Comparison of artificial neural networks and empirical equations to estimate daily pan evaporation

This study consists of two parts. In the first part, daily pan evaporation estimations are achieved by a suitable artificial neural network (ANN) model for the meteorological data recorded from the automated GroWheather meteorological station near Lake Egirdir, which lies in the Lake District of western Turkey. At this station six meteorological variables are measured simultaneously, namely, air temperature, water temperature, solar radiation, air pressure, wind speed and relative humidity. The ANN architecture has only one output neuron with up to four input neurons representing air and water temperatures, air pressure and solar radiation. Prior to ANN model construction the classical correlation study indicated the insignificance of wind speed and relative humidity in the Egirdir Lake area. Hence, the final ANN model has three input neurons in the input layer with one at the output layer. The hidden layer neuron number is found to be six after various trial and error model runs. In the second part, daily evaporation values are estimated using classical approaches such as the Priestley–Taylor, Brutsaert–Stricker, Makkink and Hamon methods. The comparison was first made using the original constant values involved in each equation, and then using the calibrated constant values. The results show that when the original constant values were used, the Priestley–Taylor, Brutsaert–Stricker and Makkink methods underestimated evaporation values, but the Hamon method overestimated them. When calibrated constant values were substituted for the original constant values, all four equations improved to estimate evaporation. While the mean square error (MSE) values varied between 6.27 and 49.2 for original constant values, they varied between 3.43 and 4.33 for calibrated constant values. Of the evaporation methods, the Hamon method improved well to estimate evaporation values. It is also noted that the ANN model is superior even to the classical approaches of the Priestley–Taylor, Brutsaert–Stricker, Makkink and Hamon methods. Copyright © 2008 John Wiley & Sons, Ltd. Cette etude se compose de deux parties. Dans la premiere partie, des evaluations quotidiennes d'evaporation sont realisees par un modele approprie de reseau neuronal (ANN) pour les donnees meteorologiques enregistrees a partir de la station meteorologique automatisee de GroWheather pres du lac Egirdir qui se situe dans la Region des lacs de la Turquie occidentale. A cette station six variables meteorologiques sont mesurees simultanement, a savoir, temperature de l'air, temperature de l'eau, rayonnement solaire, pression atmospherique, vitesse de vent et humidite relative. L'architecture de l'ANN a seulement un neurone en sortie avec jusqu'a quatre neurones en entree representant les temperatures de l'eau et de l'air, la pression atmospherique et la radiation solaire. Avant la construction du modele d'ANN l'etude classique de correlation a indique la non-signification de la vitesse du vent et de l'humidite relative dans la region du lac Egirdir. Par consequent, le modele final d'ANN a trois neurones en entree dans la couche d'entree et un dans la couche de sortie. Le nombre de neurones caches a ete finalement de six apres plusieurs passages du modele. Dans la deuxieme partie, les valeurs quotidiennes d'evaporation sont estimees en utilisant des approches classiques comme les methodes de Priestley–Taylor, de Brutsaert–Stricker, de Makkink et de Hamon. La comparaison a d'abord ete faite en utilisant les valeurs d'origine des constantes impliquees dans chaque equation, et puis en utilisant les valeurs calees des constantes. Les resultats prouvent que quand les valeurs d'origine ont ete employees, les methodes de Priestley–Taylor, de Brutsaert–Stricker et de Makkink ont sous-estime les valeurs d'evaporation, mais les methodes de Hamon les ont surestimees. Quand les valeurs calees ont ete substituees aux valeurs d'origine des constantes, pour les quatre equations pour les estimations de l'evaporation ont ete ameliorees. Tandis que les valeurs de l'erreur moyenne quadratique varient de 6.27 a 49.2 pour les valeurs d'origine des constantes, elles ont varie de 3.43 a 4.33 pour les valeurs calees par le modele. Parmi les methodes d'evaporation, la methode de Hamon s'est bien amelioree pour estimer des valeurs d'evaporation. On note egalement que le modele d'ANN est superieur meme aux approches classiques Priestley–Taylor, aux methodes de Brutsaert–Stricker, de Makkink et de Hamon. Copyright © 2008 John Wiley & Sons, Ltd.

