting”, 33, pp. 883-699.
[48]. Abdel-Aal R.E. *, 2004, “Hourly temperature
[33]. Michaelides Silas Chr,*, Pattichis Constantinos S. and forecasting using abductive networks”, Elsevier, 17, pp.
Kleovouloub Georgia, 2001, “Classification of rainfall 543–556.
variability by using artificial neural networks”, International
Journal Of Climatology, pp. 1401–1414. [49]. Lee Tsong-Lin, 2004, “Back-propagation neural
network for long-term tidal predictions”, Elsevier, 31, pp.
[34]. Chang Fi-John, Liang Jin-Ming, and Chen Yen- 225–238.
Chang,2001, “Flood Forecasting Using Radial Basis
Function Neural Networks”, IEEE Transactions On [50]. Chaudhuri Sutapa, Chattopadhyay Surajit, 2005, “Neuro-
Systems, Man, And Cybernetics, vol. 31, no. 4, pp. 530- computing based short range prediction of some
535. meteorological parameters during the pre-monsoon season”,
Springer-Verlag, 9, pp. 349–354.
[35]. Brath A., Montanari A. and Toth E., 2001, “Neural
networks and non-parametric methods for improving [51]. Lin Gwo-Fong * and Chen Lu-Hsien, 2005,
ealtime flood forecasting through conceptual hydrological “Application of an artificial neural network to typhoon
models”, Hydrology and Earth System Sciences, 6(4), pp. rainfall forecasting”, Hydrological Processes, 19, pp. 1825–
627-640. 1837.
[36]. Rajurkar M. P., Kothyari U. C., Chaube U. C., 2002, [52]. Vandegriff Jon *, Wagstaff Kiri, George Ho, Plauger
“Artificial neural networks for daily rainfall-runoff Janice, 2005, “Forecasting space weather: Predicting
modelling”, Hydrologkal Sciences-Journals, 47(6), pp. 865- interplanetary shocks using neural networks”, Elsevier, vol
877. 36, pp. 2323–2327.
[37]. Harun Sobri, Nor Irwan Ahmat & Mohd. Kassim Amir [53]. KISI Ozgur, 2005, “Daily River Flow Forecasting
Hashim, 2002, “Artificial neural network model for rainfall- Using Artificial Neural Networks and Auto-Regressive
runoff relationship”, Journal Technology, Malaysia, pp. 1- Models”, Turkish J. Eng. Env. Sci., vol 29, pp. 9 - 20.
12.
[54]. Maqsood Imran, Khan Muhammad Riaz, Huang Guo
[38].Iseri, Y. G. C., Dandy, R., Maier, A., Kawamura, and Jinno, H., Abdalla Rifaat, 2005, “Application of soft computing
K., 2002, “Medium Term Forecasting of Rainfall Using models to hourly weather analysis in southern
Artificial Neural Networks”, Part 1 background and Saskatchewan, Canada”, Elsevier, vol 18, pp. 115–125.
methodology, Journal of Hydrology , 301 (1-4), 1834-1840.
[55]. Somvanshi V.K., Pandey O.P., Agrawal P.K.,
[39]. Silva, A. P., 2003, “Neural Networks Application to Kalanker1 N.V., Prakash M.Ravi and Chand Ramesh, 2006,
Spatial Interpolation of Climate Variables”, Carried Out By, “Modelling and prediction of rainfall using artificial neural
STSM on the Framework of COST 719 ZAMG, Vienna 6- network and ARIMA techniques”, J. Ind. Geophys. Union,
10 October 2003. Vol.10, No.2, pp.141-151.
[40].Snell., Seth, E., Gopal, Sucharita, Kaufmann, Robert, K., [56]. Srikalra Niravesh and Tanprasert Chularat, 2006,
2003, “Spatial Interpolation of Surface Air Temperatures “Rainfall Prediction for Chao Phraya River using Neural
Using Artificial Neural Networks: Evaluating Their Use for Networks with Online Data Collection”, Malaysia, pp. 13-
Downscaling GCMs”, Journal of Climate, 13 (5), 886-895. 15.
[41].Blender, R., 2003, “Predictability Study of the Observed [57]. Kumarasiri A.D.and Sonnadara D.U.J., 2006,
and Simulated European Climate using Linear Regression”, “Rainfall Forecasting: An Artificial Neural Network
Q. J. R. Meteorol. Soc. (2003), 129, 2299–2313. Approach”, Proceedings of the Technical Sessions, vol 22,
pp. 1-13.
[42]. Maqsood Imran, Khan Muhammad Riaz, Abraham
Ajith, 2004, Neural Comput & Applic, 13, pp. 112–122. [58]. Kumar D. Nagesh, Reddy M. Janga and Maity Rajib,
2006, “Regional Rainfall Forecasting using Large Scale

