A comparison of climatological observing windows and their impact on detecting daily temperature extrema
Abstract
Climatological observing window (COW) is defined as a time frame over which continuous or extreme air temperature measurements are collected. A 24-h time interval, ending at 00UTC or shifted to end at 06UTC, has been associated with difficulties in characterizing daily temperature extrema. A fixed 24-h COW used to obtain the temperature minima leads to potential misidentification due to fragmentation of “nighttime” into two subsequent nighttime periods due to the time discretization interval. The correct identification of air temperature extrema is achievable using a COW that identifies daily minimum over a single nighttime period and maximum over a single daytime period, as determined by sunrise and sunset. Due to a common absence of hourly air temperature observations, the accuracy of the mean temperature estimation is dependent on the accuracy of determination of diurnal air temperature extrema. Qualitative and quantitative criteria were used to examine the impact of the COW on detecting daily air temperature extrema. The timing of the 24-h observing window occasionally affects the determination of daily extrema through a mischaracterization of the diurnal minima and by extension can lead to errors in determining daily mean temperature. Hourly air temperature data for the time period from year 1987 to 2014, obtained from Toronto Buttonville Municipal Airport weather station, were used in analysis of COW impacts on detection of daily temperature extrema and calculation of annual temperature averages based on such extrema.
1 Introduction
The determination of daily minimum and maximum temperatures, often used in calculating the daily mean temperature, has been surprisingly difficult. On 1st July 1961, the climatological day was redefined in Canada to end at 06UTC instead at 00UTC in order to correct for the potential mischaracterization of air temperature minima. This change to the new climatological observing window (COW), however, introduced a cold bias to the observations by increasing the potential for recording minimal temperatures on two consecutive days from the same diurnal minimum, in effective double counting particularly cold days. Between 15 and 38% of days in Canada at different stations has been affected by this cold bias annually (Vincent et al. 2009, 2012). At the same time, it was found that the change in COW did not significantly affect the recording of daily maximum temperature. Unlike night time, day time with the new COW was not fragmented and the maximum temperature was unambiguously determined.
Non-climatic variations, such as somewhat arbitrarily defined COWs, affect proper assessment of climate trends and a number of techniques to date have been developed to detect inhomogeneities in data records. Homogenization was applied to the original temperature data recorded at 120 synoptic stations throughout Canada to correct for the bias in minimum temperature due to the change of COW. The temperature adjustment on individual days varied from 0.5 °C to as much as 12.5 °C (Vincent et al. 2012). It is clear that the COW affects the determination of daily air temperature extrema. This draws attention to the importance of developing a COW that corresponds better to the physical time frame of daily air temperature extrema.
Due to the common lack of continuous temperature observations (i.e., hourly), knowledge of daily air temperature extrema is crucial for the calculation of daily mean temperatures and in turn all other larger time periods such as monthly, seasonal, annual, and decadal. Daily mean temperature is the key variable in the field of climate observations, modeling, and climate change impacts. Climatological studies typically use mean temperature obtained from 24 observations at intervals of 1 h (true mean), derived from daily extrema (mean of daily extrema), (T_{max} + T_{min}) / 2, or fixed sampling at coarser than hourly temporal resolution. As a result of scarcity of hourly data records in general, many algorithms for the calculation of daily mean temperature have been developed (Hartzell 1919; Collison and Tabony 1984; Reicosky et al. 1989; Weber 1993; Harris and Pedersen 1995; Zeng and Wang 2012; Trewin 2004; Weiss and Hays 2005; Bonacci et al. 2013; Gough and He 2015).
Correct identification of daily air temperature extrema, i.e., minima and maxima measurements, is climatologically a vitally important, but surprisingly complex, task. Daily air temperature extrema are the points, on the temperature-time curve, in which the trend of daily air temperature cycle changes its sign. Temperature trend changes happens at least two times per solar day, with maximum air temperature typically occurring during the daytime, and the minimum air temperature during the nighttime as one expects from radiative considerations. The complications associated with local and regional geography, season, and with atmospheric circulation modify this picture enabling a wider range of times for the minimum and maximum temperature of the day to occur. For example, winds associated with midlatitude cyclones can cause the maximum temperature to occur at night and the minimum to occur during the daytime. As a general practice, the discrete determination of daily extreme values when performed on a continuous temperature data set searches for both minima and maxima over a 24-h time period time frame, fixed at a starting point regardless of the season or geographic location. The objective of this study is to assess the applicability of different climatological observing windows in addressing the problem of finding the true (mathematical) daily extrema.
