Climatology and trends in some adverse and fair weather conditions in Canada,

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2005jd006155, 2006 Climatology and trends in some adverse and fair weather conditions in Canada, Xiaolan L. Wang 1 Received 29 April 2005; revised 17 October 2005; accepted 12 December 2005; published 13 May [1] This study characterizes the climatology and trends of some adverse and fair weather occurrence in Canada, on the basis of reliable long-term records of hourly weather and bright sunshine observations at about 90 Canadian stations for the period The results show that fair or no-weather (i.e., no precipitation or visibility obscuration) trends are generally consistent with the sunshine trends, showing significant increases in southern Canada, with decreases in the Canadian Arctic. The increase of no-weather is most extensive in spring and summer, while the decrease is most extensive in autumn. For the same period, freezing precipitation has become more frequent in the region north of 50 N (especially in spring and autumn) but less frequent in southern British Columbia (BC), central Prairies, and the Great Lakes area in autumn-winter, as well as in northeastern Canada in winter. Blowing snow occurrence has decreased significantly almost everywhere across Canada, with the most significant decline in southwestern Canada in winter. In every season, fog occurrence has significantly increased in the Prairies-Yukon-Northwest Territories and northern BC but decreased in eastern Canada and southern BC. The frequency of low ceiling conditions (<304.8 m, i.e., 1000 feet) has increased in Alberta-BC interior and the Great Lakes area, with the most significant increases in Alberta-BC interior in autumn-winter and in the Great Lakes area in spring; while the other regions have experienced a negative trend, which prevails across the country in summer. Citation: Wang, X. L. (2006), Climatology and trends in some adverse and fair weather conditions in Canada, , J. Geophys. Res., 111,, doi: /2005jd Introduction [2] By definition, adverse or potentially hazardous weather refers to weather phenomena that have adverse impacts on the socioecosystem of the Earth. It can be associated with various types of weather, such as extreme temperatures, precipitation (amount and types), wind, and visibility restrictions, and so on. Adverse weather events are of particular concern because of their adverse impacts on the local human and other inhabitants and systems. For example, freezing precipitation could jeopardize surface transportation and increase mortality of young animals, among its other adverse impacts. Weather phenomena such as fog and blowing snow are significant problems for the safety of transportation (surface, marine, and aviation), because of their obstruction to vision. Therefore it is of crucial importance to characterize the variability and trend in the frequency and intensity of adverse weather events. Results of research in this area are of tremendous applicability. Analyzing observed data for the last 5 decades or so, many previous studies have reported trends in temperature, precipitation, and cyclone activity in Canada [e.g., Zhang et al., 2000; Serreze et al., 2000; Bonsal et al., 2001; Shabbar and 1 Climate Research Division, Atmospheric Science and Technology Directorate, Environment Canada, Toronto, Ontario, Canada. Published in 2006 by the American Geophysical Union. Bonsal, 2003; Stone et al., 2000; Wang et al., 2006a, 2006b]. However, very limited effort has been put on assessing trends associated with other types of adverse weather in Canada, particularly sensible weather elements or phenomena such as freezing precipitation, blowing snow, fog, and low ceiling conditions. Only recently, Hanesiak and Wang [2005] analyzed trends in the occurrence frequency of the above four types of adverse weather, and of no-weather (i.e., no precipitation or visibility obscuration) occurrence in the Canadian Arctic. (The four types of adverse weather conditions were chosen for the study because they all have significant impacts on transportation/aviation.) Their results show that the frequency of freezing precipitation (FZ) has increased almost everywhere across the Canadian Arctic, but the occurrence of blowing snow (BS) has become significantly less frequent. Significant changes were also identified in the occurrence frequency of fog, low ceiling (LC), and no-weather (NoWx) events. All these changes were found by the same authors to be consistent with changes identified in temperature, precipitation, and cyclone activity in the same region during a similar time period. [3] This study aims to extend the work by Hanesiak and Wang [2005] by assessing observed changes in the frequency of the adverse weather events (FZ, BS, fog, LC) as compared to fair weather (NoWx and bright sunshine hours) across Canada. The definitions for both adverse and no-weather follow those of Hanesiak and Wang [2005] and were 1of27

