The variables considered in the modeling are the wind speed of the cyclone, the height of the water rise, the distance from the path of the cyclone, the population, the amount of livestock, the number of permanent and temporary houses and the length of paved and unpaved roads. It has been observed that the height of the water rise has the most damaging effect on the victims. The neural network technique has been further extended to analyze the effect of noise in data using 'Monte-Carlo Simulation'.
CHAPTER-l
- GENERAL
- BACKGROUND OF THE STUDY
- POTENTIAL OF THE SYSTEM
- OBJECTIVES OF THE STUDY
- ORGANIZA TION OF THE THESIS
- SUMMARY
The main objective of the research work is to develop a forecasting system to predict the damage caused by cyclones. To investigate the characteristics of the damage caused by the cyclone, i.e. the number and types of casualties caused by the cyclone and their relationship with the parameters of the cyclone, such as wind speed, height of the water wave and distance from the center, etc. b. The assessment can be made through the development of prediction models, which is the main goal of the research work presented in this thesis.
CHAPTER- 2
LITERA TURE REVIEW
- INTRODUCTION
- STUDIES REGARDING DAMAGES CAUSED BY CYCLONE IN BANGLADESH
- STUDIES RELATED TO PREDICTION OF DAMAGES CAUSED BY NATURAL DISASTER
- OVERVIEW OF THE METHODS OF ANALYZING DAMAGES CAUSED BY NATURAL DISASTER
- SUMMARY
The study is very qualitative and cannot assess the relief and rehabilitation requirements in relation to the condition of the infrastructure and the strength of the cyclone. Here the dependent variable is related to the independent (explanatory) variables using theories and experimental results of the interaction between them. The 'knowledge' that the network has going from training (calibration) is stored in the form of connection weights between nodes.
STUDY DESIGN AND METHODOLOGY
- INTRODUCTION
- OVERVIEW OF NEURAL NETWORK (NN) METHOD
- THEORY AND STRUCTURE OF NEURAL NETWORK AND BACK - PROPAGATION
- SUMMARY
The back-propagation algorithm mainly includes the two phases, i.e., the training phase and the prediction phase of operation, where the mean square error between the actual output of a multi-layer feed-forward perception and the desired output is minimized. The sigmoidal function is used as the transfer function of the nodes in the hidden layer and the output layer. If i is the index of a neuron, ei denotes the weighted sum of the pulses it receives:.
Most of the networks used in forecasting applications are organized in layers and hence known as layered networks, in Figure-3.2. Information transfer occurs by propagation from the input layer to the output layer. This means that the neurons in layer C receive data from layer C-I and transmit their pulses to layer C+ I. Sigmoidal function is used as transfer function of the nodes in the hidden layer and output layer.
The term backpropagation is used because part of the error gradient for previous layers is used when updating the connections in a layer. The error formula is manifested only in the connections of the penultimate layer. The general methodology of the neural network, its structure and back-propagation method have been explained in this chapter.
Using the method described in this chapter, the neural network model will be trained with a set of data on the casualties caused by cyclone.
DATA COLLECTION AND ANALYSIS
INTRODUCTION
BACKGROUND OF COLLECTION OF DATA
Bangladesh Meteorologicill Department, observe the cyclone position, movement or path, wind speed and date of cyclone. Data on human death victims, damaged livestock, damaged house and damaged roads are collected including some cyclone disaster photos.
TYPES AND CHARACTERISTICS OF DATA
In Noakhali, the total population loss was 8878, the total damaged livestock was 6697, the total total damaged house was 173880, the total partially damaged house was 160450, the total damaged paved road was 10 km, the total damaged unpaved road was 60 km and average wind speed was 150 km/h, gust height 3 m, distance 81 km. In Bhola, the total population loss was 221, the total damaged livestock was 14593, the total total damaged house was 47218, the total partially damaged house was 105001, the total damaged paved road was 36 km, the total damaged unpaved road was 292 km and average wind speed was 178 km/h, gust height 4 m, distance 84 km. In Lakshmipur, total population loss was 10785, total livestock damaged was 556, total houses completely damaged were 10785, total houses partially damaged were 25665, total asphalted road damaged was 33 km, total unasphalted road damaged was 210 km and the average value of the wind speed was 140 kmlh, the gust height was 4 m, the distance was 112 km.
In Barguna, the total population loss was 7, the total number of damaged livestock was 2640, the total number of fully damaged houses was 10913, the total partially damaged house was 44577, the total damaged paved road was 16 km, the total damaged unpaved road was 95 km and the average value of the wind speed was 195 kmlh, wave height was 3m, distance was. In Feni, total population loss was 5, total damaged livestock was 2233, total fully damaged house was 23602, total partially damaged house was 37173, total. In Jhalokathi the total loss of population was I, the total damaged livestock was 2505, the total fully damaged house was 4104, the total partially damaged house was 14620, the total damaged paved road was 25 km, the total damaged unpaved road was 152 km and the average value of the wind speed was 150 km/h, the wave height was 3 m, the distance was 138 km.
