Rainfall Estimation with Artificial Neural Network Based on Non-Rainfall Weather Data in Shiraz, Mashhad and Kerman Regions
Rainfall due to its noise and random nature has structural changes at different times. Because of large uncertainty, fluctuations in the amount of rainfall forecast is created the prediction of which has been difficult. In this article, precipitation predictability was carried out rescaled by range...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | fas |
| Published: |
University of Tabriz
2013-05-01
|
| Series: | نشریه جغرافیا و برنامهریزی |
| Subjects: | |
| Online Access: | https://geoplanning.tabrizu.ac.ir/article_4_dd9efeabe9156c5a53838a63be8c69ca.pdf |
| Summary: | Rainfall due to its noise and random nature has structural changes at different times. Because of large uncertainty, fluctuations in the amount of rainfall forecast is created the prediction of which has been difficult. In this article, precipitation predictability was carried out rescaled by range analysis (R/S) technique in Shiraz, Mashhad and Kerman regions. SnapshotHurst (H) showed that rainfall parameter has the ability of predictability, because H was higher than 0.5 and much closer to the value 1. Minimum Hurst value was 0.8 in Mashhad and maximum Hurst value was 0.92 in Shiraz. In order to predict rainfall we used artificial neural network. Type of input parameters based on Pearson correlation test between data from non-rainfall, were a combination of temperature and humidity data. Number of input parameters, the number of middle layers, and other information related to artificial neural network randomly were selected. As a whole, rainfall estimation was calculated through Peresptron multi-layer neural network for comparing the performance of neural network. Results showed that the use of 3 and 4 meteorological parameters has the best rank estimator. Proposed layouts for the Shiraz station is 1-21-21-3, for Kerman 1-25-25-3 and for Mashhad 1-19-19-4 in which 1-25-25-3 of have correlation coefficients more than 91 percent. Validation rainfall models showed that network designed for rainfall parameters has best performance rainfall in Mashhad, Shiraz and Kerman stations with 4, 11 and 14 percent error respectively. As a whole, results showed that neural network method with considering the temperaturel and humidity data for describing the process and their combination in predicting good results were offered. |
|---|---|
| ISSN: | 2008-8078 2717-3534 |