2 Figure 1. Locations of (a) 95 stations providing long-term (30+ years) hourly current weather observations and (b) 86 stations providing bright sunshine data used in this study. Circles indicate stations of shorter time series of data (30 49 years; see Tables A1 and A2). The black, green, and red plus signs indicate stations that are grouped as the Arctic, SBC, and GLA stations (see Tables 1 and 2). Abbreviations are defined as follows: YT, Yukon; NWT, Northwest Territories; NU, Nunavut; BC, British Columbia; SBC and NBC, southern and northern BC; ALTA, Alberta; SASK, Saskatchewan; MAN, Manitoba; ONT, Ontario; QUE, Quebec; NB, New Brunswick; NS, Nova Scotia; NFLD, Newfoundland (excluding Labrador); PEI, Prince Edward Island; and GLA, the Great Lakes area. extracted from the Manual of Surface Weather Observations (MANOBS [Environment Canada, 1977, hereinafter referred to as EC77]). [4] Freezing precipitation includes freezing rain and freezing drizzle, which are drops of rain which freeze on impact with the ground or with other objects at or near the Earth s surface and should be reported when rain or drizzle is freezing on the Ice Accretion Indicator or on other objects at or near the Earth s surface [EC77, p. 3 4]. It does not include ice pellets, nor snow pellets. Blowing snow events are an obstruction to vision other than precipitation but due to snow particles raised by the wind to sufficient heights above the ground to reduce the horizontal visibility at eye level (1.8 m above the surface) to 9654 m (i.e., 6 miles) or less [EC77, pp. 2 1 and ]. It does not include drifting snow (which by definition is a large quantity of snow particles raised by the wind so that very low objects are veiled or hidden and yet the visibility at eye level is not appreciably restricted [EC77]). Fog is a suspension of very small water droplets in the air, reducing the visibility at the Earth s surface [EC77, p. 3 7] and is reported as an official event if it reduces the visibility to less than 1 km. [5] Ceiling height is defined as the lesser of (1) the height above ground of the base of the lowest layer aloft, at which the summation of opacity is 6/10 or more of the whole sky, or (2) the vertical visibility in a surface-based layer which completely obscures the whole sky [EC77]. (A layer aloft is a layer whose base is sufficiently high above the surface to show clear spaces beneath. A surface-based layer is a layer whose base is at the ground level; e.g., fog, smoke, falling or blowing snow.) In this study, low ceiling (LC) is used to refer to any ceiling height that is lower than m (i.e., 1000 feet). [6] By definition, LC events may include some, but not all, fog events. For example, fog patches are not included in LC events, because they are a surface-based layer which does not completely obscure the whole sky (by definition, fog patches consist of fog extending to at least 2 m above ground level and whose areal extent comprises less than 50% coverage of the ground normally visible from the observing point [see EC77, p. 3 7]). [7] Any hourly weather report with no precipitation or visibility obscuration occurrence is defined as a no-weather (NoWx) event, which is used as a fair weather indicator. Another fair weather indicator used in this study is bright sunshine hours. In Canada, as in most countries throughout the world, sunshine data are obtained from the Campbell- Stokes Sunshine Recorder. The instrument provides a record of bright sunshine, i.e., sunshine that is bright enough to burn or scorch a standard card upon which the rays from the sun have been concentrated by the glass sphere of the recorder [Environment Canada, 1973]. The amount of bright sunshine for each hour is measured and expressed in tenths of an hour [Environment Canada, 1973]. [8] As described by Hanesiak and Wang [2005], the hourly observations of current weather and ceiling heights are mainly subjective human observations. Potential issues with these subjective observations include: observer error and inconsistencies among the different observers, observational errors due to low visibility or darkness, and so on. Although observers are highly trained for current weather observations, inconsistencies can still arise among them, especially for freezing precipitation that is mixed with other types of precipitation. Estimation of ceiling heights near 1000 ft can pose some errors among observations (daylight and darkness) during periods in which ceilometers or pibals (pilot balloons) were not available. Ceilometers also have some degree of error, although it is very much smaller (about 3 10 m) compared to human estimation [Hanesiak and Wang, 2005]. Pibals were regularly used in Canada; thus ceiling height errors are considered not to be problem- 2of27

3 atic, although a ceilometer relocation or type change could cause a mean shift (discontinuity) in the data time series (to be dealt with in section 2). Inherent visibility errors associated with blowing snow and fog are also considered to be relatively small (near 10%) since all stations used special distance markers, and in more recent times, a combination of distance markers and visibility instruments. Quantifying these biases and any other errors in the observations would be very difficult and was not attempted in any great detail in this study. However, all significant artificial steps were removed prior to the trend analysis, in order to obtain a more realistic estimate of trend. The technique for removing artificial step changes is described in the next section (section 2). [9] The remainder of this paper is arranged as follows. The data set, as well as the homogeneity test and trend analysis methods, are briefly described in section 2. The climatology of each of the six variables (FZ, BS, fog, LC, NoWx, sunshine) in each season is briefly discussed in section 3. The observed changes in the period are presented in section 4, with a summary and some discussions in section Data and Methods of Analysis [10] In this study, any hourly report indicating FZ (BS, fog, LC, NoWx) occurrence is defined, and hence counted, as an FZ (BS, fog, LC, NoWx) event. Monthly occurrence counts (and frequencies) of each of the adverse/fair weather events were derived from hourly reports of current weather (including all the five elements: FZ, BS, fog, LC, and NoWx) at 95 stations across Canada (see Table A1 in Appendix A and Figure 1a). Note that the 15 Arctic stations (see Figure 1a) that were analyzed by Hanesiak and Wang [2005] are also included here to show a complete picture for Canada, but with limited discussions to avoid redundancy. [11] These 95 stations were selected because they have nearly complete records of the related hourly weather reports recorded in the digital data archive of Environment Canada (EC) for at least 30 years during the period from January 1953 to July 2004 (see Table A1; note that 1 January 1953 is the first day of digital data for most elements in the EC digital archive, which is the reason for 1953 being the initial year for this study; we do not have the related digital data for periods before 1953). Some hourly reports are missing at some stations, which are often due to observing program changes (e.g., from 24 hours 7 day to daytime only ) or station automation or relocations/site constructions, and so on (some reasons are unknown). To diminish effects of biased/inadequate sampling (e.g., daytime only observations are not representative of nighttime climates), any month of 50% or more missing hourly reports was excluded from the analysis (note that large percentages of missing reports are often associated with observing program changes). In other words, no explicit interpolation for months of missing observations was done in this study. Any month of 50% or more missing hourly reports (both the frequency/count data and the corresponding time t) was excluded from the regression analysis below (i.e., regression of the response variable in question on the relevant time/ month t of observation, which are not necessarily consecutive throughout the period). For each element at each station, the number of months excluded is listed in Table A1. In the retained months, the overall percentage of missing hourly reports is generally below 1.4%, with the exception of two stations that have 7% (Miramichi A, NB) and 8.3% (Kamloops A, BC) missing respectively (Table A1). [12] In addition, bright sunshine data from 86 stations in Canada (see Table A2 and Figure 1b) are also analyzed similarly to assess trends in the occurrence of fair weather. Here, monthly or seasonal sunshine amount is represented by the percentage of time with bright sunshine in each calendar month or each season (this has some similarity to the transformation of count data described below). For each station, the time series of monthly/seasonal sunshine amount was derived and analyzed for trends after a homogeneity test. [13] As discussed in details by Hanesiak and Wang [2005], it is necessary to properly transform the count data (which have binomial distributions), so that conventional regression models can be used to estimate trends in time series of frequency. Also, there exist various sources (such as station relocation or instrument/observer changes, etc.) for artificial step changes (mean shifts) in the time series of occurrence count/frequency or the bright sunshine data, which need to be taken into account in the related trend analysis to obtain a more realistic estimate of trend. Data transformation and homogenization and trend analysis were thus completed based upon the methods of Hanesiak and Wang [2005]. Note that, prior to the homogeneity and trend analyses, the long-term mean annual cycle was removed from the monthly time series to reduce the data dependence between consecutive months. [14] Letting s t denote the occurrence count of a type of event in month/season t of m t observations (so f t = s t /m t is the occurrence frequency), then, for the case of homogeneous time series, the regression model: s t þ 0:5 h t ¼ log ¼ m þ bt þ e t m t s t þ 0:5 was used to estimate the linear trend b in time series of (empirical) log odds h t, where e t denotes a zero-mean Gaussian process. As shown by Hanesiak and Wang [2005], the above linear trend b is equivalent to an exponential trend in the corresponding frequency time series, which is nearly linear over the period of around 50 years of data. [15] The homogeneity test for time series h t is based on this two-phase regression model [Wang, 2003]: h t ¼ m 1 þ bt þ e t ; N 1 t t c m 2 þ bt þ e t ; t c < t N 2 where t c denotes a possible point of artificial mean shift of size d c =(m 2 m 1 ) 6¼ 0, which divides the time series h t for t 2 {N 1, N 1 +1,..., N 2 } into two segments: t 2 {N 1, N 1 +1,..., t c } and t 2 {t c +1,t c +2,..., N 2 }(1N 1 < N 2 N). For each and every trial value of t c 2{N 1 + N min, N 1 + N min +1,..., N 2 N min } (where N min is a selected minimum length of segment), the sum of squared errors (SSE) of model (2) was compared with that of model (1). The time t c that is associated with the maximum reduction in the SSE (among all the trial values of t c ) and statistically significant ð1þ ð2þ 3of27