In Pirojpur, the total population loss was I, the total damaged livestock was 8, the total completely damaged house was 417, the total partially damaged house was 500, the total damaged paved road was 29 km, the total damaged dirt road was 118 km and the average value of wind speed was 148 km/h, peak height was 3 m, distance was. In Chandpur, the total population loss was I, the total damaged livestock was 165,029, the total completely damaged house was 532, the total partially damaged house was 2063, the total damaged paved road was 10 km, the total damaged dirt road was 73 km and the average value of wind speed was 100 km/h, peak height was 2 meters, distance was. In Rangamati, the total population loss was 12, the total damaged livestock was 389, the total completely damaged house was 15998, the total partially damaged house was 20197, the total damaged paved road was 27 km, the total damaged dirt road was 68 km and the average value of wind speed was 235 km/h, peak height was Om, distance was 118 km.
In Barisal total loss of population was 0, total number of livestock damaged was 98, total fully damaged house was 1944, total partially damaged house was 10958, total damaged paved road was 33 km, total damaged unpaved road was IS km and the average value of the wind speed was 160 kmIh, the wave height was I m, the distance was.
4'" SUMMARY
CHAPTERS
DEVELOPMENT AND CALIBRA nON OF THE NEURAL NETWORK MODEL
- INTRODUCTION
- DEVELOPMENT OF THE MODEL
- DEFINITION OF THE VARIABLES
- DESIGN OF NEURAL NETWORK
- RESULT OF CALIBRA nON
- Calibration Result of Livestock Casualty Model
- Calibration Result of House Damage Model
- Calibration Result of Road Damage Model
- RESULT OF VALIDA TION TEST
- MODELS INCORPORATING ACCESSIBILITY
- Calibration Result of Human Casualty Model
- Calibration Result of Livestock Casualty Model
- Result of Validation Test
- SUMMARY
At the initial programming stage, all variables were included in a single model. Input variables related to cyclone strength were included in all models. The following section explains each of these steps in detail. i) Choice of variables in models.
In this case, it was found that the training could be done successfully. ii) Selection of the number of hidden layers. The criteria for choosing the number of nodes in the hidden layer is stated as "the number of nodes that minimizes the total error at the optimum with the minimum (or reasonable) number of iterations". It was found that the results of the developed models are very consistent with the observed values.
It is observed that the output of the models agrees quite satisfactorily with the observed data. The observed and predicted values of the output variables are shov,ll in Figures 5.60 and 5.6b. This chapter described the development of the neural network model for cyclone loss and damage analysis.
Validation tests prove that the explanatory power of the developed models is also very good.
MODEL APPLICATION AND RESULT
- INTRODUCTION
- EFFECT OF CYCLONE PARAMETERS
- ANALYSIS OF NOISE IN DATA
- SUMMARY
Figures 6.la - 6.lf show the effects of changing wind speed on all 6 types of accidents. Figures 6.2a - 6.2f show the effects of the change in water surge height on different types of accidents. Here, it is assumed that all variables, except the height of the water wave, are constant.
Considering all these effects, it is observed that water rise is the most prominent destructive element among the three variables in question. But it is recognized that these data may be incorrect due to errors in measurement, errors in reporting and the tendency of various organizations to over- and under-estimate the numbers of casualties. To analyze the effect of errors in the data, 'Monte Carlo Simulation' approach has been used.
For the analysis of the effects of errors in the data, it is assumed that the reported data are the mean of the input variables. The coefficient of variance, when expressed as a percentage, indicates the percentage of error in the data. It has been found that the height of water rise is the most destructive element among the variables considered.
This means that the models can predict quite accurately as long as the error in the input data remains within a tolerable limit (in this case 20%).
DISTANCE, km
CHAPTER-7
CONCLUSION AND FUTURE RESEARCH DIRECTION
CONCLUSION
The 'Neural Network' modeling technique was found to be the most appropriate for predicting damage caused by cyclones. The variables included in the input side of the model were the characteristics of the cyclone defined by wind speed, water rise height and distance from the path of the cyclone, as well as the socio-economic variables which include population, livestock, number of houses and the amount of paved and unpaved roads. Data had been collected from water sources such as Cyclone Preparedness Program of Bangladesh Red Crescent Society, Bangladesh Meteorological Department and CARE, Bangladesh.
Using the data and technique described above, the neural network models were calibrated very successfully. The explanatory power of the models was confirmed by regression of the results against the observed values. We also examined the effects of input noise on model results using "Monte Carlo simulation".
It was observed that the output of the models remains quite stable for errors up to twenty percent.
LIMITATIONS OF THE MODELS
FUTURE RESEARCH DIRECTION
Roads and Highways Department, Government of Bangladesh.. 1995), Integrated Planning Information Systems Disaster Planning Analysis, Journal of Urban Planning and Development, Vo1.l2l, No.1.
Photo of Damage House Casual,tt
Pboto of Damages of Cvclone 1991
Pboto of Damages of Cyclone 1991
Weight Distribution
Count