1
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

Climate Teleconnections and Artificial Intelligence [73]. Karmakar Sanjeev et al., 2008, “Development of a
Techniques”, Journal of Intelligent Systems, Vol. 16, No.4, Satellite Based Agriculture Meteorological Information
pp. 307-322. System (AGRIMETCast) in the Context of Remote Places
[59].Guhathakurta, P., 2006, “Long-Range Monsoon Rainfall of Chhattisgarh Using Remote Sensing Data” , India
Prediction of 2005 for the Districts and Sub-Division Kerala Society of Remote Sencing, Ahmedabad, India., SAC
With Artificial Neural Network”, Current Science, 90 (6), (ISRO)
pp-773-779. [74]. Karmakar Sanjeev et al., 2008, “Development of
[60]. Bustami Rosmina, l BessaihNabi, Charles Bong, Artificial Neural Network Models for Long-Range
Suhaila Suhaili, 2007, “Artificial Neural Network for Meteorological Parameters Pattern Recognition over the
Precipitation and Water Level Predictions of Bedup River”, Smaller Scale Geographical Region”, IEEE Computer
International Journal of Computer Science, vol 34:2. Society, Washington, DC, USA., 8-10 Dec. 2008, pp.1 – 6.

[61]. Paras, Mathur Sanjay, Kumar Avinash, and Chandra [75]. Hocaoglu Fatih O., Oysal Yusuf, Kurban Mehmet, 2009,
Mahesh, 2007, “A Feature Based Neural Network Model “Missing wind data forecasting with adaptive neuro-fuzzy
for Weather Forecasting”, World Academy of Science, inference system”, Springer-Verlag London, vol. 18, pp.
Engineering and Technology, vol 34, pp. 66-73. 207–212.

[62]. Hayati Mohsen, and Mohebi Zahra, 2007, [76]. Solaimani Karim, 2009, “Rainfall-runoff Prediction
“Application of Artificial Neural Networks for Temperature Based on Artificial Neural Network (A Case Study: Jarahi
Forecasting”, World Academy of Science, Engineering and Watershed)”, American-Eurasian J. Agric. & Environ. Sci.,
Technology, vol 28, pp. 275-279. vol. 5(6), pp. 856-865.

[63]. Morid Saeid, Smakhtin Vladimir and Bagherzadeh K., [77]. KOŠCAK Juraj, JAKŠA Rudolf., SEPEŠI Rudolf,
2007, “Drought forecasting using artificial neural networks SINCÁK Peter., 2009, “Weather forecast using Neural
and time series of drought indices”, Royal Meteorological Networks”, 9th Scientific Conference of Young
Society, vol. 27, pp. 2103–2111. Researchers

