Temperature trends at the Mauna Loa observatory, Hawaii

. Observations at the Mauna Loa Observatory, Hawaii, established the systematic increase of anthropogenic CO 2 in the atmosphere. For the same reasons that this site provides excellent globally averaged CO 2 data, it may provide temperature data with global signiﬁcance. Here, we examine hourly temperature records, averaged annually for 1977–2006, to determine linear trends as a function of time of day. For night-time data (22:00 to 06:00 LST (local standard time)) there is a near-uniform warming of 0.040 ◦ C yr − 1 . During the day, the linear trend shows a slight cooling of − 0.014 ◦ C yr − 1 at 12:00 LST (noon). Overall, at Mauna Loa Observatory, there is a mean warming trend of 0.021 ◦ C yr − 1 . The dominance of night-time warming results in a relatively large annual decrease in the diurnal temperature range (DTR) of − 0.050 ◦ C yr − 1 over the period 1977– 2006. These trends are consistent with the observed increases in the concentrations of CO 2 and its role as a greenhouse gas (demonstrated here by ﬁrst-order radiative forcing calcula-tions), and indicate the possible relevance of the Mauna Loa temperature measurements to global warming.


Introduction
The observations since 1958 that established the systematic increase of atmospheric CO 2 (Keeling et al., 1976) were carried out at the NOAA Observatory on Mauna Loa (altitude 3397 m a.s.l. (above sea level)), Big Island, Hawaii. It has been argued (Ryan, 2001) that this is an excellent location to make atmospheric measurements because of the isolation from localized anthropogenic and continental sources and sinks. The well-mixed atmosphere at this isolated highelevation observatory has very small variations in CO 2 con-Correspondence to: B. D. Malamud (bruce.malamud@kcl.ac.uk) centrations, and the observations have been widely taken as representative of global average values (IPCC, 2007). We suggest that this single high-altitude NOAA station at Mauna Loa Observatory provides a background site for temperature which may be of similar global significance as the CO 2 data obtained there. We note however, that unlike CO 2 , there is considerably more spatial variability amongst temperature trends. While there is a general trend of global warming, there are large variations in temperature trends in various parts of the world and in some regions there is cooling. We use the high-quality hourly temperature data (NOAA, 2009) from this observatory for the time period 1977-2006 to study (a) trends in annual mean temperature data as a function of the hour of the day, and (b) trends in annual mean diurnal temperature ranges (DTR), where the DTR is the difference between the maximum and minimum temperatures in a given 24 h period.
We recognize that temperature trends are often obtained utilizing spatial averages of results from multiple stations, rather than one station. For example, Jones and Moberg (2003) utilized 5159 stations to map global values of surface temperature trends. In another example, trends for averaged stations in the Hawaiian Islands have been given by Giambelluca et al. (2008). However, we propose that using high-quality hourly temperature data obtained from this high-altitude NOAA tropical site, might be consistent with globally averaged temperature trends.
First, the 30-yr Mauna Loa NOAA observatory temperature hourly data used, including missing observations, are described (Sect. 2). Then, temperature trends for the period of study as a function of time of day are given and a mean rate of temperature warming is determined (Sect. 3). This value is compared to IPCC (2007) global temperature trends and warming rates inferred from changes in global CO 2 concentrations. Next, the trend in DTR at this observatory is derived and compared with other studies (Sect. 4). Finally, the implications of our results are discussed (Sect. 5). The Mauna Loa Observatory hourly mean air temperature measurements used here are from the NOAA Earth System Research Laboratory (ESRL) station 31 (altitude 3397 m a.s.l., 19.54 • N, 155.58 • W) and are measured in an aspirated radiation shield at 2 m above ground (NOAA, 2009). During the 30-yr period, 1 January 1997 to 31 December 2006, three different systems were used to collect temperature data (T. Mefford, Global Monitoring Division, NOAA, personal communication, 2009): (a) January 1977-December 1983(b) January 1984-October 1993(c) October 1993-December 2006. Hourly data were based on the mean of one minute samples from October 1993 to December 2006; previous to 1993, longer time periods were averaged.
