SpeedSort Algorithm

The SpeedSort algorithm is an MCP method described fully in King and Hurley (2005).

SpeedSort is unique among MCP algorithms in that it does not operate on the original scatter plot of target speed versus reference speed. Rather, it involves sorting the data points before performing the curve fit process.

SpeedSort uses a linear model of the form y = mx + b. However, the model also incorporates a 'dog leg' from that line down to the origin for speeds below a certain threshold. Windographer calculates the curve fit parameters and the wind speed threshold using the following process:

It independently sorts reference and target speeds in ascending order, then plots a scatter plot of the sorted target speeds versus the sorted reference speeds.
It chooses the threshold wind speed for the 'dog leg' as the lesser of 4 m/s or half the long-term mean reference wind speed.
It uses orthogonal least squares to fit a line to the portion of the sorted scatter plot above the threshold reference wind speed.

The graph below shows an example of a SpeedSort curve fit. The graph shows the original (unsorted) data, the sorted data, and the best-fit line. Note the slight 'dog leg' at around 2.5 m/s (reference). The angle of the 'dog leg' will be greater in cases where the line of best fit has a y-intercept value farther away from zero.

Options

If you choose multiple direction sectors, Windographer will subdivide the concurrent data by direction sector and perform the curve fit process separately for each sector. It uses the reference direction sensor to determine the direction sector.

Similarly, if you choose multiple yearly divisions, Windographer will also subdivide the concurrent data according to month. If you choose 4 yearly divisions, for example, then January-March data will go in the first subdivision, April-June in the second, July-September in the third, and October-December in the fourth.