Paper 1 Summary

Machine Learning Techniques For
Rainfall Prediction: A Review
2017
This paper presented review of different methods used for rainfall

prediction and problems one might encounter while applying
different approaches for rainfall forecasting
Abstract
Heavy rainfall prediction is a major problem for meteorological department as it is closely
associated with the economy and life of human. It is a cause for natural disasters like flood
and drought which are encountered by people across the globe every year. Accuracy of
rainfall forecasting has great importance for countries like India whose economy is largely
dependent on agriculture. Due to dynamic nature of atmosphere, Statistical techniques fail
to provide good accuracy for rainfall forecasting. Nonlinearity of rainfall data makes
Artificial Neural Network a better technique. Review work and comparison of different
approaches and algorithms used by researchers for rainfall prediction is shown in a tabular
form. Intention of this paper is to give non-experts easy access to the techniques and
approaches used in the field of rainfall prediction.
Scholarcy Synopsis
The accuracy of rainfall forecasting is crucial for countries like
India, and Artificial Neural Network is a better technique due to the
nonlinearity of rainfall data.
Rainfall prediction is important for avoiding floods and managing water resources.
Different methods for rainfall prediction are discussed, including Autocorrelation Function
and the Partial Autocorrelation Function Back-Propagation Neural Network, Support Vector
Machines, and Self Organizing Map.
The paper also mentions the challenges of rainfall prediction and the advantages of using
Artificial Neural Network.
Scholarcy Highlights
 Rainfall prediction is helpful to avoid flood which save lives and properties of humans
 Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF) are
important analytical tools used with the time series analysis and forecasting [30]
 1) Back-Propagation Neural Network: BPNN is made of multilayer feed-forward (MLFF)
neural network which contains one input layer, hidden layers and one output layer
 Support Vector Machines were created by Vapnik and his co-workers which has been
used for supervised learning due to – (i) Better generalization ability than other NN
models (ii) SVM solution is identical, optimal and absent from local minima (iii)
Applicable to non-vectorial data (Strings and Graphs) and (iv) Very few parameters are
needed for tuning the learning m/c
 The estimation of rainfall is of great importance in terms of water resources
management, human life and their environment
 This paper presented review of different methods used for rainfall prediction and
problems one might encounter while applying different approaches for rainfall
forecasting
Scholarcy Summary
INTRODUCTION
Rainfall prediction is helpful to avoid flood which save lives and properties of humans.
It helps in managing resources of water.
Fluctuation in rainfall timing and its quantity makes rainfall prediction a challenging task
for meteorological scientists.
In all the services provided by meteorological department, Weather forecasting stands out
on top for all the countries across the globe.
Various neural networks algorithm which are used for prediction are discussed with their
steps in detail.
BACKGROUND
Two widely used methods for rainfall forecasting are: Statistical methods and Numerical
Weather Prediction (NWP) model [16].
ARMA model is only useful for stationary time-series data and forecasting of short term
rainfall.
Autocorrelation Function (ACF) and the Partial Autocorrelation Function (PACF) are
important analytical tools used with the time series analysis and forecasting [30].
Main uses of these models are to measure the statistical relationships between observations
in a single data series.
In order to identify the model, ACF and PACF have to be estimated.
They are used to guess the form of the model and to obtain approximate estimates of the
parameters as well [30]
Artificial Neural Network

ANN is a computational model that is inspired by the human brain [31]. ANN contains a big
number of interconnected neurons, which mostly operate in parallel, and are well
structured.
A single-layer feed forward (SLFF) neural network consists one input layer whose nodes
have weights assigned and one output layer.
A multilayer feed-forward (MLFF) neural network architecture can be developed by adding

hidden layers in SLFF neural network.
1) Back-Propagation Neural Network: BPNN is made of MLFF neural network which

contains one input layer, hidden layers and one output layer.
During the training process, this output may fit to the data very well, but it may provide
poor results during the testing process
This suggests that the neural network may not generalize well.
Alternate option to make the neural network generalize enough is by doing small changes in
the number of layers and neurons in the inputs, without changing the output components.
Solution is to keep the architecture of neural network relatively simple and small [34],
because complex architectures are much more prone to overfitting [35]
Support Vector Machine (SVM)

