Thursday, March 30, 2017

Methodological Notes on Flood Risk Management Research


My PhD research topic is "Developing an integrated flood risk management framework for Vietnam" under the supervision of Dr Von Meding. I have utilised both quantitative and qualitative approaches, but mainly quantitative. In this post I will share some notes on three quantitative methodology approaches applied in my study.

1) Multi-Criteria Decision-Making (MCDM) methods

MCDM methods enable us to handle quantitative variables and help decision makers in solving flood management problems such as formulating their values and preferences, quantifying these priorities, and to apply them to decision-making processes. Many MCDM techniques are widely used in the field of flood risk management such as AHP, ANP, CP, ELECTREE, MAUT, PROMETHEE, TOPSIS, VIKOR and SAW. I applied two of these methods in my study; Technique for Order Preference by Similarity to the Ideal Solution (TOPSIS) developed by Hwang and Yoon (1981) and Analytic Hierarchy Process (AHP) proposed by Saaty (1988).

The application of AHP is still quite new in the field of disaster management and is gaining more and more attention in the field. It was used to evaluate the criteria and sub-criteria of flood risk components based on decision makers’ judgement in my study. Thankfully, the author of AHP, Professor Thomas L. Saaty, has provided a free software, Super Decisions (http://www.superdecisions.com/) to solve the complicated algorithm of the method. The result using this software looks like this:




TOPSIS is one of the most widely used MDCM methods in the field of flood risk management. I used the TOPSIS method to analyse flood risk across 63 provinces and eight regions in Vietnam using the national disaster database. The results from this evaluation tool provide additional information to support decision-making at a national scale. An example of the analysis is as follows:


2) GIS and MCDM

Geographic Information System (GIS) emerged during the 1960s through the work of Roger Tomlinson in Canada and has been widely used all over the world. GIS is considered a critical tool for spatial analysis and has been applied broadly to natural hazard risk assessment. The combination of flood risk assessment and GIS framework has been implemented in many recent studies at both local and global scale. Spatial flood risk assessments are useful tools for the identification of at-risk locations and the determination of at-risk components which are a necessary foundation to establish appropriate mitigation and adaptation measures. The integration of MCDM and GIS is applied in our study. This is an example of this application:


3) Machine learning (statistics) approach

Machine learning (ML) is the combination of classical statistics and computer science. Prediction is a major endeavour of ML, to answer questions of how to select relevant factors for prediction and how to build good predictive models. ML is divided into supervised and unsupervised learning. The difference is that the supervised learning has an outcome or target variable and the unsupervised one does not. The supervised learning includes such techniques, regression, decision tree, neural networks, k nearest neighbour, support vector machines, Bayesian learning, and random forests. The unsupervised learning includes such techniques, cluster analysis, principal component models and hidden Markov models. The applications of ML are increasingly popular in genetics, medical science, business and flood risk studies.

I applied regression models and tree-based methods of supervised ML in my study. The statistical R software is very powerful for ML implementation. The book “An Introduction to Statistical Learning” (available download at http://www-bcf.usc.edu/~gareth/ISL/) provides a simple explanation of many ML techniques. The authors of this book also provide many lectures with slides and videos (https://www.r-bloggers.com/in-depth-introduction-to-machine-learning-in-15-hours-of-expert-videos/) to illustrate all contents in the book, so it is very convenient to start learning ML (suitable for beginners like me). This is one of my applications using tree-based methods:



My PhD journey is going to finish at the end of this year. This is the most beautiful and meaningful journey in my life. When walking on this trip, I found some interesting ‘transportations’, including the three approaches that I shared in this post. I hope that it can prove useful for other researchers.