4 Figure 2. Time series of monthly log odds of fog occurrence observed at (a) Sandspit A, (b) Prince Rupert A, and (c) Dauphin A. (with climate ID given in parentheses). The dashed line is the trend line estimated from the raw data (without taking into account the mean shifts shown). Figure 3. Long-term mean frequencies (in percentage) of freezing precipitation (FZ) occurrence in (a) autumn, (b) winter, and (c) spring. 4of27

5 Figure 4. Long-term mean annual cycles of the occurrence frequency (in percentage) of the adverse and fair weather conditions ((a) FZ, (b) BS, (c) fog and LC, (d) NoWx, and (e) bright sunshine) at the indicated stations. improvement in the fit of model (2) (over model (1)) is chosen as a possible change point (point of mean shift). This procedure was repeated until all segments are either deemed to be homogeneous at the selected significance level or too short to be divided further (i.e., each segment has fewer than 2 N min data points). During the process of detecting change points, visual inspection of the time series in question was carried out to help determine whether or not to include a change point in the time series; and metadata (station history and inspection reports; if available) were used to check the veracity of the mean shifts detected. However, there exist undocumented change points (i.e., change points that were identified to be significant statistically and visually, but no metadata were available to verify them or no reason was found to explain them), which need to be and were corrected. We also used spatial consistency of trend as another indicator to check the veracity of mean shifts detected, while keeping in mind that inconsistent trends can exist at nearby stations because of local effects. Apparently, this data homogenization procedure includes some subjective analysis. However, such subjectivity is necessary because the so-called Type I and Type II errors (i.e., mistakenly reject and accept the null hypothesis) are inherent in any statistical test, which in this case means that there is always a small possibility for the statistical test to identify a change point that does not exist or fail to identify a real change point. The subjective checking of spatial consistency of trend and the time series visualization aim to reduce the inherent errors of statistical tests. Also, metadata investigation is very important here, 5of27

6 Figure 5. Long-term mean frequencies (in percentage) of blowing snow (BS) occurrence in (a) autumn, (b) winter, and (c) spring. because the distribution of the test statistic for identifying undocumented change points is very different from that for assessing documented change points (i.e., change points that are associated with documented reasons) and thus much higher critical values should be used to identify undocumented change points [cf. Wang and Feng, 2004; Wang, 2003; Lund and Reeves, 2002]. [16] Then, if the time series in question was found to have no mean shift, model (1) was used to estimate the trend b. If K > 0 points of artificial mean shifts were identified in the time series, a (K + 1)-phase regression model was fitted to the time series to estimate the trend b, as well as the size of the k-th mean shift d k =(m k+1 m k ) for k =1,2,..., K. For example, the time series shown in Figures 2a or 2b was identified to have three mean shifts (i.e., K = 3), a fourphase regression model was fitted to the time series, resulting in four different intercept values (and hence fitted lines; see the solid straight lines in Figures 2a or 2b). A student t- test was then completed to determine whether the trend b was statistically different from zero, to assess the statistical significance level of the estimated trend [von Storch and Zwiers, 1999]. [17] Since the exponential trend curves in the frequency time series are nearly linear over the period of data, the average rate of change over the period can be estimated and reexpressed in percentage of the long-term mean frequency. This was done for the time series of frequency (or sunshine amount) averaged over all 95 (or 86) stations, and over provinces or some regions that were selected because of their spatially consistent change/climate (see section 4 below). [18] The effects of artificial mean shifts on the estimate of trend are illustrated in Figure 2, and also by Hanesiak and Wang [2005]. It has been shown clearly that trends estimated from the raw data, ignoring artificial mean shifts (if any), could be very misleading. In particular, station automation was found to be especially problematic for the fog and blowing snow events, because automated stations such as AWOS (Automated Weather Observing Systems) do not report on the occurrence of fog or blowing snow event while nonoccurrence of the event (instead of a missing flag) was always recorded in the digital data archive of Environment Canada, which would lead to an extremely inaccurate estimate of trend if ignored (see Figure 2c). This is a general problem with AWOS stations. [19] The above homogeneity and trend analyses were carried out for time series of monthly log odds (for all 12 months consecutively, with the annual cycle removed) of each type of event, and for time series of monthly bright sunshine data at each station, separately. To investigate the seasonality of trends, we also estimated trends for each season separately, using seasonal averages of the monthly data that have been adjusted for detected artificial mean shifts if any (adjusted to the longest segment). The time series of annual frequency was not analyzed for trend here, because it is just a kind of average of the four seasonal mean time series; it has a much smaller sample size (and hence larger sampling variability) than the monthly time series described above, although both the annual and the monthly time series indicate the same all-season or annual trend. Also, the above monthly 6of27