Prior to analysis, seven leap days were removed and missing hourly data were gap-filled. For the period of record, 1 January 1977-31 December 2006, there were n = 262 968 hourly measurements, of which 9695 values (3.7 %) were missing. Missing values were found to be approximately uniformly distributed by year and also as a function of the hour of day. Interpolation to replace missing data was done as follows: (a) if data for the missing hour were available for that same hour within 7 days on both sides, these two values were averaged; (b) if data as described in "a" were not available, values were substituted from the subsequent year for the same hour and day. The two types of interpolation accounted for about 50 % each in terms of missing values. The longest gaps were 29 March-28 April 1984 and 14 August-15 September 1997; the first gap was caused by a volcanic eruption at Mauna Loa which severed power lines to the observatory and the second gap was caused by a major lightning strike which affected all projects at the observatory (T. Mefford, Global Monitoring Division, NOAA, personal communication, 2009).

Temperature trends
Our first objective is to determine, as a function of each hour of the day (h), the linear trends of the Mauna Loa temperature data over the 30 yr of record, 1 January 1977 to 31 December 2006. We illustrate our data analyses using hourly temperatures T measured at h = 12:00 LST (noon) and 24:00 LST (midnight). In Fig. 1, we show the daily temperatures for the period of record at these times of day. Also shown are the annual means of the daily values and the best-fit linear trend (using ordinary least-squares) to the annual values. The slope of the annual trend line for 12:00 LST has a small rate of annual cooling, dT /dt = −0.014 ± 0.014 • C yr −1 (uncertainties are ±1 s.e. (standard error) of the slope). The slope of the annual trend line for 24:00 LST has a modest rate of annual warming, (dT /dt = 0.039 ± 0.010 • C yr −1 ). Fitting a single linear trend to the 30 yr of data can be questioned because of their relatively large scatter; this is addressed in Sect. 4. In addition to scatter, another source of variability is large scale synoptic influences (Giambelluca et al., 2008) such as the Pacific Decadal Oscillation (PDO). We emphasise linear temperature trends in our analyses, so as to compare them with the linear trend in global CO 2 concentrations at the same site.
The dependence of T , the mean of all hourly temperatures at a specified time of day, h, as a function of h, is given for the entire period  in Fig. 2a. Values for T range from a maximum T = 11.0 ± 3.0 • C at h = 12:00 LST (uncertainties are ± s.d. (standard deviation) of the daily values given in Fig. 1) to a minimum T = 3.7 ± 2.2 • C at h = 05:00 LST (Fig. 2a). The dependence of the linear temperature trends for the entire period, dT /dt, as a function of the time of day, h, have values that range from a warming trend dT /dt = 0.040 ± 0.009 • C yr −1 (±1 s.e. of the slope) at h = 02:00 LST to a slight cooling trend dT /dt = −0.014 ± 0.014 • C yr −1 at h = 12:00 LST (i.e., the slope of the trend given in Fig. 1a) (Fig. 2b). This type of hour of day variability of dT /dt has been considered previously for urban data in Japan (Fujibe, 2009).
The difference between the maximum and minimum dT /dt in Fig. 2b is 0.054 • C yr −1 . At night (22:00 to 06:00 LST) there is a near uniform heating trend with values near dT /dt = 0.040 • C yr −1 . During the day, in the morning there is a reduction in the annual warming rate, with a localized minimum (and overall cooling per year) at 12:00 LST of dT /dt = −0.014 • C yr −1 , and then an increase again of the warming trend during the afternoon. It is interesting to note the reflection symmetry about a horizontal line between the shapes of the diurnal variability of the mean temperature ( We next consider whether the hourly values of the annual temperature trends (dT /dt) over the 30 yr of record have seasonal variability. In Fig. 3a we present temperature trends as a function of hour of day on a seasonal basis, with the four seasons defined as: spring (March, April, May), summer (June, July, August), fall (September, October, November), and winter (December, January, February). Winter includes January and February from year t and December from year t − 1. Also included in Fig. 3a are the annual mean values from Fig. 2b ("all months"). From Fig. 3a we observe that the seasonal variability at a given hour is on average only about 25 % of the overall daily variability, with the dependence of temperature trends on time of day very similar, in terms of the overall shape of the five curves. The rates of warming are highest in the spring and lowest in the fall and winter. Comparing the mean annual warming (cooling) rates of dT /dt as a function of hour of day for "all months" to each season, spring values of dT /dt are overall higher, summer values very similar, winter values lower, and fall values mostly lower, except for 07:00-10:00 LST when they are higher.