[64]. Hayati Mohsen, and Shirvany Yazdan, 2007, [78]. Karamouz M.; . Fallahi M, Nazif S. and Farahani M.
“Artificial Neural Network Approach for Short Term Load Rahimi, 2009, “Long Lead Rainfall Prediction Using
Forecasting for Illam Region”, World Academy of Science, Statistical Downscaling and Arti cial Neural Network
Engineering and Technology, vol. 28, pp. 280-284. Modeling”, Transaction A: Civil Engineering, Vol. 16, No.
2, pp. 165-172.
[65]. Lucio P. S., Conde F. C., Cavalcanti I. F. A., Serrano
A. I., Ramos A. M., and Cardoso A. O., 2007, [79]. Widjanarko Bambang Otok, Suhartono, 2009,
“Spatiotemporal monthly rainfall reconstruction via “Development of Rainfall Forecasting Model in Indonesia
artificial neural network – case study: south of Brazil”, by using ASTAR, Transfer Function, and ARIMA
Advances in Geosciences, vol. 10, pp. 67–76. Methods”, European Journal of Scientific Research, Vol.38
No.3, pp.386-395.
[66]. Hartmann, Heikea * Becker Stefanb and Kinga
Lorenz, 2007, “Predicting summer rainfall in the Yangtze [80]. Nekoukar Vahab, Taghi Mohammad, Beheshti Hamidi,
River basin with neural networks”, Royal Meteorological 2010, “A local linear radial basis function neural network
Society. for financial time-series forecasting”, Springer Science, vol.
23, pp. 352–356.
[67]. Aliev R. A., Fazlollahi B., Aliev R. R., Guirimov B.,
2008, “Linguistic time series forecasting using fuzzy [81]. Weerasinghe H.D.P.,. Premaratne H.L and Sonnadara
recurrent neural network”, Soft Comput, vol. 12, pp. 183– D.U.J., 2010, “Performance of neural networks in
190. forecasting daily precipitation using multiple sources”,
J.Natn.Sci.Foundation Sri Lanka, vol. 38(3), pp. 163-170.
[68]. Chattopadhyay Surajit and Chattopadhyay Goutami,
2008, “Identification of the best hidden layer size for three [82]. Luenam Pramote, Ingsriswang Supawadee,
layered neural net in predicting monsoon rainfall in India”, Ingsrisawang Lily, Aungsuratana Prasert, and Khantiyanan
Journal of Hydroinformatics, vol. 10(2), pp. 181-188. Warawut, 2010, “A Neuro-Fuzzy Approach for Daily
Rainfall Prediction over the Central Region of Thailand”,
[69]. Hung N. Q., Babel M. S., Weesakul S., and Tripathi N. ISSN 2010, vol. 1.
K., 2008, “An artificial neural network model for rainfall
forecasting in Bangkok, Thailand”, Hydrology and Earth [83]. Wu C. L., Chau1 K. W, and Fan C., 2010, “Prediction
System Sciences, vol. 5, pp. 183–218. of Rainfall Time Series Using Modular Artificial Neural
Networks Coupled with Data Preprocessing Techniques”,
[70]. Aytek Ali, Asce M and Alp Murat, 2008, “An Journal of Hydrology, Vol. 389, No. 1-2, pp. 146-167.
application of artificial intelligence for rainfall–runoff
modeling”, J. Earth Syst. Sci., vol. 117, No. 2, pp. 145–155. [84]. Nastos Panagiotis, Moustris Kostas, Larissi Ioanna,
and Paliatsos Athanasios, 2010, “Rain intensity forecast
[71]. Chattopadhyay Goutami, Chattopadhyay Surajit, Jain using Artificial Neural Networks in Athens, Greece”,
Rajni, 2008, “Multivariate forecast of winter monsoon Geophysical Research Abstracts, Vol. 12.
rainfall in India using SST anomaly as a predictor:
Neurocomputing and statistical approaches”, [85]. Patil C. Y. and Ghatol A. A., 2010, “Rainfall
forecasting using local parameters over a meteorological
[72]. Win Khaing Mar and Thu Naing Thinn, 2008, station: an artificial neural network approach”, International
“Optimum Neural Network Architecture for Precipitation J. of Engg. Research & Indu. Appls, Vol.3, No. II, pp 341-
Prediction of Myanmar”, World Academy of Science, 356.
Engineering and Technology, vol. 48, pp. 130-134.