7 Figure 6. Long-term mean frequencies (in percentage) of (a and b) fog and (c and d) low ceiling (LC) occurrence in the indicated season. time series has 12 data per year, and the seasonal mean time series has one data per year (four seasons in four different time series), so that the length of the seasonal mean time series is about a twelfth of the length of the relevant monthly series. Thus estimates of trends from the seasonal mean time series are generally subject to larger sampling variability, which might result in a lower spatial consistency of trends. [20] Here, we defined the four seasons as winter (January March), spring (April June), summer (July September), and autumn (October December). Note that such a definition of seasons is more convenient and allows us to use all years of data available, whereas defining December February as winter would result in one less season of data to use and hence was not chosen here. Since winter is much longer than 3 months in most areas of Canada and the climate of October December often bears substantial similarity to that of January March, this definition of seasons generally have no significant impacts on the results, although trivial impacts could result. 3. Climatology of Adverse/Fair Weather [21] Knowledge on the climatology and annual cycle of the adverse/fair weather occurrence is of general importance, and is also useful for trend analysis. Generally speaking, it is more difficult to assess changes associated with infrequent events because one has limited knowledge about them because of their infrequent occurrence. Thus it is necessary to characterize the climatology, and statistical and spatial distributions of the adverse/fair weather occurrence prior to a trend analysis, especially for the climatologically relatively infrequent types of adverse weather such as freezing precipitation or blowing snow (these have a much smaller long-term mean frequency of occurrence than does the occurrence of no-weather or fair weather conditions; see Figures 3 7). 7of27

8 Figure 7. Long-term mean frequencies (in percentage) of no-weather (NoWx) occurrence in (a) winter and (b) summer. [22] Figures 3 8 show the long-term mean frequencies of the adverse/fair weather occurrence in selected seasons. Climatologies are not shown for freezing precipitation and blowing snow in summer because these types of events rarely occur in most areas of Canada (except the Arctic region) during summer. For fog, low ceiling height, and noweather conditions, autumn and winter have a similar pattern of frequency climatology, and so do spring and summer. Thus the climatologies for autumn and spring are not shown either. The long-term mean annual cycle of frequency was also derived for each type of event at each station, which is discussed below but shown only for five selected stations (representing different climates across southern Canada). [23] As shown in Figure 3, the long-term mean frequency of freezing precipitation (FZ) occurrence is below 1% in all seasons at most locations across Canada (except a few stations in the east coast area where the long-term mean frequency of FZ in winter ranges 1 4%; see Figure 3b). It is generally below 0.1% in BC (British Columbia) interior all year around, and at 81 of the 95 stations in spring. The FZ occurrence climatologies shown here are in agreement with those shown by Stuart and Isaac [1999, see Figures 5 6 and 9]. Note that FZ occurrence in Canada has a strong seasonality (see Figures 3 and 4a). The frequency is largest in winter in the east coast and the Great Lakes area (see Figure 1a), and in both autumn and winter in south-central Canada. In the Canadian Arctic, freezing precipitation occurs more often in the transition seasons (autumn and spring) than in winter. This is because the winter air temperature in the Arctic is usually too cold for freezing precipitation to occur (in such cold air temperature, precipitation often occurs in the form of snow). Freezing precipitation occurs only when the air temperature at the time of precipitation is around the freezing point. Therefore the nature of annual cycle and climatology of FZ occurrence depends greatly on the annual cycle and climatology of air temperature and precipitating systems occurrence. [24] The occurrence of blowing snow (BS) in Canada also has a strong seasonality, which bears substantial similarity to the seasonality of freezing precipitation occurrence (see Figures 4a and 4b). The BS frequency is largest in northern, eastern, and central Canada in winter, and also in northeastern Canada (from northern Quebec to Nunavut) in autumn and spring (Figure 5). In winter, the long-term mean frequency of BS occurrence is about 8 21% at stations in Nunavut, and 11.4% at Churchill Airport (Manitoba). It is generally below 10% in other seasons and at other locations, and below 0.6% in southern Canada in the transition seasons (Figures 5a and 5c). The climatology of winter BS occurrence shown here is in very good agreement with the pattern of average annual number of days with blowing snow shown by Phillips [1990], and of the mean annual number of blowing snow events derived from the European Centre for Medium Range Forecasts Reanalysis for the period [Dery and Yau, 1999]. The nature of the annual cycle and climatology of BS occurrence depends greatly on the annual cycle and climatology of the local near-surface winds and the time, amount and state of snow available on the ground. [25] As shown in Figure 6, across Canada, fog and low ceiling (LC) occurrence is climatologically most frequent in the Atlantic provinces and the Great Lakes area, and least frequent in the region from the Prairie provinces (ALTA, SASK, and MAN) to Yukon and Northwest Territories (NWT). The long-term mean frequency of fog occurrence is about 10 44% and peaks in spring and early summer in the Atlantic provinces (NB, NS, PEI, and NFLD), while in the Great Lakes area, it is about 10 20% and peaks in autumn. The mean frequency of fog occurrence is below 5% in all seasons in the region from the Prairie provinces to Yukon and NWT, and about 5 15% in the west coast area. It mainly peaks in autumn (see Figure 4c). A similar pattern of fog occurrence climatology in Canada is also shown by Phillips [1990], with the east coast having the highest number of days with some fog. The climatology and seasonality of LC occurrence frequency are generally similar to those of fog occurrence. However, the LC frequency is higher than the corresponding fog frequency in the Arctic in all seasons, and in Prairie provinces in all seasons except 8of27

