Both observations and climate models dictate that the global warming will amplify evaporation process. It is expected that
this will be a dominant factor of decreasing water resources with the increasing world population in the near future. Thus a
realistic planning of water resources is of crucial importance.
In this study, prediction of evaporation amounts is realized using Support Vector Regression (SVR) which arises from
Support Vector Machine (SVM) and widely applied to nonlinear time series and prediction problems. SVR?s main idea is to
minimize the prediction error on the training set and also model complexity. The SVR maps the original and nonlinear input
datain to a high dimensional feature space by nonlinear mapping to yield and solve a linear regression problem in this feature
space. A regression function is generated by applying a set of high dimensional linear functions. SVR has many advantages such
as being model independent or computationally efficient. Also it guarantees to converge to optimal solution. In literature, SVR
gives excellent results in atmospheric variables when compared with conventional or modern approaches.
In application of SVR, preparation of input data does matter considerably. In this study, chaotic approach on complex time
series is used for setting data. To prepare input data, phase-space reconstruction approach is utilized. Embedding parameters,
namely embedding dimension and delay time, are extracted from the original time series. The prediction results of evaporation
time series are very promising.
Ozlem Baydaroglu is graduated from Yıldız Technical University as an environmental engineer in 2002. Then she has completed Master of Science
in environmental engineering and Master of Business Administration, respectively. Now she is lasting her PhD and working as a research assistant
in Atmospheric Sciences Department in Istanbul Technical University. Her research subjects are chaos, hydrology, renewable energy and statistics
in atmospheric sciences.
Peer Reviewed Journals
Make the best use of Scientific Research and information from our 700 + peer reviewed, Open Access Journals