Support Vector Machine is one type of feed forward network.
Support vector machines are applicable for tasks like pattern classification, nonlinear
regression etc.
Support Vector Machines were created by Vapnik and his co-workers which has been used
for supervised learning due to – (i) Better generalization ability than other NN models (ii)
SVM solution is identical, optimal and absent from local minima (iii) Applicable to non-
vectorial data (Strings and Graphs) and (iv) Very few parameters are needed for tuning the
learning m/c.
Very few scientists have applied this technique for rainfall prediction and results were
acceptable
Self Organizing Map (SOM)
SOM is a part of artificial neural network.
Neurons in the network compete between themselves to get activated.
One neuron only can get activated at one time.
This scheme is called as “winner takes all” scheme.
Base of SOM is unsupervised learning, in which human interpretation is not required during
the learning process.
Structure of neurons is in one or two dimensional lattice.
Self Organizing Map is data visualization technique invented by Prof.
Teuvo Kohonen which reduces the dimensions of data through self-organizing neural
networks [37].
SOM reduces dimensions & display similarities ([37], [38])
LITERATURE REVIEW
A.Sahai et al [3] used error back propagation algorithm for Summer Monsoon Rainfall
prediction of India on monthly and seasonal time series
They used data of previous five years of monthly and seasonal mean rainfall values for
rainfall prediction.
They used past four months rainfall data as inputs to neural network model.
Khalifa [13] used yearly, biannually, quarterly and monthly rainfall data for rainfall
prediction
They trained four different Focused Time Delay Neural Networks (FTDNN) for rainfall
forecasting.
Ghosh [14] contributed towards developing K- mean clustering technique combined with
decision tree algorithm, CART, is used for rainfall states generation from large scale
atmospheric variables in a river basin.
Min-Max temperature, relative humidity, pan evaporation Rainfall, climatic indices,

atmospheric temperature, solar data
CONCLUSION
The estimation of rainfall is of great importance in terms of water resources management,
human life and their environment.
It can be met with the incorrect or incomplete estimation problems because rainfall
estimation is affected from the geographical and regional changes and properties.
This paper presented review of different methods used for rainfall prediction and problems
one might encounter while applying different approaches for rainfall forecasting.
Due to nonlinear relationships in rainfall data and ability of learning from the past makes
Artificial Neural Network a preferable approach from all available approaches
Contributions
The estimation of rainfall is of great importance in terms of water resources management,
human life and their environment. It can be met with the incorrect or incomplete estimation
problems because rainfall estimation is affected from the geographical and regional changes
and properties. This paper presented review of different methods used for rainfall
prediction and problems one might encounter while applying different approaches for
rainfall forecasting. Due to nonlinear relationships in rainfall data and ability of learning
from the past makes Artificial Neural Network a preferable approach from all available
approaches.

Paper 1 Summary

Uploaded by

Copyright:

Available Formats

Paper 1 Summary

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Paper 1 Summary

Uploaded by

Copyright:

Available Formats

Machine Learning Techniques For

Rainfall Prediction: A Review

This paper presented review of different methods used for rainfall

It helps in managing resources of water.

In order to identify the model, ACF and PACF have to be estimated.

Artificial Neural Network

A multilayer feed-forward (MLFF) neural network architecture can be developed by adding

1) Back-Propagation Neural Network: BPNN is made of MLFF neural network which

Support Vector Machine (SVM)

Neurons in the network compete between themselves to get activated.

One neuron only can get activated at one time.

This scheme is called as “winner takes all” scheme.

Structure of neurons is in one or two dimensional lattice.

Self Organizing Map is data visualization technique invented by Prof.

SOM reduces dimensions & display similarities ([37], [38])

Min-Max temperature, relative humidity, pan evaporation Rainfall, climatic indices,

You might also like