9 Figure 8. Long-term mean percentages of time with bright sunshine in (a) autumn, (b) winter, (c) spring, and (d) summer. summer (see Figure 9); while it is lower in the west coast and the Great Lakes area to the St. Lawrence valley (QUE) in all seasons but winter, and in most areas of the east coast in summer (see Figure 9b). [26] Climatologically, the highest monthly frequency of no-weather (NoWx) occurrence in Canada ranges between 68% and 95%, and the lowest monthly frequency, between 38% and 78% (see Figure 4d). The NoWx frequency peaks in the warm seasons (spring and summer) almost everywhere across Canada, with the lowest frequency seen in January or December (see Figure 4d). The latter indicates that climatologically weather (precipitation or visibility obscuration) occurrence in Canada peaks in January or December, when the no-weather conditions occur least frequently. Note that each hour is associated with weather if not with no-weather, and vise versa. Thus a larger number of hours with no-weather is always associated with a smaller number of hours with weather occurrence, and vise versa. Trends of no-weather occurrence always have an opposite sign to trends of weather occurrence. As shown in Figure 7, the NoWx frequency is generally lower in the Atlantic provinces and the Great Lakes area than in the Prairie provinces; it is lowest in northeastern Canada (Nunavut and the region around Hudson Bay) in winter and autumn (not shown). This is not surprising, because the major storm track and polar front activity regions, such as the east coast and northeastern Canada, are associated with more frequent weather, and hence less frequent no-weather occurrence. [27] As would be expected from the fact that daylight lasts much longer in the warm seasons, the amount of bright sunshine peaks in summer or late spring (June) everywhere across Canada, with the smallest amount observed during November-January (see Figure 4e). During the peak season (June August), the long-term mean monthly sunshine amount is about 30 53% in the lower Canadian Arctic (between 58 N and 65 N; note that none of the 86 stations is located north of 65 N), 24 44% in BC, 35 47% in the Prairie provinces, 33 42% in Ontario, 30 37% in Quebec, and 25 34% in the Atlantic provinces (see Figure 4e). As shown in Figure 8, the long-term mean seasonal sunshine amount is smallest in autumn (below 10% in northern 9of27

10 Figure 9. Differences between the long-term mean frequencies (in percentage) of fog and low ceiling (LC) occurrence in (a) winter and (b) summer (LC minus fog). Canada, and about 10 20% at most of the other locations; see Figure 8a). It is about 10 20% in winter, and 21 40% in spring and summer (Figures 8c and 8d). The Prairie provinces have a larger amount of bright sunshine than most of the other locations in Canada. Note that we have bright sunshine mostly during hours with no-weather, although there can be bright sunshine in some hours with weather when the weather condition lasts only for part of the hour. Thus the patterns of climatologies of bright sunshine bear substantial similarity to those of no-weather occurrence frequency. In general, the inherent relationships between the adverse and fair weather conditions are, to a great extent, manifested in the climatologies discussed above. 4. Results of Trend Analysis [28] For each station under investigation, as described in section 2, changes associated with the frequency of each of the five types of event (FZ, BS, fog, LC, and NoWx) were assessed by means of logistic regression; and linear trends were also estimated for time series of monthly/seasonal sunshine amount. The results are presented in Figures 10 15, with three levels of statistical significance (1 p) (for p 0.95, 0.80 p < 0.95, and p < 0.80). Again, trend patterns of freezing precipitation and blowing snow are not shown for summer because these events rarely occur in most areas of Canada during summer. [29] In general, the frequency of freezing precipitation occurrence has decreased in southern BC and central Alberta and Saskatchewan, while it has increased significantly in the region from northern BC to the Canadian Arctic, as well as in the region eastward of Manitoba (inclusive) and between N (Figure 10a). The increase in Manitoba is mainly seen in winter, and so is the decline in the Great Lakes area (Figures 10b 10d). In autumn, decreases are seen at most locations in southern Canada, with significant increases in the region around Hudson Bay (Figure 10b). In spring, freezing precipitation was found to have become less frequent in the Prairies but more frequent in most of the other areas (Figure 10d). As listed in Table 1, northern BC and the Arctic have the highest rates of increase in the areal mean monthly frequency, which are highly significant; while Manitoba and the Atlantic provinces (e.g., NBNS and NFLD) have experienced moderate increases of marginal significance. The high rate of decrease in southern BC ( 8%) is mainly due to the small long-term mean count of FZ occurrence (i.e., 0.07 per month) in this area. [30] As shown in Figure 11, blowing snow has become less frequent almost everywhere across Canada, especially in winter in southern Canada. Could this general decline be an artifact of some changes in the definition of blowing snow? To answer this question, we first investigated all related versions of MANOBS [Environment Canada, 1951, 1957, 1961, 1970, 1977]. We found that there were a couple of slight changes in the definition of blowing snow regarding to the degree the horizontal visibility at eye level is reduced, that is, the change from...visibility at a height of 6 feet is reduced to...visibility at eye level is generally very poor (effective 1 January 1957) [Environment Canada, 1951, 1957], and a further change to...visibility at eye level is reduced to 6 miles or less (effective 1 January 1970) [Environment Canada, 1961, 1970]. These changes could have caused temporal data discontinuities. However, no statistically significant or visible change points were identified around January 1957 and January Nevertheless, we took the two change points into account and repeated the trend analysis, which resulted in a very similar trend pattern (almost identical to Figure 11a and thus not shown). In other words, the general decline did not arise from the slight changes in the definition of blowing snow. Visual inspection of the time series also confirm the veracity of the general decline. Note that trends in surface wind speed or in the time/amount/state of snow available on the ground could be the causes for this general decline. Work on homogenization and assessment of trends in surface wind speed is in progress and results will be compared to see if trends in blowing snow occurrence and in surface wind speed are consistent physically. In addition, local effects such as a gradual increase in the amount of nearby buildings or refor- 10 of 27