In order to further quantify for 1977-2006 the seasonal effects on annual mean warming (cooling) as a function of time of day, we give the number of hours per day that the temperature trends have a specified range of annual warming or cooling values. We illustrate our approach in Fig. 3b for spring (March, April, May) finding there are 1.8 + 2.4 = 4.2 h (out of 24 h) that the temperature trend is within the range 0.00 ≤ dT /dt < 0.01 • C yr −1 , i.e. for spring during the 30yr period 1977-2006, 18 % of the hours during each calendar day, on average, have zero to slight warming trends. In Fig. 3c we present the number of hours per day with dT /dt in a given temperature trend range (over 30 yr), cool- ing (dark blue to light blue over three trend ranges) to warming (light red to dark red, five trend ranges), as a function of four seasons and "all months". Darker colours indicate larger amounts of cooling (blue) or warming (red) over the 30-yr record. Summer and "all months" are very similar in the distribution of hours amongst the different trend ranges, with 17.7-17.9 h warming, on average out of the 24 h (74-75 % of the time), and 6. The mean of the Mauna Loa Observatory annual warming and cooling trends given in Fig. 2b gives an overall warming trend (dashed horizontal line) of dT /dt = 0.021 ± 0.011 • C yr −1 (± mean of error bars in Further to our Mauna Loa single instrument station (micro-climate) analysis, regional analysis is performed using NCEP (National Climate Environmental prediction) monthly reanalysis data (Kalnay et al., 1996;NOAA, 2010). Sea Surface Temperatures (SST) for the region 18.1-21.9 • N and 153.8-157.5 • W for the period 1977-2006 were analysed. This 3.8 • × 3.7 • region contains the Mauna Loa Observatory (MLO) site in Hawaii, and reflects the temporal pattern of SST in the region of Hawaii. The resultant trend (Fig. 4) of SST as a function of year, gives a gradient over the 30 yr period of d(SST)/dt = 0.018 ± 0.006 • C yr −1 (±1 s.e. of slope), consistent with our Mauna Loa mean warming trend results, dT /dt = 0.021 ± 0.011 • C yr −1 .
We now move from the local (micro-climate) and regional trends, to global trends. The mean rate of warming inferred from averaged annual global surface temperature measurements, 1980-2005, by the IPCC (2007) is dT /dt = 0.018 ± 0.005 • C yr −1 . We note the consistency of the regional and global temperature trend values. Both of these are also consistent with our local value at Mauna Loa. One interpretation of these similarities is that the Mauna Loa Observatory value is representative of the mean rate of global warming. This is in direct analogy to the widely accepted association of the mean rate of increase in the concentration of CO 2 , dC/dt = 1.6 ppmv yr −1 (ppm by volume per year) measured at the observatory (CDIAC, 2009) over the same period, with the rate of increase of CO 2 globally. An alternative interpretation is that the regional SST trend is influencing the local Mauna Loa Observatory temperature trend.
Examination of the theoretical relation between changes in global mean atmospheric CO 2 concentrations and changes in global mean atmospheric temperatures relate to the change in radiative forcing F (Myhre et al., 1998) with F in W m −2 , C the concentration of CO 2 in ppmv, and where λ is the equilibrium climate sensitivity. Studies  1) and (2) gives: with T in • C, λ in • C m 2 W −1 , and T 0 a reference temperature. At the end of Sect. 4, this equation will be used to further explore the association between temperatures and CO 2 concentrations at the Mauna Loa Observatory.

