the latest version

Measure Correlate Predict Module

Correlate With Reference Data

The new Windographer MCP module correlates your data set with a nearby reference data set, and lets you extend your data set in time or fill gaps using the reference data.  Watch the video below for a demonstration.


The first tab of the MCP window lets you import two data sets. The target data set is usually the one you have measured with a met tower or remote sensing device. The reference data set can be a long-term data set from mesoscale modeling, a nearby airport or ASOS station, or just a nearby met tower. The two data sets can have different time steps and can even be offset in time. You can adjust the filter settings, the time step of comparison, and the time offset, and immediately see the effect on the scatter plots and coefficients of determination.

The second tab compares the two data sets in several graphical and tabular formats, so you can determine the degree of similarity between the two data sets, and see patterns that might suggest trimming one of the data sets or subdividing by direction.  Along with scatter plots of speed and direction, you can see graphs comparing the time series, diurnal profiles, seasonal profiles, vertical profiles, wind roses, and distributions of the two data sets.

Things get interesting on the third tab, which lists seven algorithms for correlating the wind speed data: Linear Least Squares, Orthogonal Least Squares, Variance Ratio, Weibull Fit, SpeedSort, Vertical Slice, and our own Matrix Time Series algorithm. You can use the same algorithm every time if you wish, or use the performance comparison feature to experimentally find the algorithm that works best in each situation. This optional feature can help reveal which algorithm and what settings lead to the best results in each situation. It divides the concurrent data into a training period and a test period, and then for each algorithm it analyzes the correlation within the training period, predicts target data in the test period, and then calculates error statistics by comparing the predicted target data with the true target data in the test period. The bar graphs summarize the results of the experiment, and a separate window provides details.

The fourth tab deals with wind direction data. It calculates the mean veer, which is the clockwise change in direction from reference to target, and then when predicting the target direction it simply adds that mean veer value to the reference direction. You can choose to subdivide the analysis into multiple direction sectors if you wish.

On the final tab you you choose the algorithm you wish to use to synthesize the wind speed data, and the date range you wish to synthesize. The lower portion of the screen show a comparison of the target data before and after the MCP process, so you can see the effects on the mean wind speed, mean power density, wind speed distribution, diurnal and seasonal patterns, and so on. Different speed correlation algorithms will have different effects on the final data set, and some may introduce considerable distortions, so it is often worth creating the final data set multiple times using different speed algorithms, and comparing the results. When you export the final data set to a new .windog file, Windographer flags the synthetic data points in that file, so that long after you create it you can see which data points were measured at the target site and which were estimated via the MCP process.