11 Figure 10. Statistical significance of change in the (a) monthly, (b) autumn, (c) winter, and (d) spring frequency of precipitation occurrence. The large, medium, and small dots indicate changes of at least 5%, 5 20%, and lower than 20% significance (i.e., p 0.95, 0.80 p < 0.95, and p < 0.80). Orange dots superimposed by a plus sign indicate positive changes (i.e., increased frequency), and blue dots indicate negative changes. Open circles indicate zero trend. estation, could be a cause for the decrease in the number of recorded events. However, this would be difficult to verify, because we do not have the data needed to do so. [31] The decline of blowing snow occurrence in Canada is more extensively significant in winter than in the other seasons; it is much smaller in the east and west coast areas in autumn and spring, with significant increases in southern BC in spring (see Figures 11b 11d). The rate of decrease in the areal mean monthly frequency is highest in Alberta and Saskatchewan, and lowest in the Atlantic provinces (see Table 1). In terms of areal mean change, the decline in blowing snow occurrence is highly significant in all regions other than the Atlantic provinces (Table 1). In general, blowing snow rarely occurs in southern BC, with only 4 occurrences per 100 months (0.04/month) on average (Table 1). [32] As shown in Figure 12, trends in the frequency of fog occurrence in Canada show little seasonality. The patterns of change are all characterized by significant decreases in eastern Canada and southern BC, with significant increases in the region from southern Prairies northwestward to NWT-Yukon and northern BC. The area of decrease is a little more extensive in summer than in the other seasons, while the area of increase is least extensive in autumn (Figures 12b 12e). Differences among seasons are noticeable only in the region surrounding Hudson Bay and Labrador. The most significant changes are the increase of 17% per decade in Alberta and the decrease of about 10% per decade in the Great Lakes area; the weakest change is seen in Manitoba (see Table 1). [33] The occurrence frequency of low ceiling (LC) conditions was identified to have increased in Alberta-BC interior and the Great Lakes area, but decreased at most of the other locations (see Figure 13a). The area of increase is most extensive in winter and least extensive in summer; but the area of decrease is most extensive in summer, while the increase in the Great Lakes area is most significant in spring and least significant in autumn (Figures 13b 13e). In 11 of 27

12 Figure 11. Same as in Figure 10 but for statistical significance of change in the (a) monthly, (b) autumn, (c) winter, and (d) spring frequency of blowing snow (BS) occurrence. terms of areal mean change, Saskatchewan, Manitoba, and the Arctic were identified to have experienced the largest, significant decreases of about 5 8% per decade; while Alberta and the Great Lakes area have experienced increases of marginal significance (Table 1). [34] As shown in Figure 14, the patterns of change in the occurrence frequency of no-weather (NoWx) are basically characterized by significant increases in southern Canada, with significant decreases in the north. The area of increase is most extensive in spring and summer, but the area of decrease is most extensive in autumn (Figures 14b 14e). The most noticeable differences among the four seasons are seen in central-western Ontario: it has significant decreases in winter and autumn, with significant increases in spring and summer (Figures 14b 14e). Besides, Alberta and Saskatchewan have experienced significant increases in winter, with small decreases in autumn (Figures 14b and 14e). In terms of areal mean change, the largest increases are seen in the Great Lakes area and the Atlantic provinces, with the largest decrease in the Arctic (Table 1). Note that the rates of change here are much lower than those for the four types of adverse weather (especially FZ and BS). This is because the long-term mean frequency of no-weather occurrence is much higher (see the long-term mean monthly counts listed in parentheses in Table 1). An increase of 2 occurrences is only a 0.4% increase in a mean value of 500 (e.g., NoWx), but a 100% increase in a mean value of 2 (e.g., FZ or BS). [35] Although there are not as many stations in the Canadian Arctic that have good long-term records of bright sunshine data, and most of the 86 stations are not colocated with the 95 stations of hourly weather records (see Tables A1 and A2), the patterns of change in the monthly/ seasonal sunshine amount bear some similarity to the corresponding patterns of change in the no-weather frequency, but have lower spatial consistency (see Figure 15). The latter is mainly due to the fact that the time series of sunshine data are shorter than 50 years at most locations (see Figure 1b), and that the periods of data are more variable than the time series of hourly weather observations. Overall, the sunshine amount has decreased in southern Quebec and southern Prairies, but increased at most of the 12 of 27

13 Figure 12. Same as in Figure 10 but for statistical significance of change in the (a) monthly, (b) winter, (c) spring, (d) summer, and (e) autumn frequency of fog occurrence. 13 of 27

