SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[1] The homogenized U.S. Historical Climatology Network (HCN) data set contains several statistical adjustments. One of the adjustments directly reflects the effect of instrument changes that occurred in the 1980s. About sixty percent of the U.S. HCN stations were adjusted to reflect this instrument change by use of separate constants applied universally to the monthly average maximum and minimum temperatures regardless of month or location. To test this adjustment, this paper reexamines the effect of instrument change in HCN using available observations. Our results indicate that the magnitudes of bias due to the instrument change at individual stations range from less than −1.0°C to over +1.0°C and some stations show no statistical discontinuities associated with instrument changes while others show a discontinuity for either maximum or minimum but not for both. Therefore, the universal constants to adjust for instrument change in the HCN are not appropriate.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[2] Scientists realize that not all thermometers will produce readings of equal uncertainty, and even identical thermometers might produce different readings when housed in various shields depending on the characteristics of the shields and features of the local microclimate. Therefore, changes in temperature instrumentation along with a nominal relocation from one type to another can create a warming or cooling bias in surface air-temperature time series. A significant instrument change in surface temperature observations took place in the U.S. cooperative observer network in the mid- and late-1980s. The traditional liquid-in-glass thermometer housed in a cotton-region-shelter (CRS) was replaced by a Maximum-Minimum Temperature System (MMTS) comprised of a thermistor housed in a cylindrical plastic radiation shield. Over sixty percent of about 5,000 cooperative stations now employ the MMTS to measure surface temperatures. Due to this important instrument change during the last century and the need for homogeneous temperature series, Quayle et al. [1991] concluded that a constant cooling bias (−0.4°C) in the MMTS for monthly mean maximum temperatures and a constant warming bias (+0.3°C) in the MMTS for monthly mean minimum temperatures was present in the data set. By directly applying these findings, starting from 1996, these two-constant-bias adjustments were made to the MMTS time series contained in the monthly U.S. Historical Climatology Network (HCN) data set [Easterling et al., 1996]. Our intent in this study is to reexamine the effects of the CRS to MMTS transition, by using a relatively longer time series coupled with metadata, to clarify the context of this instrument discontinuity.

2. Background

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[3] Instrument changes associated with surface temperature observations may involve more than the change in temperature sensor and shield. There may also be changes in observation time, relative position to site surroundings, site location (elevation and distance), and local microclimate of site. Due to the need for surface temperature bias adjustment, Quayle et al. [1991, hereinafter referred to as Quayle] selected 424 MMTS stations (stations that switched from CRS to MMTS) as well as 675 CRS stations (no switch but, instead remained CRS) from the cooperative network in the contiguous U.S. Quayle selected stations, based on the metadata, for which there were no major relocations and no changes of observing times for either the CRS or MMTS stations. A difference time series between MMTS and CRS [MMTS-CRS] for each MMTS station series and the associated CRS series was created in two ways: (a) the CRS station with the highest correlation to the MMTS series; and (b) a weighted mean of the five CRS stations with the highest correlations to the MMTS series. The weight factors were correlation coefficients calculated from annual mean temperatures during the pre-MMTS era (1960–1979). Both methods in Quayle's study produced nearly the same results on an average [MMTS-CRS] time series, that is, a cooling step change of 0.4°C and a warming step change of 0.3°C for monthly mean maximum and minimum temperatures, respectively. These two constant biases have been universally applied for bias adjustments of instrument change in every current MMTS station in the monthly U.S. HCN data set.

[4] However, Pielke et al. [2002], stating that it may not be appropriate to use a network-wide Quayle adjustment factor for any given station, decided to use the unadjusted data from U.S. HCN to evaluate regional and local temperature trends. Recent studies [Davey and Pielke, 2005; Mahmood et al., 2006] expressed concerns about the changes of microclimate exposures of stations in the CRS and MMTS stations. They illustrated a few HCN stations with poor siting and pointed out that such sites are not at all representative of their surrounding region. The site microclimate effects on temperature observations occurred whether using annual HCN data or daily HCN data [Balling and Roy, 2004; Robeson and Doty, 2005]. On the other hand, Peterson et al. [1998] in a comprehensive review article stated that the MMTS universal-constant adjustment in HCN is just a regional average; the exact effect at individual stations may vary somewhat depending on local environmental or climate factors such as the amount of direct sunlight on the shelter, and the Quayle adjustment for the transition from liquid-in-glass thermometer to the MMTS should be reevaluated [Peterson, 2003].

