Multilayer Perceptron Model for Nowcasting Visibility from Surface Observations: Results and Sensitivity to Dissimilar Station Altitudes
The reduction in the visibility during fog significantly influences surface as well as air transport operations. The prediction of fog remains difficult despite improvements in numerical weather prediction models. The present study aims at identifying a suitable neural network model with proper arch...
Saved in:
Published in | Pure and applied geophysics Vol. 172; no. 10; pp. 2813 - 2829 |
---|---|
Main Authors | , , , |
Format | Journal Article |
Language | English |
Published |
Basel
Springer Basel
01.10.2015
Springer Nature B.V |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The reduction in the visibility during fog significantly influences surface as well as air transport operations. The prediction of fog remains difficult despite improvements in numerical weather prediction models. The present study aims at identifying a suitable neural network model with proper architecture to provide precise nowcast of the horizontal visibility during fog over the airports of three significantly affected metropolises of India, namely: Kolkata (22°32′N; 88°20′E), Delhi (28°38′N; 77°12′E) and Bengaluru (12°95′N; 77°72′E). The investigation shows that the multilayer perceptron (MLP) model provides considerably less error in nowcasting the visibility during fog over the said metropolises than radial basis function network, generalized regression neural network or linear neural network. The MLP models of different architectures are trained with the data and records from 2000 to 2010. The model results are validated with observations from 2011 to 2014. Our results reveal that MLP models with different configurations (1) four input layers, three hidden layers with three hidden nodes in each layer and a single output; (2) four input layers with two hidden layers having one hidden node in the first hidden layer and two hidden nodes in the second hidden layer, and a single output layer; and (3) four input layers with two hidden layers having two hidden nodes in each hidden layer and a single output layer] provide minimum error in nowcasting the visibility during fog over the airports of Kolkata, Delhi and Bengaluru, respectively. The results show that the MLP model is well suited for nowcasting visibility during fog with 6 h lead time, however, the study reveals that the MLP model sensitive to dissimilar station altitudes in nowcasting visibility, as the minimum prediction error for the three metropolises having dissimilar mean sea level altitudes is observed through different configurations of the model. |
---|---|
Bibliography: | ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 23 |
ISSN: | 0033-4553 1420-9136 |
DOI: | 10.1007/s00024-015-1065-2 |