Diurnal Temperature Range (DTR)
To obtain the trend of the annual mean values of DTR at the Mauna Loa Observatory for 1977-2006, we first obtain the maximum and minimum hourly temperatures, for each 24 h period, h = 01:00 to 24:00 LST. The mean of these values over each calendar year gives the annual mean maximum, T max , and annual mean minimum, T min , temperatures (Fig. 5a). The best-fit linear slope (using ordinary least-squares) for the annual T max data for 1977-2006 is dT max /dt = −0.011 ± 0.013 • C yr −1 (±1 s.e. of the slope), which is, as expected, close to the 30-yr cooling trend of dT /dt = −0.014 ± 0.014 • C yr −1 given in Figs. 1a and 2b for 12:00 LST (noon). The best-fit linear trend for the minimum temperature is dT min /dt = 0.038 ± 0.009 • C yr −1 , which is close to the 24:00 LST (midnight) night-time warming trend given in Fig. 1b and the overall night-time warming trends given in Fig. 2b. The difference between the annual mean maximum and annual mean minimum temperatures is the annual mean DTR (illustrated for 1977 in Fig. 5a); this is equivalent to taking the mean of the daily DTR values for the year. The annual mean values of DTR are given in Fig. 5b as a function of time t, for the period 1977-2006 at the Mauna Loa Observatory. The best-fit trend using ordinary leastsquares is d(DTR)/dt = −0.050 ± 0.007 • C yr −1 (±1 s.e. of the slope). This absolute value of d(DTR)/dt is very close to the difference between the maximum and minimum dT /dt annual trends (0.054 • C yr −1 ) given in Fig. 2b as a function of hour of the day. The application of a single trend line to the 30 yr of data at the Mauna Loa Observatory can certainly be questioned because of the relatively large scatter of the annual data. To partially address this question, we divided the annual DTR data (presented in Fig. 5b) into two 15-yr periods and obtained the best-fit trends using ordinary least-squares. For 1977-1991, we obtain d(DTR)/dt = −0.064 ± 0.021 • C yr −1 (±1 s.e. of the slope) and for 1992-2006, d(DTR)/dt = −0.045 ± 0.013 • C yr −1 . As expected, decreasing the number of values in our sample from n = 30 to n = 15, corresponds to an increase of the s.e. of the slope. However, both d(DTR)/dt values are in reasonably good agreement with the 30-yr trend given above (d(DTR)/dt = −0.050 ± 0.007 • C yr −1 ).
In the DTR analyses above, the annual means, T max , T min and DTR, are based on the means of daily minimum and maximum hourly temperatures obtained from calendar days (h = 01:00 to 24:00 LST). However, in doing so, there is the possibility that an hourly minimum temperature for two successive calendar days might be from the same evening (e.g. one value is at 23:00 LST, the other at 01:00 LST). To reduce the possibility of two successive daily minimum temperatures being chosen from the "same" evening, one can also consider an observational day (vs. a calendar day), where the 24 h period considered begins in the morning (e.g. 08:00 LST) or evening (e.g. 19:00 LST). A detailed discussion of the influence of different 24 h periods chosen on daily minimum and maximum temperatures is given by Janis (2002). The analyses done in this section are therefore repeated, but using an observational day from h = 08:00 LST (of a given day) to 07:00 LST (on the next day). We find that the percent difference between individual annual mean values using calendar days (i.e. the diamonds, triangles and squares given in Fig. 5a and b) and observation days (be-ginning at 08:00 LST), to be 0.0-0.2 % for T max , 1.3-5.6 % for T min , and 0.5-1.8 % for DTR. The resultant changes to the overall trends given in Fig. 5a and b were also considered, and found to be very small, with percent changes to the slope values given in Fig. 5a and b less than 1 % (dT max /dt by 0.9 %, dT min /dt by 0.0 %, and d(DTR)/dt by 0.2 %). We therefore conclude that the choice of 24 h period, when determining annual means for T max , T min and DTR, has little effect on the overall trends shown in Fig. 5a and b, and continue to use calendar days for our analyses.
