The least square approach requires the sum of errors at all calibration points to be minimum i.e.

	(14)

	(15)

Let be any of the parameters A,B,C,D or E; then, the derivatives can be evaluated as

	(16)

Each derivative has to be expanded algebraically and solved for the corresponding parameter. At this stage, other parameters would take on assumed values. When one cycle of calculations is completed, the procedure is repeated for the next iteration.

In the polynomial curve fitting approach the data points are fitted with fourth order polynomial functions. It has been found that the fourth- and fifth-order polynomials were of nearly equal accuracy (the maximum error being 0.27% for the velocity range 0.2-3.5 m/s of interest to the present work). With as calibration parameters, the fourth order polynomial is of the form

	(17)

Here, U is velocity and E is voltage. Typical calibration data and the order t are shown in Figure 3.33. It is also shown that the calibration curves for both wires operating at practically equal overheat ratios are very close to each other.