Frost events have been increasing in recent years and growers of a wide range of crops have been implementing different frost mitigation measures to protect their crops. For some frost mitigation strategies, reliable forecasts beyond the same day are not as necessary, since they can be implemented on a moment’s notice (e.g. wind machines and irrigation). For other strategies such as row covers or deciding when to transplant tender annuals, it is important to have a reliable forecast a few days in advance. For example, for ginseng, it can take up to two days to put row cover over the crop.
As a result, we compared 5 forecasting sites over a month from early April to early May to see how accurate they were for the Simcoe (Norfolk County) area. Forecasts were determined for Simcoe, and the forecasts were compared to a weather station at the Simcoe Research Station approximately 2 km from Simcoe. It should be noted that this was not a scientific study and was done for only a relatively small amount of time. April is a highly variable month and forecasts would be less reliable at this time of year compared to later in the summer. The Simcoe Research Station weather station, while following many the rules for weather monitoring, is not an official Environment Canada station, so there may be some deviation from exact temperatures. This study is meant to provide only a relative comparison between the five sites.
All temperature forecasts and official temperature readings are based on a thermometer placed inside a white louvered box (to keep the thermometer out of the sun but with plenty of air flow) over a mowed grass surface at 1.5 m off the ground and 30 m away from trees, pavement or other obstacles. Keep in mind that thermometers placed closer to the ground will be warmer during the day and colder at night than what is forecast. The Simcoe Research Station weather station thermometer is at 1 m off the ground. Thus, temperatures will be warmer during the day and colder at night on average than official temperature readings.
First of all, here is a comparison of the sites and the information provided by each. Smart phones, radio station and website weather forecasts will usually be based on one of these forecasts.
Environment Canada (the official weather forecasting service of Canada): 7 day high and low temperature and condition forecasts, official weather advisories and warnings and hourly forecasts for 24 hours for temperature, conditions and wind speeds (click on “Today” in the local forecast). Also provides official historical records for many locations.
Weather Network: Same as Environment Canada but also 36 hours of dewpoint, wind speed, temperature and condition forecasts. Also provides 14 day forecasts (read further for comments on long range forecasts)
Weather Underground: Provides 10 days of hourly temperature, wind speed, cloud cover, conditions, rainfall, dew point and humidity forecasts (click on “Table” and then the day you want to view for the details).
Meteoblue: Provides 7 days with 3-hour forecast periods for temperature, precipitation, conditions and wind speeds.
Accuweather: Most of the same information as Weather Underground but hourly forecasts beyond three days require a subscription. Daily high and low temperature forecasts extend to 90 days (read further for comments about long-range forecasts).
To assess the accuracy of a forecast you need to know both how skewed the forecast is on average and how much it deviates from that skew. The skew is the number of degrees (+ or -) that the forecast site is normally off from the actual high or low. So a forecast site with a negative skew is usually underestimating the actual temperature, and a positive skew indicates that it is on average overestimating the actual temperature. The standard deviation is how variable the forecast site is. A high standard deviation indicates that the forecast temperature can be much higher or lower than what was forecast beyond the skew. For example, a forecast low temperature for a day is 0oC. If the skew for that forecast is 2oC and the standard deviation is 3oC then the actual temperature on average will be -2oC and in the range of -5oC to +1oC 50% of the time. The other 50% of the time, the actual temperature will be below -5oC or above 1oC.
A perfect forecast should have low skew and low standard deviation. The next best thing would be a site with high skew but low standard deviation. This is because this would represent a site that is consistently higher or lower than the actual temperature. If you know a site is always 3 degrees higher than it should be, then it is easy enough to correct for that difference.
Average skew (Table 1) and standard deviation (Table 2) values for the five forecast sites are provided. Note that all of the forecast sites over all periods are negatively skewed during the day and positively skewed at night. Some of this skew can be attributed to the fact that the thermometer at the Simcoe Station is closer to the ground than it should be. Even with this in mind, the actual temperatures are usually above the forecast during the day and below the forecast at night. Weather Underground had both the lowest skew and the lowest standard deviation of the five sites for daily high temperature forecasts. Accuweather had the lowest skew for daily low temperature forecasts, but also high variability. Environment Canada had the lowest variability in the daily low temperature forecasts.
When I looked at only those nights with clear skies and calm winds, which are the important nights for predicting frost events, Environment Canada had both the lowest skew and the lowest standard deviation. Even this forecast though, had a standard deviation of 2oC over the one to four day forecast periods and was skewed 2oC warm, which is probably closer to 1oC when you factor in the lower height of the thermometer in Simcoe. So if the forecast low is -1oC, the actual low will on average be about -2oC but between 0oC and -4oC 50% of the time. Some of this variability is probably due to a few forecasts that predicted cloud cover at night when it was actually clear. So in reality, the variability on nights that are known to be clear many days in advance would probably be a lot lower.
Forecasts for the Simcoe area may have higher variability than other locations in Ontario. This is because Norfolk County has very sandy soils that heat faster during the day and cool faster at night than other soils that retain more moisture. This would be especially true in the spring when most fields are bare. Weather models used for forecasting, probably do not account for these local microclimates very well.
All of the sites begin to become very unreliable beyond the 5 or 6 day forecasts. Examining forecasts beyond this period is not useful.
So what is the best site to use? Based on the variability in the forecasts, it is best to look at several forecasts and err on the side of caution. Keep in mind that beyond the variability in the forecasts discussed above, there can be significant local differences in actual temperatures within a small area. Five weather monitoring stations within 5 km of Simcoe over this same period ranged up to 3oC on some of the calm clear nights in the monitoring period.
So what is the best strategy for growers to use when trying to forecast frost events?
- Look at several different forecast sites and look at the range of forecast low temperatures.
- Err on the side of caution and assume the worst case scenario.
- Examine specifically the wind speeds, cloud cover and dew points to give an idea of how low temperatures may go.
- Keep in mind that temperatures can be significantly lower on the ground than the actual air temperature, and the drier the air, the larger this difference can be. This is important for crops that are close to the ground.
- Keep track of forecasted low temperatures and your actual low temperatures so you can adjust forecasts to more reliably match your local microclimate over time.