POST GRADUATE PROGRAM CIVIL ENGINEERING DEPT.

- ITS
FACULTY OF CIVIL ENVIRONMENTAL AND GEO ENGINEERING

MODEL TREES AS AN ALTERNATIVE TO NEURAL NETWORKS

IN RAINFALL-RUNOFF MODELLING
the 3rd Data Driven and
Lecturer : Dr.techn. Pujo Aji.,S.T.,M.T Computational Intellegence
Assignment
By Mar theana Kencanawati/Id Students Number :
03111860010003
MODEL TREES AS AN ALTERNATIVE TO NEURAL
NETWORKS IN RAINFALL RUNOFF MODELLING
DIMITRI P. SOLOMATINE..\Model trees as an alternative to neural networks in
rainfall runoff modelling.pdf
International Institute for Infrastructural Hydraulic and Environment Engineering (IHE)
KHADA N. DULAL
Department of Hydrology and Meteorology, PO BOX 406, Kathmandu, Nepal
Abstract. This paper investigates the comparative performance of two data driven modelling techniques, namely
artificial neural networks (ANNs) and model trees (MTs), in rainfall-runoff transformation. The applicability of
these techniques is studied by predicting runoff one, three and six hours ahead for European catchment. The results
shows that both ANNs and MTs produce excellent results for 1-h ahead prediction, acceptable results for 3-h ahead
prediction and conditionally acceptable results for 6-h ahead prediction. Both techniques have almost similar
performance for 1-h head prediction of run off, but the result of the ANN is slightly better than the MT for higher lead
times. However, the advantage of the MT is that the results is more understandable and allow one to build a family of
models of varying complexity and accuracy.
Keywords rainfall; runoff; artificial neural networks, M5 model tree; prediction; committee accuracy

8/26/2019 2
 INTRODUCTION

Prediction of Precipitation Runoff River Major

Problems
variables stages Hydrology

8/26/2019 3
INTRODUCTION (2)

Importants Task
Hydrologycal For
Modelling

Control Planning

Operation

Water Resources Project

8/26/2019 4
INTRODUCTION There are
basically
two
approaches
for
hydrological The theory-driven
modelling (conceptual and Conceptual
physically-based) models represent
Area of rainfall- approach, and the the general
runoff modelling internal sub
data driven
Numerous runoff (empirical and processes and
forecasting physical
techniques have black box) mechanisms of the
been suggested and approach often hydrological
used in the past.. associated by cycle.
practitioners
statistical
modelling.

8/26/2019 5
Parameter are generally Physically based models
assumed are based on are based on the

Lumped presentation of Underlying physical behavior of

understanding
the Basin the system (hydrological cycle)
Characteristic

Typically, they involve the

BATAS D.A.S solution of a system of partial
diferrential equations that
represent
MAIN

TRI RI
BUT VE
ARY R

OUTLET Flow Processes in the

Courtesy: Umboro Lasminto,2007 watershed

8/26/2019 6
EXPERIMENT (1) ………M5 MODEL TREE

8/26/2019 7
 EKSPERIMEN (2)

8/26/2019 8
 EKSPERIMEN (3)
DATA ANALYSIS OF THE SIEVE CATCHMENT FOR
PREPARATION OF INPUTS

For data driven modelling of the sieve catchment, the

casual variable is the rainfall, while the hydrological
response is the discharge at the outlet. Visualization of the
input and output shows that the maximum value of peak to
peak lags of rainfall and runoff is close to 7h.

8/26/2019 9
EXPERIMENT (5)
INPUT SET PREPARATION
The meaning of the symbols
used is as follows:
8/26/2019
The goal of a data driven
model (DDM), for example of
a regression model or ANN. Is ANN model set up
to generalize a relationship of
the form. 𝑌 (𝑚) = 𝑓(𝑥 𝑛 )
A Multi-layered perceptron
(MLP) network trained with
Software package Neural
the back-propagation algoritm
Machine(2002) and Neuro
was used because of its
Solution(2002) were used for
simplicity and capability to
the ANN Modelling.
learn the rainfall-runoff
relationship.
10
EKSPERIMEN (6)……DATA ANALYSIS

8/26/2019 11
RESULTS AND DISCUSSION (1)
Prediction of runoff Prediction of runoff Prediction of runoff
one hour ahead three hour ahead six hour ahead

8/26/2019 12
RESULTS AND
DISCUSSION
(2)
Prediction of
Qt+1, Qt+6,
Qt+3 using
ANNs and
MTs

8/26/2019 13
 CONCLUSION
For the runoff modelling experiments of the Sieve catchment using ANNs and MTs, it was
found that both techniques performed very well for runoff prediction with short lead times (1
and 3 h), while both failed to produce good results for runoff prediction with higher lead
times (6 h).

The prediction of runoff 1 h ahead is satisfactory because the input space contains the most
recent information as well as appropriately lagged information

In this study, the higher errors for the higher prediction lead times may be due to the
following factors : inadequacy in the information carrying capacity of the available data
and inability of the model to identify saturation excess runoff and infiltration excess runoff.

8/26/2019 14
8/26/2019 15