14 Figure 13. Same as in Figure 10 but for statistical significance of change in the (a) monthly, (b) winter, (c) spring, (d) summer, and (e) autumn frequency of low ceiling (LC) occurrence. 14 of 27

15 Figure 14. Same as in Figure 10 but for statistical significance of change in the (a) monthly, (b) winter, (c) spring, (d) summer, and (e) autumn frequency of no-weather (NoWx) occurrence. 15 of 27

16 Figure 15. Same as in Figure 10 but for statistical significance of change in the (a) monthly, (b) winter, (c) spring, (d) summer, and (e) autumn sunshine amount. 16 of 27

17 Table 1. Areal Mean Rates of Change (per Decade) Expressed in Percentage of the Long-Term Mean Monthly Occurrence Count (Given in Parentheses) Averaged Over the Stations a FZ BS Fog LC NoWx Rate (Count) p Rate (Count) p Rate (Count) p Rate (Count) p Rate (Count) p NBC 14.4% (0.18) % (0.17) % (29.5) % (28.5) % (571.8) SBC 8.0% (0.07) % (0.04) % (21.1) % (13.5) % (612.8) ALTA 2.6% (0.33) % (0.33) % (8.99) % (20.9) % (627.5) SASK 1.6% (0.62) % (1.20) % (13.5) % (31.6) % (618.5) MAN 4.7% (0.82) % (1.77) % (14.8) % (38.8) % (575.0) GLA 0.5% (0.76) % (0.85) % (77.8) % (51.9) % (518.1) NBNS 5.3% (0.79) % (0.97) % (109.3) % (130.2) % (516.3) NFLD 4.4% (1.56) % (1.77) % (74.8) % (128.2) % (498.8) Arctic 7.0% (0.49) % (2.99) % (8.50) % (30.2) % (518.0) South 3.0% (0.58) % (0.83) % (33.5) % (45.3) % (563.9) All 3.9% (0.56) % (1.07) % (27.1) % (42.6) % (557.0) a Abbreviations and number of stations are as follows: NBC, northern BC (7 stations); SBC, southern BC (7 stations) (see Figure 1a); ALTA, Alberta (9 stations); SASK, Saskatchewan (6 stations); MAN, Manitoba (6 stations); GLA, Great Lakes area (13 stations) (see Figure 1a); NBNS, New Brunswick and Nova Scotia (10 stations); NFLD, Newfoundland (excluding Labrador) (5 stations); Arctic (15 stations); South, southern Canada (80 stations); and All, all 95 stations. Changes with p are in bold (note that (1 p) is the statistical significance). other locations in southern Canada (see Figure 15a). The decrease in southern Quebec and southern Prairies is least significant in winter; it is most significant in spring in southern Quebec but in summer in southern Prairies (see Figures 15b 15e). The increase in BC is most significant in summer, with significant increases being identified at all but one station in BC (Figure 15d); it is least significant in spring. In the lower Canadian Arctic (see the Arctic stations in Figure 1b), the amount of bright sunshine seems to have increased in winter but decreased in spring and autumn. In terms of areal mean change, the amount of bright sunshine has increased significantly in BC, Ontario, New Brunswick and Nova Scotia (NBNS); but it has decreased in the lower Canadian Arctic (see Table 2). 5. Summary and Discussions [36] Climatology and trends in some adverse weather and fair weather occurrence in Canada have been characterized in this study. The occurrence frequencies of freezing precipitation (FZ), blowing snow (BS), fog, and low ceiling (LC) events were analyzed to assess adverse weather trends; while the no-weather occurrence and bright sunshine amount were analyzed to assess fair weather trends. All Canadian stations of reliable long-term (at least 30 years) records of related hourly weather and bright sunshine observations were analyzed. All the data time series were first subject to a homogenization procedure, to diminish the effects of artificial abrupt changes (if any) on the estimate of climatic trend. [37] In general, time series of sunshine amount were found to be relatively homogeneous in comparison with other elements, especially those obtained through subjective human observations, such as cloudiness or fog. This is because sunshine amounts are objective measurements from the instrument that did not undergo any change during that period. With the exception of five stations that were automated in the 1990s (see the effect of automation in section 3 above; the period after automation was not used for the final results), the frequency time series of BS occurrence are also relatively homogeneous, with only eight stations that were found to have significant change point(s). The patterns of trend estimated from the raw and adjusted BS data are almost identical (only one of the 95 time series was corrected for trend because it is inconsistent with its nearby stations because of some missing observations in the period since 1992). For the FZ, fog, LC, and NoWx events, the total number of stations that were identified (by means of statistical analysis and metadata investigation as described in section 2) to have change point(s) is 20, 37, 66, and 47, respectively. Ceiling height measurement is temporally most inhomogeneous because of the many changes in the method/instrument of observation (e.g., from without to with ceilometers or pilot balloons) and station/instrument relocation, etc. [38] The results show that the frequency of freezing precipitation has increased almost everywhere in the region north of 50 N, especially in spring and autumn. Significant decreases were only identified in southern BC, central Table 2. Areal Mean Rates of Change (per Decade) Expressed in Percentage of Long-Term Mean Sunshine Amount (Mean Amount) Averaged Over the Stations a Area BC ALTA SASK SMAN ONT SQUE NBNS Lower Arctic All Mean amount Rate 2.7% 1.1% 0.2% 1.1% 1.6% 0.5% 1.9% 1.8% 1.1% p a Station abbreviations are as follows: BC, British Columbia; ALTA, Alberta; SASK, Saskatchewan; SMAN, southern Manitoba (all but the north-most station Churchill); ONT, Ontario; SQUE, southern Quebec (all but the north-most station Inukjuak); NBNS, New Brunswick and Nova Scotia; lower Arctic (i.e., the north-most 8 stations shown in Figure 1b); and All, all 86 stations. Changes with p are in bold (note that (1 p) is the statistical significance). 17 of 27

