Flood forecasting using support vector machines

Abstract

This paper describes an application of SVM over the Bird Creek catchment and addresses some important issues in developing and applying SVM in flood forecasting. It has been found that, like artificial neural network models, SVM also suffers from over-fitting and under-fitting problems and the over-fitting is more damaging than under-fitting. This paper illustrates that an optimum selection among a large number of various input combinations and parameters is a real challenge for any modellers in using SVMs. A comparison with some benchmarking models has been made, i.e. Transfer Function, Trend and Naive models. It demonstrates that SVM is able to surpass all of them in the test data series, at the expense of a huge amount of time and effort. Unlike previous published results, this paper shows that linear and nonlinear kernel functions (i.e. RBF) can yield superior performances against each other under different circumstances in the same catchment.The study also shows an interesting result in the SVM response to different rainfall inputs, where lighter rainfalls would generate very different responses to heavier ones, which is a very useful way to reveal the behaviour of a SVM model.

Streamflow Modelling: A Primer on Applications, Approaches and Challenges

“…Because of the choices made by the developers of physically oriented models, each is good at simulating different parts of the hydrological cycle. In addition, the subjective choices made in soft computing approaches can result in models that perform well at different times (Han et al, 2007). By combining predictions from different models and from different modelling approaches, it is possible to take advantage of the expertise of each of them, theoretically resulting in better overall predictive capabilities.…”

“…For example, Behzad et al (2009) reported that SVMs are able to generalize better than ANNs, though there is still some danger of under-or overfitting to the training data (Han et al, 2007) (true to some extent of virtually any model). The SVMs are also able to learn from a much smaller training set than ANNs, and the global minimum of the linear optimization is easily obtainable, whereas there is a risk of becoming trapped in a local minimum of the non-linear ANN objective function (Behzad et al, 2009).…”

“…Han et al (2007) compared the RBF and linear kernel functions and found that even within the same catchment, the ideal function can change under different circumstances. Despite being a relatively new approach to hydrological modelling, SVMs and SVRs have been applied to many of the same problems as ANNs, including rainfallrunoff modelling for water resources planning and flood forecasting at various lead-times (Asefa et al, 2006;Han et al, 2007;Behzad et al, 2009;Rasouli et al, 2012), hydraulic modelling (Liong and Sivapragasam, 2002) and downscaling of GCM output (Tripathi et al, 2006). Genetic Programming (GP; Koza, 1992) is another soft computing approach to non-linear modelling and is based on Darwin's theory of evolution by natural selection.…”

Understanding Land Use, Land Cover and Woodland-Based Ecosystem Services Change, Mabalane, Mozambique

“…Current LULC maps were created using remote sensing data (Landsat 8 from May and October, sentinel 1 from November and Advanced Land Observing Satellite 2 acquired in November 2014), and image fusion techniques; the classification was done using Support Vector Machine classifier (SVMC) using ENVI version 5.2. SVMC is a non-parametric classifier, which includes a set of related learning algorithms that are used for classification and regression (Han, Chan, & Zhu , 2007) (see appendix C). The goal of image fusion techniques is to improve the spatial resolution, geometric precision and classification accuracy.…”

“…Image fusion techniques and the classification based to Support Vector Machine classifier (SVMC) using ENVI version 5.2. SVMC is a non-parametric classifier, which includes a set of related learning algorithms that are used for classification and regression (Han et al, 2007). Kappa statistics and overall accuracy assessment was applied to compare the accuracy of classified data.…”