1
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

[86]. Tiron Gina, and Gosav Steluţa, 2010, “The july 2008 International Journal of the Physical Sciences, Vol. 6(12),
rainfall estimation from BARNOVA WSR-98 D Radar pp. 2875-2888.
using artificial neural network”, Romanian Reports in
Physics, Vol. 62, No. 2, pp. 405–413. [100]. Tripathy Asis Kumar, Mohapatra Suvendu, Beura
Shradhananda, Pradhan Gunanidhi, 2011. “Weather
[87]. Goyal Manish Kumar, Ojha Chandra Shekhar Prasad, Forecasting using ANN and PSO”, International Journal of
2010, “Analysis of Mean Monthly Rainfall Runoff Data of Scientific & Engineering Research, Volume 2, Issue 7,
Indian Catchments Using Dimensionless Variables by pp.1-5.
Neural Network”, Journal of Environmental Protection, vol.
1, pp. 155-171. [101]. El-Shafie A, Noureldin A., Taha M. R., and Hussain A.,
2011, “Dynamic versus static neural network model for
[88]. .Vamsidhar Enireddy Varma K.V.S.R.P..Sankara Rao rainfall forecasting at Klang River Basin, Malaysia”,
P satapati Ravikanth, 2010, “Prediction of Rainfall Using Hydrol. Earth Syst. Sci., vol. 8, pp. 6489–6532.
Backpropagation Neural Network Model”, International
Journal on Computer Science and Engineering, Vol. 02, No. [102]. Geetha G.,. Selvaraj R Samuel, 2011, “Prediction of
04, pp. 1119-1121. monthly rainfall in Chennai using back propagation neural
network model”, International Journal of Engineering
[89]. Haghizadeh Ali, Teang shui Lee, Goudarzi Ehsan, Science and Technology, Vol. 3 No. 1, pp. 211-213.
2010, “Estimation of Yield Sediment Using Artificial
Neural Network at Basin Scale”, Australian Journal of [103]. Reshma T., Reddy K. Venkata, Pratap Deva, 2011,
Basic and Applied Sciences, vol. 4(7), pp. 1668-1675. “Determination of Distributed Rainfall- Runoff Model
Parameters Using Artificial Neural Network”, International
[90]. Subhajini A . C. and Joseph Raj V., 2010, Journal of Earth Sciences and Engineering, Volume 04, No
“Computational Analysis of Optical Neural Network 06, pp. 222-224
Models to Weather Forecasting”, International Journal of
Computer Science Issue, Vol.7, Issue 5, pp. 327-330. [104]. Raju, M.Mohan,. Srivastava R. K, Bisht Dinesh C. S.,
Sharma H. C., and Kumar Anil, 2011, “Development of
[91]. Omer Faruk Durdu, 2010, “A hybrid neural network Artificial Neural-Network-BasedModels for the Simulation
and ARIMA model for water quality time series of Spring Discharge”, Hindawi Publishing Corporation,
prediction.”, Elsevier, vol 23, pp. 586–594 Volume 2011, pp. 1-11.
[92]. . Soman Saurabh S, Zareipour Hamidreza, Malik Om, [105]. Kavitha M.Mayilvaganan,.Naidu K.B, 2011, “ANN
and Mandal Paras, 2010, “A Review of Wind Power and and Fuzzy Logic Models for the Prediction of groundwater
Wind Speed Forecasting Methods With Different Time level of a watershed”, International Journal on Computer
Horizons”. Science and Engineering, Vol. 3 No. 6, pp. 2523-2530.
[93]. Khalili Najmeh, Khodashenas Saeed Reza, Davary [106]. El-shafie A., Mukhlisin M., Najah Ali A. and Taha M.
Kamran and Karimaldini Fatemeh, 2011, “Daily Rainfall R., 2011, “Performance of artificial neural network and
Forecasting for Mashhad Synoptic Station using Artificial regression techniques for rainfall-runoff prediction”,
Neural Networks”, International Conference on International Journal of the Physical Sciences, Vol. 6(8), pp.
Environmental and Computer Science, vol.19, pp. 118-123. 1997-2003.
[94]. Pan Tsung-Yi, Yang Yi-Ting, Kuo Hung-Chi, Tan Yih- [107]. Afshin Sarah, Fahmi Hedayat, Alizadeh Amin, Sedghi
Chi, Lai1 Jihn-Sung, Chang Tsang-Jung, Lee Cheng-Shang, Hussein and Kaveh Fereidoon, 2011, “Long term rainfall
and Hsu Kathryn Hua, 2011, “Improvement of Statistical forecasting by integrated artificial neural network-fuzzy
Typhoon Rainfall Forecasting with ANN-Based Southwest logic-wavelet model in Karoon basin”, Scientific Research
Monsoon Enhancement”, Terr. Atmos. Ocean. Sci, Vol. 22, and Essays, Vol. 6(6), pp. 1200-1208.
No. 6, pp. 633-645.
[108]. Siou1 Line Kong A, Johannet Anne, Borrell Valérie,
[95]. Joshi Jignesh, Patel Vinod M., 2011, “Rainfall-Runoff Pistre Séverin, 2011, “Complexity selection of a neural
Modeling Using Artificial Neural Network (A Literature network model for karst flood forecasting: The case of the
Review)”, National Conference on Recent Trends in Lez Basin (southern France)”, Journal of Hydrology, vol.
Engineering & Technology. 403, pp. 367-380.
[96]. El-Shafie Amr H., Shafie A. El-, Mazoghi Hasan G. El, [109]. Saima H., Jaafar J., Belhaouari S., Jillani T.A., 2011,
Shehata A. and Taha Mohd. R., 2011, “Artificial neural “Intelligent Methods for Weather Forecasting: A Review”,
network technique for rainfall forecasting applied to IEEE,
Alexandria, Egypt”,
[110]. Sawaitul Sanjay D., Prof. Wagh K. P., Dr. Chatur P.
[97]. Mekanik F., Lee T.S. and Imteaz M. A., 2011, N., 2012, “Classification and Prediction of Future Weather
“Rainfall modeling using Artificial Neural Network for a by using Back Propagation Algorithm-An Approach”,
mountainous region in West Iran”. International Journal of Emerging Technology and
[98]. Kaur Amanpreet, Singh Harpreet, 2011, “Artificial Advanced Engineering, Volume 2, Issue 1, pp. 110-113.
Neural Networks in Forecasting Minimum Temperature”, [111]. Karmakar, S., Kowar, M. K., Guhathakurta, P., Evolving
International Journal of Electronics & Communication and Evaluation of 3LP FFBP Deterministic ANN Model for
Technology, Vol. 2, Issue 3, pp. 101-105. District Level Long Range Monsoon Rainfall Prediction, J.
[99]. El-Shafie A., Jaafer O. and Akrami Seyed Ahmad, Environmental Science & Engineering, , National
2011, “Adaptive neuro-fuzzy inference system based model Environmental Engineering Research Institute, Nagpur,
for rainfall forecasting in Klang River, Malaysia”, INDIA, ISSN 0367-827, Vol 51, No. 2, pp. 137-144, April
2009. URL: http://www.neeri.res.in/jese.html.

