What is nonparametric wind regression (NWR)? NWR is a pollutant source apportionment model that can be used to identify and quantify the possible source regions of pollutants as defined by wind direction sectors and wind speed bins. NWR quantifies the expected average concentration of a pollutant as a function of wind direction and wind speed using nonparametric regression and kernel smoothing methods, or a type of weighted average. The average concentration of a pollutant for a particular wind direction and speed bin is calculated as the weighted average of the concentration in a window around the bin, so that concentrations near the bin (i.e., similar wind speed and direction) have a greater influence than concentrations farther away (i.e., much different wind speed and direction) on the expected average concentration. Higher wind speed and direction (user-defined) smoothing constants lead to a larger smoothing window, which incorporate more pollutant data in the kernel smoothing process.

What inputs and settings are required for NWR? with both hourly air pollutant and wind speed and wind direction that are displayed on the map is required. Upon a site selection, available parameter, POC(s), and time information (year(s), month(s), day(s) of week, and hour(s)) are displayed below. POC and time information will update upon parameter selection depending on the data availability. Optional inputs include the removal of calm winds < 1 mph, date(s) exclusion to remove known high-pollutant events, and maximum limits on wind speed and parameter concentration. The size of the wind speed and direction smoothing constants are automatically set to appropriate values for one full year of hourly data, but can be modified. Higher smoothing constants may be needed for shorter time periods to incorporate more pollutant data in the smoothing process. Lower smoothing constants may be useful when analyzing multiple years of data to help distinguish more features in subsequent plots.

What are the results of NWR? NWR results are displayed in different types of plots. Polar Plots display the results of the NWR analysis as a continuous surface of average pollutant concentration by wind speed and direction in polar coordinates, where the wind direction represents winds heading toward the center point (location of the monitoring site) and wind speeds increase away from the center point. The colors of the surface represent the NWR-estimated average pollutant concentration for each combination of wind speed and direction as shown by the legend. Cartesian Plots displays the same NWR results in cartesian (x and y) coordinates with wind direction on the x-axis and wind speed on the y-axis. The Polar and Cartesian plots share the same data and legend. Both plots also display contour lines of the signal to noise ratios (SNR) calculated as the mean divided by the standard deviation SNR contours equal to 2 and 3 are shown in red and white, respectively. Results outside the contours (SNR < 2) indicate that there were too few data points for a reliable estimate.

Two additional plots use the NWR analysis results to apportion the expected average pollutant concentrations to wind direction sectors, which are ranges of wind directions. For these plots, the width of the wind direction sectors is defined by the wind direction smoothing constant. The Sector Apportionment Density Plot displays the sector apportionment density curve (blue line) calculated as the NWR-estimated average pollutant concentration for each wind direction sector centered at each wind direction degree (1-360°) and integrated over all wind speeds for that sector, then normalized to (i.e., divided by) the overall maximum NWR-estimated concentration. Peaks in the sector apportionment density curve indicate that a larger fraction of the average pollutant concentration is associated with the wind direction sectors. At least one peak in the sector apportionment density curve will equal 1.0 and indicates which wind sector had the highest NWR-estimated concentration. The cumulative sum of the sector apportionment density values is also shown (dark red line) and can be used to quantify the fraction of the average pollutant concentration associated with each peak. Placing the cursor over the sector apportionment curve displays the values below the plot. Subtract the cumulative density curve value at the start of a peak from the cumulative density curve value at the end of the peak to get the fraction for that peak. The Sector Apportionment Mean with Uncertainty Plot displays NWR-estimated average pollutant concentrations over the wind sector window (i.e., wind direction smoothing constant) centered at each wind direction degree (1-360°), along with uncertainty estimates calculated as ± two standard deviations (dashed lines). Like the Sector Apportionment Density plot, peaks in the sector apportionment mean curve indicate that a higher average pollutant concentrations occurred from the wind direction sectors that contains the peak. Larger uncertainty estimates for peaks in the sector apportionment mean curve indicate that some high pollutant concentrations occurred that impacted the NWR estimate resulting in a high standard deviation relative to the mean. These large uncertainty estimates will also coincide with high NWR-estimated average pollutant concentrations with SNR contours < 2 in the Polar and Cartesian plots.