[1]  Mohamed Mohandes,et al.  USE OF RADIAL BASIS FUNCTIONS FOR ESTIMATING MONTHLY MEAN DAILY SOLAR RADIATION , 2000 .

[2]  Karen Warnaka,et al.  Analyses of equations for free water evaporation estimates , 1988 .

[3]  Özlem Terzi,et al.  Artificial Neural Network Models of Daily Pan Evaporation , 2006 .

[4]  Harvey E. Jobson,et al.  Comparison of techniques for estimating annual lake evaporation using climatological data , 1982 .

[5]  F. I. Morton Climatological estimates of lake evaporation , 1979 .

[6]  Vijay P. Singh,et al.  EVALUATION AND GENERALIZATION OF 13 MASS‐TRANSFER EQUATIONS FOR DETERMINING FREE WATER EVAPORATION , 1997 .

[7]  Ulrich Anders,et al.  Model selection in neural networks , 1999, Neural Networks.

[8]  Louis Sanzogni,et al.  Feed-Forward Artificial Neural Network Model For Forecasting Rainfall Run-Off , 1998 .

[9]  Robert B. Stewart,et al.  A simple method for determining the evaporation from shallow lakes and ponds , 1976 .

[10]  R. S McKenzie,et al.  Evaluation of river losses from the Orange River using hydraulic modelling , 2001 .

[11]  Vijay P. Singh,et al.  Evaluation and generalization of temperature‐based methods for calculating evaporation , 2001 .

[12]  K. P. Sudheer,et al.  Modelling evaporation using an artificial neural network algorithm , 2002 .

[13]  Wossenu Abtew,et al.  Evaporation Estimation for Lake Okeechobee in South Florida , 2001 .

[14]  B. Choudhury,et al.  Evaluation of an empirical equation for annual evaporation using field observations and results from a biophysical model , 1999 .

[15]  Y. Travi,et al.  Lake evaporation estimates in tropical Africa (Lake Ziway, Ethiopia) , 2001 .

[16]  T. O. Halawani,et al.  A neural networks approach for wind speed prediction , 1998 .

[17]  V. Singh,et al.  Sensitivity of mass transfer‐based evaporation equations to errors in daily and monthly input data , 1997 .

[18]  Ashish Sharma,et al.  A study of optimal model lag and spatial inputs to artificial neural network for rainfall forecasting , 2000 .

[19]  Dimitri P. Solomatine,et al.  River flow forecasting using artificial neural networks , 2001 .

[20]  Shuichi Ikebuchi,et al.  Evaporation from Lake Biwa , 1988 .

[21]  Özlem Terzi,et al.  Estimating Evaporation Using ANFIS , 2006 .

[22]  de Bruin,et al.  A Simple Model for Shallow Lake Evaporation , 1978 .

[23]  T. Sathish,et al.  River Flow Forecasting using Recurrent Neural Networks , 2004 .

[24]  Joseph A. Jervase,et al.  Solar radiation estimation using artificial neural networks , 2002 .

[25]  Contour maps for sunshine ratio for Oman using radial basis function generated data , 2003 .

[26]  Shafiqur Rehman,et al.  Application of neural networks for the prediction of hourly mean surface temperatures in Saudi Arabia , 2002 .

[27]  M. Roderick,et al.  A simple pan‐evaporation model for analysis of climate simulations: Evaluation over Australia , 2006 .

[28]  Sevket Durucan,et al.  River flow prediction using artificial neural networks: generalisation beyond the calibration range. , 2000 .

[29]  Slobodan P. Simonovic,et al.  Short term streamflow forecasting using artificial neural networks , 1998 .