2 Data and methods
Characteristics of three climatological observing windows and their related time frame for determination of diurnal air temperature extrema
COW | Beginning at | Ending at | Method of identifying T_{max} | Method of identifying T_{min} |
---|---|---|---|---|
COW_{0–24} | 0:00 LST | 23:00 LST | \( {T}_{\max }=\underset{0\le h\le 23}{ \max }{T}_h \) | \( {T}_{\min }=\underset{0\le h\le 23}{ \min }{T}_h \) |
COW_{6–6} | 06:00 LST | 05:00 LST (next day) | \( {T}_{\max }=\underset{6\le h\le 5}{ \max }{T}_h \) | \( {T}_{\min }=\underset{6\le h\le 5}{ \min }{T}_h \) |
COW_{ND} | t_{rise} + 1 h LST | t_{rise} + 1 h LST (next day) | \( {T}_{\max }=\underset{t_{\mathrm{rise}}\le h\le {t}_{\mathrm{set}}}{ \max }{T}_h \) | \( {T}_{\min }=\underset{t_{\mathrm{set}}\le h\le {t}_{\mathrm{rise}}}{ \max }{T}_h \) |
This small sample shows the effects of the observing window selection on the identification of daily minimum temperature. Values of daily maximum extrema obtained through three different observing windows are identical for all for days presented in Fig. 1 while two out of the 4 days have different minimum extrema as a result of the COW chosen. Method COW_{ND} detects all four minima within the appropriate nighttime periods. Also, COW_{ND} is the only method that detects the nighttime minimum during the third night in this example. All methods coincide in the minimum value for nights two and four but double counting of minimum value occurs for COW_{6–6} method on the second night and for both COW_{0–24} and COW_{6–6} methods on the fourth night. To illustrate these points, we walk sequentially through the 4 days. The sequence begins at night with the temperature falling with a minimum just before sunrise. Both COW_{ND} and COW_{0–24} identify this as a minimum (T_{nND} and T_{n} respectively). However, since the COW_{6–6} begins after this minimum, the first minimum for this window is the second night. This illustrates how the COW_{6–6} and COW_{0–24} typically identify minima from different night time periods. However, the third night introduces another wrinkle. All three methods capture the minimum temperature occurring before sunrise. However since the fourth night is substantially warmer than the third night, only the COW_{ND} identifies that minimum. For COW_{6–6}, the temperature at sunrise after the second night is the “third night” minimum, whereas for COW_{0–24} the minimum occurs at midnight of the fourth night. In both cases, a cold bias is introduced as these two temperatures are colder than the temperature actually experienced (and detected by COW_{ND}) on the third night.
The obvious reason for this incorrect representation of minima is the definition of climatological observing window associated with the determination of these extrema. Recording of daily minima using current observing practices involves searching for the daily minimum value using fragments of two different climatological events, i.e., two different nights. To avoid this fragmentation of nights and inconsistent association of minima to the incorrect nighttime segment, a nighttime-daytime discretization of time is proposed for the determination of daily extrema. Two qualitative and two quantitative criteria are used to objectively assess the ability of different COWs to correctly identify extrema in large temperature data sets. We first visually compare air temperature extrema and times of their occurrences obtained by different COWs to indicate the apparent magnitude and frequency of discrepancies between the methods. These observations are discussed in section 3.1. Section 3.2 illustrates how linear interpolation function, in conjunction with two different COW methods, tracks measured air temperature data. The first quantitative comparison focuses on the first two statistical moments of the distributions of differences between measured hourly temperatures and corresponding temperatures calculated by linear interpolation between consecutive extrema as obtained by two COWs. Distributions of differences are in fact error distributions of the associated COW methods. Comparison of mean and standard deviation is performed to determine the conformity of the COWs with observed temperature variation and presented in section 3.3. The second quantitative comparison focuses on the overall differences between minima, maxima, daily mean, and daily temperature range derived by COW selection of climatological parameters on annual basis. Results and observations of these comparisons are presented in section 3.4.
