This paper formulates and tests a mathematical model for the prediction of historical average daily maximum and minimum temperatures. First, the model is derived based on a physical understanding of the problem. Then some available weather data are analyzed by three traditional regression methods, namely, a nonlinear regression method, a linear regression method, and a Fourier transform method. Using the coefficients obtained by these analyses, temperatures are predicted for an independent test set and the root‐mean‐square errors (rms) determined. In an alternative approach, an artificial neural network (ANN) is trained using the same weather data. The trained ANN is made to predict the temperatures for the test set and its rms determined. The variation of rms values with the change in ANN structure is also observed. A comparative study shows that the ANN is a general tool for data analysis and capable of producing results comparable in accuracy to those obtained by the more conventional methods.

This content is only available via PDF.
You do not currently have access to this content.