18 Prairies, and the Great Lakes area in the cold seasons (autumn and winter), as well as in northeastern Canada in winter. On the contrary, blowing snow occurrence has decreased significantly almost everywhere across Canada, with the most significant decreases in southwestern Canada (southern BC-Saskatchewan) in winter. It was also shown that, in every season, fog occurrence has significantly increased in the region from the Prairies to southern Yukon-Northwest Territories and northern BC, but decreased in eastern Canada (east of 95 W) and southern BC. The occurrence of low ceiling conditions was identified to have become more frequent in Alberta-BC interior and the Great Lakes area, but less frequent at most of the other locations. The positive trends are most significant in Alberta-BC interior in the cold seasons, and in the Great Lakes area in spring; while a negative trend prevails across the country in summer. The frequency of no-weather was identified to have increased significantly in southern Canada, but decreased in the Canadian Arctic. The increase is most extensive in spring and summer, while the decrease is most extensive in autumn (extending to northern Ontario in winter, and further to the Prairies in autumn). Generally, the bright sunshine trends are consistent with the no-weather trends, showing significant increases in southern Canada, with some decreases in the north. [39] Comparison of the no-weather trends with the adverse weather trends above indicates that the increase of noweather occurrence (i.e., decrease of weather occurrence) in southern Canada is at least partly due to a decline in freezing precipitation and blowing snow occurrences in the Prairies-southern BC, and due to a decline in blowing snow and fog occurrences in the east coast, the Great Lakes area, and southern BC (see Table 1). While the decline in noweather occurrence (i.e., increase in weather occurrence) in northern Canada is largely due to an increase in the occurrence of freezing precipitation (and probably other types of precipitation). This agrees very well with previously reported climatic changes in Canada, namely, more frequent cyclone activity and increased precipitation amount in northern Canada due to a northward shift of the storm track [Zhang et al., 2000; Wang et al., 2006a]. In particular, the patterns of change in the freezing precipitation occurrence (see Figure 10) bear substantial similarity to the patterns of change in surface air temperature [see Zhang et al., 2000, Figures 10 and 11], with the areas of increasing temperature corresponding reasonably well to the areas of more frequent freezing precipitation occurrence. It is thus speculated that probably the temperature in the cold seasons has been rising in such a manner that there are more times during which the temperature is just right for freezing precipitation to occur (it would have been snow or other types of solid precipitation if there were no rise in temperature). [40] Using mainly the occurrence frequencies of cyclone deepening events and cyclone deepening rates, which were derived from hourly mean sea level pressure data observed at 83 Canadian stations for up to 50 years ( ), Wang et al. [2006a] report that there are significant trends in cyclone activity in Canada (especially during winter), and that cyclone activity in Canada is closely related to the North Atlantic Oscillation (NAO), the Pacific Decadal Oscillation (PDO), and the El Niño Southern Oscillation (ENSO), simultaneously or with the NAO/PDO/ENSO indices leading for one to three seasons. Close relationships could also exist between these major circulation regimes and the occurrence frequencies of the adverse or fair weather conditions analyzed in this study, because weather phenomena are closely associated with cyclone activity. Thus a preliminary analysis of the relationships was completed, using the same NAO/PDO/ENSO indices as Wang et al. [2006a]. The resulting correlation coefficients are listed in Tables B1 B5 in Appendix B, which are briefly summarized below. [41] Indeed, there exist highly significant relationships between freezing precipitation (FZ) or blowing snow (BS) occurrence in Canada and the NAO and PDO indices, especially in the transition seasons, while the ENSO-FZ or ENSO-BS relationships are generally much weaker (Table B1). In particular, more frequent freezing precipitation occurrence can be expected in spring when a strong positive NAO is observed in the previous winter or autumn; and the frequency of freezing precipitation occurrence in spring (autumn) is highly positively (negatively) correlated with PDO (Table B1). Also, a strong positive NAO in winter is closely associated with more frequent blowing snow occurrence in the following spring but less frequent in the same winter and the following autumn, and vice versa (Table B2). A strong positive NAO in spring is also closely related to more frequent blowing snow occurrence in the following autumn (Table B2). In general, PDO is correlated positively with spring frequencies, but negatively with autumn and winter frequencies of blowing snow occurrence in Canada (Table B2). The frequency of fog occurrence in Canada is more extensively significantly correlated with PDO than with NAO, especially in spring; and its correlation with ENSO is generally weak (Table B3). The spring fog-pdo correlations are positive for the region from southern BC to Saskatchewan, but generally negative for central-southeastern Canada (from MAN eastward to NFLD; Table B3). Similarly, the frequency of low ceiling (LC) conditions is also more extensively significantly correlated with PDO than with NAO; while the PDO-LC relationships are less organized spatially, and are strongest in summer but weakest in winter (Table B4). Overall, the frequency of no-weather occurrence in Canada is more closely associated with PDO than with NAO; while its correlation with ENSO is generally weak (Table B5). [42] The above relationships indicate that the observed climatic changes shown in this study could be related to changes/shifts in large-scale circulation regimes, such as NAO or PDO. In general, there is evidence indicating relationships between climatic changes in the Northern Hemisphere high latitudes and shifts in atmospheric circulation [e.g., Serreze et al., 1997, 2000; Wang et al., 2006a]. However, this area of research needs to be and is being pursued further. Works toward more detailed interpretation or attribution of these observed climatic changes are in progress. Appendix A [43] This appendix provides station and data information of the 95 stations of the hourly current weather reports (Table A1) and of the 86 stations of bright sunshine data (Table A2) analyzed in this study. 18 of 27