1
 International Journal of Computer Applications (0975 – 8887)
Volume 51– No.18, August 2012

[112] Karmakar, S., Kowar, M. K., Guhathakurta, P., District, IEEE Xplore 2.0, IEEE Computer Society,
Development of an 8-Parameter Probabilistic Artificial Washington, DC, USA, ISBN 978-1-4244-2806-9, pp. 1-6,
Neural Network Model for Long-Range Monsoon Rainfall Dec. 2008.
Pattern Recognition over the Smaller Scale Geographical URL:http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?isnumber
Region –District, IEEE Computer Society, IEEE Xplore =4798312&arnumber=4798370&count=184&index=54.
2.0,, DC, USA, ISBN 978-0-7695-3267-7, pp. 569-574,
July 2008. [114]. Karmakar, S., Kowar, M. K., Guhathakurta, P., Long-
http://www2.computer.org/portal/web/csdl/doi/10.1109/ICE Range Monsoon Rainfall Pattern Recognition & Prediction
TET.2008.225. for the Subdivision ‘EPMB’ Chhattisgarh Using
http://portal.acm.org/citation.cfm?id=1445475 Deterministic & Probabilistic Neural Network, IEEE
Xplore 2.0, IEEE Computer Society, Washington, DC,
[113] Karmakar, S., Kowar, M. K., Guhathakurta, P., USA, ISBN 978-0-7695-3520-3, pp. 376-370, Feb. 2009.
Development of Artificial Neural Network Models for http://www2.computer.org/portal/web/csdl/doi/10.1109/IC
Long-Range Meteorological Parameters Pattern APR.2009.24.
Recognition over the Smaller Scale Geographical Region-

Table 1. Performance of BPN in deterministic forecast during training and independent Period
Data Training period (1951-1991) Independent/testing period (1991-2004)

Mean SD MAD SD MAD CC Mean SD MAD SD MAD CC

(% of (% of (% of (% of
mean) mean) mean) mean)

Normalized 0.62 0.03 0.02 5.51 3.21 0.6 0.62 0.03 0.02 5.50 2.96 0.8

De- 1257.8 293.6 174.7 23.3 13.8 0.6 1293.2 353.7 188.2 27.4 14.5 0.8
Normalized
(in mm)