3. Data Source and Analysis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[5] To reexamine the effects of CRS to MMTS transition, the 1221 station series in the U.S. HCN data set are targeted for analysis. Similar to the Quayle method, we selected two groups of time series (CRS and MMTS). Due to the fact that most of MMTS stations were installed during the 1980s, our selection of both CRS and MMTS stations were limited to those that had no station moves, no instrument height changes, and no other type instruments used in the period from 1970 to 2003. The data we used were from the HCN data set that had been adjusted for time of observation (TOB) bias, which bias is related to the time of observation, site location, and observational season. We chose data adjusted for TOB bias because data not adjusted for TOB effects could seriously contaminate the effort to isolate instrument change effects. In addition, the MMTS stations were confined to stations that had a length of monthly time series on each side of the MMTS transition more than 171 months. Compared to the 43-month post transition data set used in Quayle's study, our relatively longer time series made it possible for us to observe the variations of MMTS bias over a longer time period. Any missing months in our selected data set are retained as missing in our entire analysis. We found 163 CRS and 116 MMTS stations from the 1221 U.S. HCN stations that met our criteria. These stations are well distributed across the contiguous U.S (Figure 1).

image

Figure 1. Average [MMTS - CRS] time series of aggregated monthly mean and standard deviation (STD) for (a) maximum and (b) minimum temperatures in the selected 116 MMTS stations across the contiguous United States. The month 1 is the month when the MMTS was commissioned. Embedded map shows 163 CRS (white circle) and 116 MMTS stations (black circle).

Download figure to PowerPoint

[6] In the Quayle method, it should be noted that either the MMTS paired with the single best correlated CRS or with the weighted mean of the five most highly correlated CRS stations achieved the same results. Thus, we used the weighted mean approach for interpolating the CRS time series and if there were less than five highly correlated nearby CRS stations we simply used the available stations but at least three for the weighted mean interpolation for the MMTS station. Only CRS stations with greater than 0.7 correlation coefficients (Quayle used 0.6) and paired distance less than 400 km separation were used in our analysis.

[7] In order to increase the robustness of our analysis, not only is the data analysis method used in this study the same as the Quayle method for reproducing the effect of instrument change, but also we used two statistical temperature homogeneity adjustment techniques: (a) multiple linear regression (MLR) [Vincent, 1998]; and (b) standard normal homogeneity test (SNHT) [Alexandersson and Moberg, 1997]. Both MLR and SNHT were used to identify discontinuity positions in the MMTS series that match the dates when CRS was replaced by MMTS, as well as to quantify the discontinuity magnitudes of inhomogeneous MMTS time series. Note that both MLR and SNHT techniques are considered as more reliable techniques for the identification of homogenous series and the detection of single most probable discontinuity in temperature time series [DeGaetano, 2006]. The reference series used for both MLR and SNHT method were generated using an anomaly-correlation-coefficient-weighted average [Menne and Williams, 2005], which stems from SNHT studies [Alexandersson and Moberg, 1997]. The most probable discontinuity in the series was considered in this study if metadata confirmed the change.

4. Instrument Change Bias and Inhomogeneous Attributes

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[8] The maximum or minimum temperature bias, defined as a difference between CRS and MMTS [CRS-MMTS] observations, showed step changes which are quite similar to Quayle's results in the sense of an aggregated average (Figure 1). The striking discontinuity was also identified for monthly maximum and minimum temperatures starting at the time of instrument change. Note that from the month of MMTS installation throughout the year 2003 the mean [CRS-MMTS] series didn't show obvious variations although the time series length was much longer than that in Quayle's work. The station average temperature biases caused by instrument changes were 0.57 and −0.35°C for monthly maximum and minimum temperatures (Figure 1), which are slightly larger magnitudes than found in Quayle's results. However, the magnitudes of standard deviation (STD, calculated each year for the CRS-MMTS differences from all stations) for both maximum and minimum temperature were 0.83°C on average in the process of so-called Quayle's interpolation (Figure 1). This result suggests that the aggregated monthly average biases, 0.57 and −0.35°C in this study, are only appropriate for representing an average of 116 MMTS stations and are inappropriate for removing instrument change bias at any individual stations. Figure 2 clearly demonstrates that the transition from the CRS to MMTS caused positive biases at some sites and negative bias at others (Figure 2a vs Figure 2b and Figure 2c vs Figure 2d). The [CRS-MMTS] series in Figure 2 does not support a constant adjustment factor as is suggested by Figure 1 and the increase or decrease of the difference [CRS-MMTS] with time obviously exists in Figure 2. In addition, Figure 2 indicates that relatively large biases occurred before the MMTS installation, which implies that Quayle's method might not be applicable for comparing the CRS and MMTS stations over various spatial scales. To contrast with the annual correlation coefficient method used in Quayle's study, we used 12 monthly time series (i.e., separate January, February… and December series) to calculate the correlation coefficients and then interpolated the MMTS temperatures on a monthly basis. Each box and whisker in Figure 3 represents station variation among 116 stations and monthly variation in a year. The result indicates that the station/spatial variations (Figures 3b and 3d) were much larger than the monthly variations (Figures 3a and 3c). This result suggests that working with monthly and station-specific bias adjustments is important if local and regional trends are to be assessed.

image

Figure 2. Average of the five most positive and five most negative temperature difference [MMTS-CRS] series taken from Figure 1. (a) For the most negative and (b) for the most positive maximum temperatures; and (c) for the most positive and (d) for the most negative minimum temperatures.