We compare our DTR results with those of other studies (Table 1), ranging from analyses done on one single station to 7000 spatially averaged stations. Grant et al. (2005) obtained an annual DTR change (≈ −0.0020 • C yr −1 , 1935-2003) for a high-altitude station at Mt Washington that is approximately an order of magnitude smaller than our Hawaii results (≈ −0.050 • C yr −1 , 1977-2006). Although both sites are high altitude, Mt Washington is influenced by a continental location and is closer to anthropogenic influences than the Mauna Loa Observatory. There are also differences in local winds, cloud cover, synoptic systems, and the period studied is twice as long. Giambelluca et al. (2008) studied records from 21 stations on Hawaii. They spatially averaged four high-altitude stations (>900 m a.s.l.) over the period 1975-2006 and found an annual DTR change ≈ −0.036 • C yr −1 , slightly less than the value we obtain. Vose et al. (2005) obtained maximum temperatures, minimum temperatures and DTR trends globally for the period 1979-2004, approximately the same period we consider. With maximum temperature data averaged over 7018 stations and minimum temperature and DTR data averaged over 6970 stations, they found dT max /dt ≈ 0.0287 • C yr −1 , dT min /dt ≈ 0.0295 • C yr −1 , and an annual DTR change ≈ −0.0001 • C yr −1 . Differences with our analysis in Hawaii are expected since most stations in the Vose et al. (2005) global study are continental and at low elevations.
The seasonal variability of dT max /dt, dT min /dt and d(DTR)/dt, is assessed for summer months (June, July, August) and winter months (December, January, February) for the 30-yr time period (Fig. 6). Also included are the annual  Table 2, along with those from equivalent analyses for fall and spring. mean values ("all months") from Fig. 5. In each case the best-fit line (using ordinary least-squares) to the data is also shown, with the resultant slopes summarized in Table 2, along with those for spring (March, April, May) and fall (September, October, November). From Table 2, it is seen that there is some seasonal variability for dT max /dt and dT min /dt trends, but very little variability in d(DTR)/dt. The lack of seasonal variability in d(DTR)/dt suggests a minimal role of local and regional influences (e.g. cloud cover, winds, synoptic systems), and does not reject the hypothesis that the decrease in DTR may be associated with the increase in global anthropogenic CO 2 . We now address how changes over the period 1977-2006 in atmospheric CO 2 relate to changes in Mauna Loa annual mean temperature (T ), annual mean maximum (T max ) and minimum (T min ) temperature (Fig. 5), and annual mean DTR (Fig. 5). Based on Eq. (3) we show (Fig. 7) the dependence of these annual mean variables as a function of 5.35 ln(C/C 0 ), where C is the annual mean concentration of CO 2 and C 0 = 334 ppmv is for 1977 (CDIAC, 2009). For each case, the best-fit line (using ordinary least-squares) is shown, along with its slope, which from Eq. (3) is the value of the equilibrium climate sensitivity λ. However, we recognise that this proportionality is directly applicable only to T , the annual mean temperature, i.e. the temperature averaged over the 24 h calendar day (and all days of the year). We obtain λ[T ] = 0.84 ± 0.45 • C m 2 W −1 (±1 s.e. of the slope), λ[T max ] = −0.48 ± 0.53 • C m 2 W −1 , λ[T min ] = 1.58 ± 0.38 • C m 2 W −1 , and λ[DTR] = −2.06 ± 0.28 • C m 2 W −1 . As previously noted in Sect. 3, our value λ[T ] = 0.84 ± 0.45 • C m 2 W −1 is in quiet good agreement with preferred global values (Gregory et al., 2002;IPCC, 2007) , can be taken as proxies for the relation between CO 2 and the respective variables T max , T min , and DTR. These three values of λ once again emphasize the difference between day-time temperature, night-time temperature and DTR trends. Based on these values, we hypothesize that the influence of CO 2 increase is primarily a night-time effect.

