Temperature

BCRMSE

    • For both the 00 and 12 UTC initializations, a diurnal trend is noted in the distribution of BCRMSE values, which range from 2.2 – 4.0 °C.
    • The smallest errors occur for forecasts valid from 06 – 09 UTC, and the largest errors occur for forecasts valid at 00 UTC.
    • BCRMSE increases with forecast lead time. In many cases, this increase is statistically significant (SS) for forecasts valid at the same time, 24 hours apart (e.g., the 12-hr versus the 36-hr forecast lead time BCRMSE values from 12 UTC initializations).
    • In general, all seasonal aggregations follow the characteristics of the annual aggregation discussed above, with the exception of winter, which does not have a diurnal cycle and shows a steady increase in BCRMSE values with lead time.

Bias

    • A SS cold bias is noted for all forecast lead times and both initializations for all temporal aggregations; the largest cold bias occurs for forecasts valid at 18 UTC, while the smallest cold bias is seen for forecasts valid at 09 UTC for all seasonal aggregations except winter, where the diurnal trend is shifted revealing the largest cold bias occurs for forecasts valid at 00 UTC and the smallest at 12 UTC.

Dew Point Temperature

BCRMSE

    • Similar to surface temperature, a diurnal trend is displayed in the distribution of BCRMSE values, ranging from approximately 2.3 – 5.5 °C, with the largest errors and confidence intervals (CIs) associated with the winter aggregation. A general increase in BCRMSE values is noted with lead time.
    • For all temporal aggregations except winter, the largest errors occur for forecasts valid from 21 – 00 UTC, and the smallest errors occur for forecasts valid from 06 – 09 UTC. For winter, the largest errors are seen between 09 – 12 UTC, while the smallest errors are generally valid around 18 UTC.

Bias

    • More so than any other variable or metric evaluated, the bias for surface dew point temperature is sensitive to the forecast initialization hour, lead time and temporal aggregation.
    • In general terms, a SS high bias is noted during the late afternoon into evening hours and a SS low bias is seen during the morning hours for all temporal aggregations, except winter. For the winter aggregation, a consistent SS high bias is noted for nearly all lead times, regardless of initialization hour.

Wind

BCRMSE

    • For both the 00 and 12 UTC initializations, BCRMSE values range from approximately 1.8 – 2.6 m/s. All temporal aggregations, especially for the 00 UTC initializations, exhibit and increase in error with forecast lead time. With the exception of winter, all temporal aggregations also have a notable diurnal trend.
    • The largest errors occur for forecasts valid at 21 – 00 UTC for all temporal aggregations, while the smallest errors occur for forecasts valid around 09 – 12 UTC.

Bias

    • A consistent SS high bias is noted at all forecast lead times for both initializations, and all temporal aggregations. A diurnal cycle is displayed in the distribution of values, with SS lower high wind bias noted at 00 UTC compared to many other valid times.

3-hour Precipitation Accumulation

GSS

    • Regardless of initialization hour, forecast lead time or temporal aggregation, the largest GSS values are seen at the lowest thresholds and steadily decrease to near-zero for the highest thresholds. GSS values also decrease with lead time. The CIs are generally smallest for the summer season and largest for the winter season.
    • The number of observed events (base rate) by threshold has a similar trend, with winter having the lowest base rate of any temporal aggregation.

Frequency Bias

    • A SS high bias is seen for all thresholds up to around 0.25”, where the exact threshold depends on the temporal aggregation and lead time examined. Above that threshold, for the annual and summer aggregations there tends to be a SS low bias, while the other seasons CIs often encompass one.

24-hour Precipitation Accumulation

GSS

    • A decrease in GSS with increasing threshold and lead time is noted in all temporal aggregations except winter, where the median values remain high but the CIs are very large indicating low confidence in those values. Again, a similar trend is noted in the base rate, with very few observed events of 3” accumulations in 24 hours.

Frequency Bias

    • Generally, regardless of temporal aggregation or initialization hour, the CIs above the 1.5” threshold encompass one. Below that threshold, for the annual and spring aggregations, a SS high bias is noted. For the winter aggregation the SS high bias is only seen up to the 1.0” threshold, while the summer season generally exhibits a SS high bias between the 0.5” and 1.5” thresholds only. Finally, the CIs for the fall aggregation often encompass one, with only a few exceptions depending on the lead time and threshold.

Temperature

BCRMSE

    • The overall distribution for BCRMSE generally shows a minimum error between 500 and 300 hPa and a maximum at 850 hPa for all temporal aggregations.
    • A SS increase in error with lead time is noted, with the errors ranging from approximately 0.6 – 2.2 °C, depending on lead time, pressure level, and temporal aggregation.

Bias

    • For all temporal aggregations, a SS cold bias is seen for the lower levels (850 and 700 hPa), transitioning to a SS warm bias at and above 500 hPa, regardless of forecast lead time.
    • The largest bias occurs in the upper-levels for all forecast lead times.

Dew Point Temperature

BCRMSE

    • For all temporal aggregations, as the pressure level decreases from 850 to 500 hPa and the forecast lead time increases from 12 hours to 48 hours, the BCRMSE values increase, with values ranging from approximately 5 – 11 °C.

Bias

    • In general, a SS moist bias is noted, where, as the pressure level decreases, the bias increases. The bias does not increase with lead time and, in fact, decreases and encompasses zero for the 48-hour lead time at the 850 hPa level for all temporal aggregations. The summer aggregation is an exception, which frequently shows a SS dry bias at and below 700 hPa, especially beyond the 12-hour lead time.

Wind

BCRMSE

    • In general, an increase in BCRMSE values from a minimum at the lowest levels to a maximum around 300 – 200 hPa with decreasing values above that, is noted for all temporal aggregations.
    • Overall, the errors range from approximately 1.6 – 6.5 m/s. While a SS increase in error with lead time is noted for all levels, a larger increase in errors is seen for levels below 200 hPa with lead time.

Bias

    • A SS high wind bias is seen for all forecast lead times at the lowest levels, transitioning to a SS low wind bias above 700 hPa for lead times out to 36 hours. For the 48-hour lead time, many of the levels encompass zero, with a SS low bias at 200 and 100 hPa only.