American Journal of Innovative Research & Applied Sciences
Share us on Social Media Links:
| ISSN: 2429-5396 (e) | www.american-jiras.com | |
| Web Site Form: v 0.1.05 | JF 22 Cours, Wellington le Clairval, Lillebonne | France |
| ARTICLES | Am. J. innov. res. appl. sci. Volume 4, Issue 6, Pages 240-246 (June 2017)
American Journal of innovative
Research & Applied Sciences
ISSN 2429-5396 (Online)
OCLC Number: 920041286
| JUNE | VOLUME 4 | N° 6 | 2017 |
Background: One of the major problems in water resources management is the precipitation forecasting. Given the effect of precipitation on water resources, it is found that a more accurate prediction of precipitation would enable more efficient utilization of water resources and flood management. On the other hand, climate and precipitation are highly non-linear and complicated phenomena, which require non-linear mathematical modeling and simulation for accurate prediction. Objectives: The objective of this study is to predict the precipitation for one month ahead using feed forward neural network FFNN model with the Back Propagation algorithm, with the use of a combination of meteorological parameters (evaporation, air temperature, relative humidity and monthly index) from year 1976 to 2009 for Homs Meteorological Station, Syria. Methods: A feed forward neural network FFNN model was applied to predict the precipitation on a monthly basis. The models were trained based on the Levenberg-Marquardt algorithm. They investigated the effect of the number of hidden neurons and activation function on ANN modeling. The effect of climatic parameters was also examined by training three separate ANN models with three different input data sets. Results: Evaluation of the model performance using the last four years showed that there was a good agreement between the ANN estimations and the measured precipitation values. Comparison of correlation coefficient (R) and Root Mean Square Errors (RMSE) showed that the neural Network architecture (4-35-1) is of the best performance with logsigmoid activation function for the hidden layer and tansigmoid activation function for the output layer. Sensitivity analysis performed to determine the most important parameter indicated that the most important input parameter in forecasting precipitation is relative humidity, especially when it is accompanied by air temperature. Conclusions: The study reveals that FFNN model with Levenberg-Marquardt Back Propagation algorithm can be used as an appropriate forecasting tool in predicting the monthly precipitation in Homs Meteorological Station.
Keywords: Precipitation, Prediction, Artificial neural network, Back Propagation algorithm.
*Correspondant author and authors Copyright © 2017:
| Ghatfan, Abdalkareem Ammar 1 | Badia, Youcef Haidar 2 | and | Amer, Qousai Al Darwish 3 |
1. Tishreen University | department of water engineering and irrigation | Lattakia | Syria |
2. Tishreen University | department of structuralengineering | Lattakia | Syria |
3. Tishreen University | department of water engineering and irrigation | Lattakia | Syria |
This article is made freely available as part of this journal's Open Access: ID | Ghatfan-ManuscriptRef.1-ajira160517 |
AN ARTIFICIAL NEURAL NETWORK MODEL FOR MONTHLY PRECIPITATION FORECASTING INHOMS STATION, SYRIA
| Ghatfan, Abdalkareem Ammar | Badia, Youcef Haidar | and | Amer, Qousai Al Darwish |. Am. J. innov. res. appl. sci. 2017; 4(6):240-246.
| PDF FULL TEXT | |Received | 16 May 2017| |Accepted | 23 May 2017| |Published 28 May 2017 |