Discussion
The systematic increase in atmospheric CO 2 has been convincingly demonstrated by measurements made at the Mauna Loa Observatory (Keeling et al., 1976), where the temporal change of annual mean values of CO 2 concentrations 1977-2006 is well represented by a linear trend dC/dt = 1.6 ppmv yr −1 . This site provides another example of an area with minimally-varying site characteristics, such as found in (often old) cities (Parker, 2006(Parker, , 2010Jones and Lister, 2009), which have been used to detect regional to global warming trends. The micro-scale temporal consistency of the surroundings of the site, allows detection of macro-scale (i.e. regional to global) warming trends. One focus of this paper has been to consider the potential implications of the Mauna Loa temperature trends as a proxy for global effects. As discussed above, we have found (Fig. 2b) that there is an overall annual warming trend of temperatures dT /dt = 0.021 ± 0.011 • C yr −1 at this observatory for the same period. This is very close to the Hawaii regional sea surface temperature (SST) trend d(SST)/dt = 0.018 ± 0.006 • C yr −1 for the period 1977-2006 (Fig. 4), the average "preferred" value of the IPCC (2007) for the period 1980-2005 of dT /dt = 0.018 ± 0.005 • C y −1 and our inferred CO 2 trend analysis value of dT /dt = 0.019 [0.012 to 0.029] • C yr −1 . We suggest that these similar trends may tentatively imply that the Mauna Loa temperatures can be associated with global variability, at least at low latitudes.
At the Mauna Loa Observatory, we find a systematic dependence of temperature change on the time of day. During the hours of darkness, there is a near-uniform annual warming during the 30-yr period 1977-2006 of 0.040 • C yr −1 (Fig. 2b). However, during the hours of sunlight there is a systematic decrease in this warming trend during the morning, showing a slight cooling of −0.014 • C yr −1 at 12:00 LST (noon), and then a systematic increase in the afternoon (Fig. 2b). Over the 30-yr period, this dominance of nighttime warming results in a relatively large annual decrease in the diurnal temperature range (DTR) of −0.050 • C yr −1 (Fig. 5b). Newman et al. (2010) have studied the record breaking statistics of the daily maximum and minimum temperatures in the same 30-yr Mauna Loa, Hawaii data that we use here (for 1977-2006). They consider maximum and minimum temperatures on a daily basis. For the minimum temperatures, they find an excess of the number of record setting high temperatures compared to the number of record setting low temperatures, while for maximum temperatures the numbers of record setting high temperatures and record setting low temperatures are approximately equal. These results are consistent with the moderate night-time warming and small day-time cooling reported here.
In this paper we have discussed the observed temperature trends at Mauna Loa in the context of global changes in CO 2 . There are other potential causes for these temperature trends, including local changes in wind velocity and cloud cover, and variations in the occurrences and intensity of regional synoptic systems. However, because of feedbacks, it is hard to separate cause from effect. For example, we would expect changes in temperature to impact and be impacted by cloud cover, wind, etc., on the micro-scale and indeed all scales. Doing a definitive study to separate cause and effect would require significant physically-based modelling, which is outside of this study. A principal focus of this paper has been the systematic downward trend in DTR. Environmental and seasonal variations in DTR have been discussed by several authors (e.g. Leathers et al., 1998;Scheitlin and Dixon, 2010). As shown in Fig. 6b, there is very little seasonal influence on the DTR trends at Mauna Loa.
Our basic hypothesis is that a large part of the temperature and DTR trends at Mauna Loa can be attributed to changes in CO 2 . At night, longwave radiation and turbulent sensible heat fluxes dominate heat loss. Increasing presence of green house gases will result in enhanced reradiation back towards the surface and hence warming nocturnal temperatures. During the day time, shortwave radiation dominates, particularly in tropical regions. It would be expected that the role of green house gases would be greater in the early morning before significant heating enhances boundary layer depth. At the end of the day, the boundary layer collapses. A possible explanation for the middle of the day cooling is that the enhanced surface heating is actually resulting in greater mixing and therefore a decrease in the near-surface green house gas concentration which would reduce incoming longwave radiation. These trends are consistent with the observed increases in the concentrations of CO 2 and its role as a greenhouse gas, and indicate the possible relevance of the Mauna Loa temperature measurements to global warming.
Reference temperatures over specific 30-yr periods are referred to as temperature normals. We suggest that the hourly temperature trends given in Fig. 2b, and the DTR trend given in Fig. 5b, for the 30-yr period at Mauna Loa, could be considered temperature trend and DTR trend normals, potentially representative of global values. Based on our results, we suggest that both temperature and DTR trends should be a part of studies of global warming. Specifically global climate model studies of global change should yield changes in DTR consistent with observations.