NCL Website header
NCL Home > Documentation > Functions > Interpolation, Ngmath routines

ftcurvs

Calculates a smoothing spline.

Prototype

	function ftcurvs (
		xi     : numeric,  
		yi     : numeric,  
		d  [*] : numeric,  
		xo [*] : numeric   
	)

	return_val  :  float or double

Arguments

xi

An array containing the abscissae for the input function, with rightmost dimension npts. If xi is multi-dimensional, it must have the same dimension sizes as yi.

yi

An array of any dimensionality, whose rightmost dimension is npts, containing the functional values of the input function. That is, yi(...,k) is the functional value at xi(...,k) for k=0,npts-1.

d

A user-specified value containing the observed weights; it may either be a 1D array (of length npts) or a scalar.

xo

A 1D array of length nxo containing the abscissae for the interpolated values.

Return value

An array of the same dimensionality as yi, but with the rightmost dimension replaced by nxo, containing the interpolated functional values of the smoothing spline at the points specified by xo.

The output array will be of type double if any of the input is double, and float otherwise.

Description

ftcurvs is in the Fitgrid package - a package containing 1D and 2D interpolators using cubic splines under tension.

There are some parameters that can alter the behavior of ftcurvs. These parameters all have reasonable default values. However, users may change any of these parameters by invoking ftsetp prior to calling ftcurvs. ftcurvs is called after all of the desired values for control parameters have been set.

Control parameters that apply to ftcurvs are: sig, smt, eps, sf2.

Two parameters and one function argument used to control the degree of smoothness - the parameters are smt and eps, and the function argument is d.

The argument d is a value indicating the degree of confidence in the accuracy of the input function values - it should be an approximation of the standard deviation of error. Effectively the value of d controls how close the smoothed curve comes to the input data points. If d is small, then the interpolated curve will pass close to the input data. The larger the value of d, the more freedom the smooth curve has in how close it comes to the input data values.

The parameter smt is a more subtle global smoothing parameter; smt must be non-negative. For small values of smt, the curve approximates the tension spline and for larger values of smt, the curve is smoother. A reasonable value for smt is npts.

The parameter eps controls the precision to which smt is interpreted; eps must be between 0. and 1. inclusive. A reasonable value for eps is sqrt(2./npts).

The value for the parameter sig specifies the tension factor. Values near zero result in a cubic spline; large values (e.g. 50) result in nearly a polygonal line. A typical value is 1. (the default).

You can extrapolate values with ftcurvs (that is, calculate interpolated values for abscissae outside of the domain of the input), but these values are, in general, unreliable.

Examples

begin
  xi = (/  0.00,   2.00,   5.00,   8.00,  10.00,  13.00,     \
          15.00,  18.00,  21.00,  23.00,  30.00         /)
  yi = (/  1.00,   0.81,   0.00,  -0.81,  -1.00,  -0.84,     \
          -0.56,   0.04,   0.73,   1.18,   2.0          /)

  npts = 201
  xo   = fspan(0.,30.,npts)

  yo = ftcurvs(xi, yi, 0.1, xo)
end

Errors

= 1 if the number of input points is less than 2.
= 4 if the xi values are not strictly increasing.
= 5 if a d-value is non-positive.