We now apply these COWs to a climatological data set. This data set includes hourly temperatures so that identification of true minimum and maximum temperature is unambiguous. Hourly air temperature data set was obtained from the Environment Canada Historical Climate Data Service for Buttonville A weather station for the period from June 1986 to December 2014.
The weather station is located at elevation of 198.1 masl and latitude and longitude of φ = 43°51′44″ N, λ = 79°22′12″ W respectively. The weather station is located in a suburban environment with humid continental climate characterized by severe winters, warm summers, and strong seasonality. The air temperature is measured at 1.25–2 m above the ground. In addition to hourly air temperature data, solar declination data at 18:00 UTC (corresponding to13:00 LST) were downloaded from ssd.jpl.nasa.gov website for the purpose of calculation of sunrise and sunset times for the 1986- to 2014-year time period.
3 Results
3.1 A comparison of climatological observing windows
Hourly air temperature data set was utilized for identification of daily temperature extrema with the three different climatological observing windows, COW_{0–24}, COW_{6–6}, and COW_{ND}. In addition to the respective air temperatures extrema, occurrence times of extrema were also recorded. Further, true mean daily temperature associated with each of the climatological observing windows was calculated as arithmetic mean of 24 hourly temperatures as well as those derived as the average of the minimum and maximum temperatures derived by each of the COWs.
- 1.
Differences among the observing windows are not apparent for the air temperature maxima as it is for the minima. Maxima are generally identified correctly by all three observing windows.
- 2.
Minima are frequently visibly mischaracterized. In such cases, differences between the minima obtained by COW_{ND} and other two methods are commonly substantial.
- 3.
Regular small variations between identified minima demonstrate the larger difference between COW_{ND} and other two observing windows method, COW_{0–24} and COW_{6–6}, than between the COW_{0–24} and COW_{6–6} methods.
- 4.
Minima identified by COW_{ND} method are generally higher than the minima identified by the other two observing windows.
- 5.
On days when daily air temperature cycle does not conform to the solar heating-cooling patterns, it is likely that the local weather pattern is influenced by external factors, such as the passage of midlatitude cyclone. These situations are typically characterized by prolonged increasing or decreasing temperature patterns. In such situations, determination of daily extrema within the COW_{ND} framework yields out-of-place extrema with identical or very close daily minimum and maximum values due to loss of the radiatively induced sinusoidal pattern of the temperature diurnality. Figure 7 presents two examples of misidentification of COW_{ND} method’s inability to correctly determine daily extrema on days with prolonged increasing or decreasing temperature patterns.
- 6.
Observations 1 to 5 appear to be independent of season or year
3.2 Linear tracking of air temperature data based on COW_{0–24} and COW_{ND} determined extrema
At the same time, minima and maxima detected by the COW_{ND} method coincide with true mathematical extrema, and consequently, linear function connecting extrema points tracks the temperature trend correctly.
3.3 Error distribution comparison of two contrasting climatological observing windows
Summary statistics for error distributions of COW_{0–24} and COW_{ND} methods
METHOD | Minimum (°C) | 1st quart. (°C) | Median (°C) | Mean (°C) | 3rd quart. (°C) | Maximum (°C) | St. Dev. (°C) |
---|---|---|---|---|---|---|---|
COW_{0–24} | -17.93 | -1.14 | -0.09 | -0.30 | 0.79 | 10.70 | 1.98 |
COW_{ND} | -10.96 | -0.09 | -0.07 | -0.12 | 0.75 | 10.57 | 1.53 |
Mean and standard deviation of COW_{ND} error distributions are −0.12 and 1.53 °C respectively while their values for COW_{0–24} are −0.30 and 1.98 °C. Apparently, error distribution for COW_{ND} is centered closer to zero and is slightly narrower than COW_{0–24} distribution rendering this method better in the sense of the second criterion discussed in section 2. This observation is confirmed by quartiles and distribution bounding values presented in Table 2.