Download figure to PowerPoint

image

Figure 3. Boxplots showing the 30-year temperature difference series [MMTS-CRS] in116 MMTS stations for (a) 12-month variations of maximum temperatures, (b) 116-station variations of maximum temperatures, (c) 12-month variations of minimum temperatures, and (d) 116-station variations of minimum temperatures. The box indicates the lower quartile, median and upper quartile values. The whiskers extend to 1.5 times the interquartile range and outliers are beyond the ends of the whiskers.

Download figure to PowerPoint

[9] To further examine the attributes of the instrument change in the MMTS series, both the MLR and SNHT consistently identified that there were 34 and 24 inhomogeneous MMTS series respectively, for the maximum and minimum temperatures, and for which the discontinuity positions matched exactly with the actual MMTS replacement dates. The magnitudes of the discontinuity in these matched inhomogeneous stations were varied distinctly from station to station and favorably consistent with results given by Quayle's method in this study. On the other hand, it is interesting that 25 and 27 MMTS series were commonly identified by the MLR and SNHT as homogenous for the maximum and minimum temperatures, respectively. Note that about half of 116 MMTS stations were inhomogeneous and their most probable discontinuity positions were not matched with the MMTS installations. These unmatched inhomogeneous MMTS stations might result from other instrument-related changes that occurred in the series but the metadata didn't support these change points. To illustrate the variation of [MMTS-CRS] series in a combination of maximum and minimum temperature series versus identified homogeneous and inhomogeneous series, Figure 4 presents average [MMTS-CRS] series of homogeneous and inhomogeneous MMTS series for maximum and minimum temperatures commonly obtained from the metadata supported MLR and SNHT methods. For stations with a homogeneous maximum series (Figure 4a), its corresponding average minimum series (Figure 4b) is inhomogeneous. Likewise for stations with a homogeneous minimum series (Figure 4f) the corresponding average maximum series (Figure 4e) is inhomogeneous. The inhomogeneous maximum series (Figure 4c) and the inhomogeneous minimum series (Figure 4h) also have corresponding series that are inhomogeneous (Figures 4d and 4g). Examination of Figure 4 reveals that instrument changes may cause quite different effects on maximum and minimum temperatures, which might be only partially due to instrument change itself and more likely due to site microclimate changes and anthropogenic effects.

image

Figure 4. Average difference [MMTS-CRS] for maximum and minimum temperature where homogeneous and inhomogeneous MMTS series were commonly identified by both MLR and SNHT methods. The left column is the maximum temperatures and the right column the minimum temperatures. (a and b) Stations with homogenous maximum temperatures; (c and d) inhomogeneous maximum temperatures; (e and f) homogenous minimum temperatures; and (g and h) inhomogeneous minimum temperatures.

Download figure to PowerPoint

5. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[10] This study does not address the actual bias adjustments for specific stations at the transition of CRS to MMTS. However, it does indicate that a regional average [MMTS-CRS] constant adjustment used in the U.S. HCN is not applicable for the individual stations when local or sub-regional trend analysis is to be undertaken. For example, gridded temperature values or local area-averages of temperature might be unresolvedly contaminated by Quayle's constant adjustments in the monthly U.S. HCN data set or any global or regional surface temperature data sets including Quayle's MMTS adjustments. It is clear that future attempts to remove bias should tackle this adjustment station by station. Our study demonstrates that some MMTS stations require an adjustment of more than one degree Celsius for either warming or cooling biases. These biases are not solely caused by the change in instrumentation but may reflect some important unknown or undocumented changes such as undocumented station relocations and siting microclimate changes (e.g., buildings, site obstacles, and traffic roads). We suggest that the transition from CRS to MMTS was often accompanied by geographically small but micrometeorologically significant changes in site location which in some cases enhanced the bias and sometimes canceled the bias depending on the specific microclimate variations in the vicinity of the station. To robustly use the statistical methods, full use of available metadata and potential metadata are critical to accurately quantify the inhomogeneity for maximum or minimum temperatures especially for studying changes of extreme climate. Furthermore, our study showed different discontinuity attributes for maximum and minimum temperature series. Thus, the constant Quayle surface temperature adjustment factors are not applicable for extreme temperature bias adjustments.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References

[11] Partial support of this work was provided by NOAA, Office of Global Program, Climate Change Data and Detection Element. Authors would like to thank Thomas C. Peterson of National Climatic Data Center (NCDC) for insightful comments in this work, and Claude Williams of the NCDC for providing the station metadata.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Background
  5. 3. Data Source and Analysis
  6. 4. Instrument Change Bias and Inhomogeneous Attributes
  7. 5. Conclusion
  8. Acknowledgments
  9. References