ANN, Decision Tree and Regression Methods for Forecasting Monthly Lagged Lake Water Level

Hi guys,
Here I am sharing the abstract of my presentation (Paper) in EWRA 2015 which be held June 10-13, 2015 in Istanbul. I will share more details as soon as it is published in the abstract book.

ANN, Decision Tree and Regression Methods for Forecasting Monthly Lagged Lake Water Level 

Babak Vaheddoost(1), Hafzullah Aksoy(2), Hirad Abghari(3), Saieed Zare Naghadeh(4)

(1) Department of Civil Engineering, Istanbul Technical University, Turkey, e-mail: vaheddoostb@itu.edu.tr
(2) Department of Civil Engineering, Istanbul Technical University, Turkey, e-mail: haksoy@itu.edu.tr
(3) Faculty of Natural Resources, Urmia University, Iran, e-mail: h.abghari@urmia.ac.ir
(4) Department of Civil Engineering, Dokuz Eylul University, Turkey, e-mail: saieed_zare@yahoo.com

Abstract:

The forecasting of lake fluctuations with previously observed data is a necessary task in closed basin lake hydrologic analysis. Known as a hyper-saline lake, Lake Urmia is dealing with atrophy and mismanagement through recent years. In this investigation, nine methods in the category of parametric and nonparametric approaches were applied for modelling Lake water level. Eleven months in the form of monthly lagged data were used as independent variables and used arbitrarily for each tested model. The present situation of the lake was considered as the dependent variable. The parametric approaches, used in modelling were multi linear regression (MLR), nonlinear regression (NLR) and decision tree (DT). Nonparametric approaches applied in the modelling were artificial neural networks (ANN) with different functions. Generalized regression neural network (GRNN), radial base function (RBF) and feed forward back propagation (FFBP) were used as ANN models. Three criteria of coefficient of determination, Lin’s concordance correlation coefficient and root mean square percentage error were used in comparison of the results. Each model was given a grade through each criteria of the measurement in the magnitude of zero to ten for comparison. The summation of the grades for each method was accepted as the prosperity of each method. Results show that the NLR is the superior method of all, while GRNN shows the worst results.

Key words: Decision tree, ANN, lake water elevation, hyper-saline lake, Lake Urmia

Bbak VDST