Forecasting Methodology

by Ron Hartling

sIn the absence of any authoritative source of PM2.5 forecasts for the Kingston area, I have over a six-month period, learned to base my forecasts on combining three main sources, and correlating them intuitively based on observations of how daily observation of all three actually turned out in the next day’s readings. While accurate prediction of exact readings proved impossible, my best guesses have fairly reliably predicted the magnitude and approximate timing of increasing and decreasing trends, which is what one really needs to know in order to be in a position to take effective precautions.

Based on analysis of the most recent four years of Kingston hourly PM2.5 readings compared to Environment Canada data on measured hourly wind directions and speeds, I found a strong correlation with wind direction. I coupled that with analysis which demonstrated little or no correlation between actual PM2.5 readings and  weekends versus weekdays or overnight versus working hours, thereby establishing that Kingston’s airborne particulate pollution is almost exclusively caused by emissions from other jurisdictions as opposed to local sources. Our MP2.5 rise significantly whenever winds from the south and southeast are bringing us particulates from the highly-industrialized Northeastern United States.  During wildfire season, the impact of other wind directions depends on the current location of major smoke clouds.

I also refer to the US EPA AirNow map whenever south and southeast winds are included in the forecast, which provides a preview of what’s coming our way. By default, the map does not display Level 1 particulate clouds, presumably adhering to the highly-inadequate, politically-influenced US definition of an average annual PM2.5 exposure of 12 µg/m3 being the maximum “safe” level (much of the Northeastern US frequently exceeds that limit). In order to view the extent of Level 1 clouds, click the tiny “Show green contours” check box in both the Current and Forecast tabs. Also, be aware that the forecasts automatically stop at the Canada/US border.

There is a further complication in the correlation between Environment Canada’s wind-direction forecasts for the Kingston area and the observed arrival of PM2.5 particulate clouds passing through from the US. When that forecast includes quite variable local wind conditions, the overarching Jetstream appears to be a more accurate predictor of when and if those clouds actually pass over or near Kingston. 

Finally, during wildfire season, I refer to the Canadian wildfire smoke model map, which does include Level 1 airborne pollution. Its limitations are three-fold. First, its Level 1 range is from 1 to 12 µg/m3. I consider anything under the evidence-based WHO annual average limit of 5 µg/m3 to be a reasonably healthy exposure level and therefore refer to it in my charts as “Level 0” with a green colouration, and deem higher Level 1 readings as moderately unhealthy in the long run. Unfortunately, there is no way to distinguish those in the map. Second, the underlying model is based on wildfire intensity, so is oblivious to the airborne industrial emissions which constitute the majority of Kingstonians’ annual exposure.  Third, the depicted smoke clouds from Western and Northern Canada have typically dissipated sufficiently by the time that they reach Eastern Ontario as to have minimal impact on Kingston air quality.

The cause-and-effect relationship between those three sources is situation-dependent given that winds can only blow what particulates are or are not present in the air masses which they happen to pass through.  In order to make that information useful for personal decision-making, I attempt to encapsulate into a one-line forecast summary what experience suggests is the most likely outcome of the juxtaposition of those three factors into a forecast risk level (0, 1, 2, etc.) for the next day. While no weather-related forecast can ever be perfect, my recent predictions have proven over time to be roughly 70% accurate.