3.4 Effect of COW selection on climatological parameters
Finally, the annual averages of daily temperature ranges of COW_{ND} are on average lower by ~0.97 °C than the annual averages of daily temperature range determined by COW_{0–24} and by ~1.51 °C in comparison to COW_{6–6} method (Fig. 17). The systematic nature of the observed differences and their magnitude demonstrate that the observing window selection affects the determination of daily extrema.
4 Conclusions
Position of the observing window affects the determination of daily extrema and thus potentially biases climatological assessments. Incorrect characterization of daily air temperature extrema affects diurnal minima more frequently than the maxima. Annual averages of minima determined by COW_{ND} for the location of Buttonville and time period from 1986 to 2014 are warmer approximately 0.64 °C than the annual averages of minima determined by COW_{0–24} and COW_{6–6}.
The COW_{ND} framework is proposed for the improvement of the quantification of diurnal distribution of air temperature extrema. Night and Day (Local Standard Time) Climatological Observing Window COW_{ND} detects true daily temperature extrema more reliably than climatological observing window COW_{0–24} and COW_{6–6}. COW_{ND} method errs occasionally in absence of a sinusoidal diurnality in daily temperature cycle. Loss of the oscillatory feature in daily temperature variation seems to be symptomatic of unusual weather variations and as such can be indicative of local scale atmospheric phenomena. Results of this preliminary study provide a cautionary note on the determination of daily extrema when assessing the global warming based on extrema means and its sensitivity to the particular COW used. If climate change assessment is based on means of extrema, annual averages of COW_{0–24} method are systematically lower than the annual averages obtained by COW_{ND} method with the differences ranging between 0.07 and 0.25 °C.
Footnotes
References
- Bonacci O, Zeljkovic I, Sakic Trogrlic R, Milkovic J (2013) Differences between true mean daily, monthly and annual air temperatures calculated with three equations: a case study from three Croatian stations. Theor Appl Climatol 114:271–279. doi:10.1007/s00704-012-0830-8 CrossRefGoogle Scholar
- Collison P, Tabony RC (1984) The estimation of mean temperature from daily maxima and minima. Meteorological Magazine, London 113:329–337Google Scholar
- Gough WA, He D (2015) Diurnal temperature asymmetries and fog at Churchill, Manitoba. Theor Appl Climatol 121:113–119. doi:10.1007/s00704-014-1227-7 CrossRefGoogle Scholar
- Harris SA, Pedersen JH (1995) Comparison of three methods of calculating air temperature from electronic measurements. Z Geomorph 39:203–210Google Scholar
- Hartzell FZ (1919) Comparison of methods for computing daily mean temperatures: effect of discrepancies upon investigations of climatologists and biologists. Mon Weather Rev 47:799–801CrossRefGoogle Scholar
- Reicosky DC, Winkelman LJ, Baker JM, Baker DG (1989) Accuracy of hourly air temperatures calculated from daily minima and maxima. Agric For Meteorol 46:193–209CrossRefGoogle Scholar
- Trewin B (2004) Effects of changes in algorithms used for the calculation of Australian mean temperature. Aust Meteorol Mag 53:1–11Google Scholar
- Vincent LA, Xiolan LW, Milewska EJ, Hopkinson R, Malone R (2009) Bias in minimum temperature introduced by a redefinition of the climatological day at the Canadian stations. J Appl Meteorol Climatol 48:2160–2168. doi:10.1175/2009 CrossRefGoogle Scholar
- Vincent LA, Xiolan LW, Milewska EJ, Wan H, Yang F, Swail V (2012) A second generation of homogenized Canadian monthly surface air temperature for climate trend analysis. J Geo Research 117:D18110. doi:10.1029/2012JD017859 Google Scholar
- Weber RO (1993) Influence of different daily mean formulas on monthly and annual averages on temperature. Theor Appl Climatol 47:205–213CrossRefGoogle Scholar
- Weiss A, Hays CJ (2005) Calculating daily mean temperatures by different methods: implications from a non-linear algorithm. Agric For Meteorol 128:57–65CrossRefGoogle Scholar
- Zeng X, Wang A (2012) What is monthly mean land surface temperature? Eos 93(15):156–157CrossRefGoogle Scholar