Advertisement

Crustal Velocity Field Modelling with Neural Network and Polynomials

  • Khosro Moghtased-Azar
  • Piroska Zaletnyik
Part of the International Association of Geodesy Symposia book series (IAG SYMPOSIA, volume 133)

Abstract

A comparison of the ability of artificial neural networks and polynomial fitting was carried out in order to model the horizontal deformation field of the Cascadia Subduction Zone, as determined from GPS analyses of the Pacific Northwest Geodetic Array (PANGA).

One set of data was used to calculate the unknown parameters of the model (training set 75 {%} of the whole data set) and the other set was used only for testing the accuracy of the derived model (testing set–25 {%} of the whole data set). The testing set has not been used to determine the parameters of the model.

The problem of overfitting (see Kecman (2001)) (i.e., the substantial oscillation of the model between the training points, the same problem than polynomial wiggle (Mathews and Fink 2004) can be avoided by restricting the flexibility of the neural model. This can be done using an independent data set, namely the validation data (one third part of the training set). The proposed method is the so-called “stopped search method”, which can be used for obtaining a smooth and precise fitting model. However, when fitting high order polynomials, it is difficult to overcome the negative effect of the overfitting problem.

Different order polynomial models and neural network models with different numbers of neurons were calculated. The best fitting polynomial model was 6th order with 28 parameters. The finally used neural network model contained 7 neurons in it’s one hidden layer, with radial basis activation functions, with 31 parameters. These two models, with same order of numbers of parameters, were compared. Calculating the remained errors at the training points the two models had the same fitting precision. However according to the testing point’s results, the neural network model offered more reliable results, with 2–3 times smaller errors.

The computations were performed with the Mathematica software, and the results are given in a symbolic form which can be used in the analysis of crustal deformation, e.g. strain analysis

Keywords

Crustal deformation Modelling Neural network Polynomial 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. Cherkassky, V., M. Yunqian (2003). Comparison of model selection for regression, Neural Computation 15, pp. 1691–1714CrossRefGoogle Scholar
  2. Dragert, H. (1987). The fall (and rise) of central Vancouver Island: 1930–1985, Canadian Journal of Earth Science, Vol. 24, pp. 689–697CrossRefGoogle Scholar
  3. Dragert, H., R.D. Hyndman, G.C. Rogers, K. Wang (1994). Current deformation and the width of the seismogenic zone of the northern Cascadia subduction thrust, Journal of Geophysical Research, Vol. 99, pp. 653–668CrossRefGoogle Scholar
  4. Dragert, H., X. Chen, J. Kouba (1995). GPS monitoring of crustal strain in southwest British Columbia with the Western Canada Deformation Array, Geomatica, Vol. 49, pp. {301–313.}Google Scholar
  5. Dubois, G. (2000). How representative are samples in sampling network? Journal of Geographic Information and Decision Analysis, Vol. 4/1, pp. 1–10Google Scholar
  6. Eren, K (1984). Strain analysis along the North Analtolian Fault by using geodetic surveys, Journal of Geodesy, Vol. 58(2)Google Scholar
  7. Feng, C.X., Z.G. Yu, A. Kusiak (2006). Selection and validation of predictive regression and neural network models based on designed experiments, IIE Transactions, Vol. 38(1), pp. {11–23.}Google Scholar
  8. Kavzoglu,. T., M.H. Saka (2005). Modelling local GPS/leveling geoid undulations using artificial neural networks, Journal of Geodesy 78, pp. 520–527CrossRefGoogle Scholar
  9. Kecman, V. (2001). Learning and Soft Computing, A Bradford Book, The MIT Press, Cambridge, Massachusetts, LondonGoogle Scholar
  10. Kleusberg, A., Y. Georgiadou, H. Dragert (1988). Establishment of crustal deformation networks using GPS, A case study, CISM Journal of ACSGC Vol. 42, pp. 341–351Google Scholar
  11. Mars, P., J.R. Chen, R. Nambiar (1996). Learning Algorithms: Theory and Applications in Signal Processing, Control and Communications, CRC Press, Boca Raton, FloridaGoogle Scholar
  12. Mathews, J.H., K.D. Fink (2004). Numerical Methods Using Matlab, Fourth Edition, Prentice-Hall Inc., Upper Saddle River, New Jersey, USA, pp. 270–273Google Scholar
  13. Meyer, T.H. (2004). The discontinuous nature of kriging interpolation for digital terrain modeling, Cartography and Geographic Information Science, Vol. (4) pp. 209–216CrossRefGoogle Scholar
  14. Miima, J.B., W. Niemeier 2004. Adapting neural networks for modelling structural behavior in geodetic deformation monitoring, ZfV, Vol. 129(3), pp. 160–167Google Scholar
  15. Miller, M., D.J. Johnson, C.M. Rubin, H. Dragert, K. Wang, A. Qamar, C. Goldfinger (2001). GPS-determination of along-strike variation in Cascadia margin kinematics: Implications for relative plate motion, subduction zone coupling, and permanent deformation, Tectonics, Vol. 20(2), pp. 161–176CrossRefGoogle Scholar
  16. Rogers, G.C. (1988). An assessment of the megathrust earthquakes potential of the Cascadia subduction zone, Canadian Journal of Earth Science, Vol. 25, pp. 844–852Google Scholar
  17. Shuhui, L., D.C. Wunsch, M.G. Giesselmann (2001). Comparative analysis of regression and artificial neural network models for wind turbine power curve estimation, Journal of Solar Energy Engineering, Vol. 123(4), pp. 327–332CrossRefGoogle Scholar
  18. Wang, K., H. Dragert, H.J. Melosh (1994). Finite element study of uplift and strain across Vancouver Island, Canadian Journal of Earth Science, Vol. 31, pp. 1510–1522Google Scholar
  19. Welsch, W., O. Heunecke (1999). Terminology and Classification of Deformation Models. 9th International FIG-Symposium on Deformation Measurements, Olsztyn, 27. -30.09.1999. Proceedings, pp. 416–429Google Scholar
  20. Zaiser, J. (1984). Ein dreidimensionales geometrisch-physikalisches Modell für konventionelle Beobachtungen-Beobachtungsfunktionale, Parameterschxätzung und Deformationsanalyse, Deutsche Geodätische Kommission, Bayerischen Akademie Wissenschaften, Reihe C, {Heft Nr. 298.}Google Scholar
  21. Zaletnyik, P., L. Völgyesi, B. Paláncz (2004). Approach of the hungarian geoid surface with sequence of neural networks, International Archives of Photogrammetry and Remote Sensing, Vol. XXXV, Part B8, pp. 119–122Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2009

Authors and Affiliations

  • Khosro Moghtased-Azar
    • 1
  • Piroska Zaletnyik
    • 2
  1. 1.Department of Geodesy and GeoinformaticsStuttgart UniversityGermany
  2. 2.Department of Geodesy and SurveyingBudapest University of Technology and EconomicsH-1521 Budapest

Personalised recommendations