SEARCH

SEARCH BY CITATION

Keywords:

  • Normal temperature;
  • land cover change;
  • urban heat island

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[1] Quantification of the effects of land use/land cover (LULC) changes on proximal measurements of near-surface air temperature is crucial to a better understanding of natural and anthropogenically induced climate change. In this study, data from stations utilized in deriving U.S. climatological temperature normals were analyzed in conjunction with NCEP-NCAR 50-Year Reanalysis (NNR) estimates and highly accurate LULC change maps in order to isolate the effects of LULC change from other climatological factors. While the “Normals” temperatures exhibited considerable warming in both minima and maxima, the NNR data revealed that the majority of the warming of maximum temperatures was not due to nearby LULC change. Warming of minimum temperatures was roughly evenly split between the effects of LULC change and other influences. Furthermore, the effects of LULC change varied considerably depending upon the particular type of land cover conversion that occurred. Urbanization, in particular, was found to result in warming of minima and maxima, while some LULC conversions that might be expected to have significantly altered nearby temperatures (e.g., clear-cutting of forests) did not.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[2] Globally averaged near-surface temperature has been shown to have increased substantially over the past several decades, and much of this warming is attributable to anthropogenically induced increases in atmospheric greenhouse gases, especially carbon dioxide [see, e.g., Trenberth et al., 2007]. However, warming due to increased greenhouse gas concentrations does not preclude other factors from either contributing to or moderating the observed rise in near-surface temperatures. One such potential factor is changes in land use/land cover (LULC) near stations where these temperatures are measured.

[3] It has been recognized for some time now that urbanization near meteorological observing stations could impact their temperature records through changes in local albedo, surface heat capacity, and partitioning of net radiative forcing into latent and sensible heat fluxes. Attempts have been made to both quantify this urban heat island (UHI) effect, and to correct for it in climatological records [Karl and Williams, 1987; Easterling et al., 1996]. Quantification efforts mostly have relied upon proxy measures of the degree of urbanization near a station (e.g., Census Bureau population [Karl et al., 1988]; satellite measures of nighttime lights [Gallo et al., 1999]) to facilitate comparisons between “urban” and “rural” stations. Such application of proxy data can be problematic in that population or artificial lighting are not always good indicators of the degree to which nearby land has been “urbanized.” That is, the factors that are most likely to affect near-surface temperature as a result of urbanization (e.g., increased areas of impervious material altering the Bowen ratio, albedo changes, etc.) may be correlated with population or nighttime lights, but neither of these are direct measures of the land cover changes of interest. It might be expected that conversion of pastureland to urban use would result in different near-surface temperature changes than would occur if a forested area were urbanized.

[4] Furthermore, urbanization represents only a single type of anthropogenically induced LULC change (for the areas studied in this research, urbanization accounted for 26% of all LULC change). The effects of other types of LULC change on near-surface measurements of temperature have received comparatively little attention despite the fact that they collectively comprise the majority of LULC change.

[5] The advent of high-resolution, multispectral satellite imagery has made possible detailed analyses of LULC over geographically large regions. For example, the National Land Cover Database (NLCD [Homer et al., 2004]) is providing high-resolution LULC determinations across the conterminous U.S. through the use of unsupervised classification techniques. Continuation of such analyses into the future will provide a much needed method of monitoring LULC change near meteorological observation stations, as has been recommended previously [Gallo et al., 1996; Pielke et al., 2007]. Unfortunately, these LULC data sets are typically contemporary in nature and provide little or no information regarding historical changes in LULC. A different approach has been taken by the Land Cover Trends Project [Loveland et al., 2002] in which randomly selected blocks of land are interpreted by human analysts utilizing historical satellite and aerial imagery. This methodology results in accurate LULC determination for five distinct points in time over a nearly 30-year period, although without the “wall-to-wall” coverage provided by the NLCD.

[6] Gallo et al. [1996] found that differences in LULC can manifest themselves in temperature observations recorded at stations up to 10 km distant. On the basis of this, Hale et al. [2006] examined historical trends of minimum and maximum temperatures at NOAA stations used for determining climate normals (“Normals stations” hereafter, consisting of both first order and Cooperative Network stations) that were within 10 km of Land Cover Trends Project sample blocks. Comparisons of temperature trends before and after nearby LULC change revealed significant differences. Relatively few trends before periods of greatest LULC change were of a statistically significant magnitude, and those that were tended to be equally divided between warming and cooling trends. Following the major LULC change, however, significant temperature trends were much more common, and nearly all of these were warming trends.

[7] While the correlation between LULC change and trends in temperature found by Hale et al. [2006] were highly suggestive of a LULC influence on the temperature records at Normals stations, the authors were careful to point out that the correlation “does not necessarily imply that the LULC changes are the causative factor.” Clearly, post-change temperature trends encompass a later time period than pre-change trends. Thus the observed correlation could be explained by increased warming in the 1990s relative to the 1970s.

[8] Unfortunately, establishing a causative link between LULC and temperature trends is not a straightforward task. Ideally, one could compare nearby pairs of stations for which one station underwent considerable LULC change while the other experienced no change. In practice, however, this approach is marred by several difficulties. First, relatively few stations (11% of the stations examined by Hale et al. [2006]) experienced no nearby LULC change during the 1973–2000 time period. Second, of those stations for which no change occurred in the intersection between a 10-km radius surrounding the station and a nearby Trends Project sample block, the intersection area is often quite small (i.e., the station is nearly 10 km distant from the sample block). Thus LULC change may have occurred near the station, just not in the intersection area for which LULC was determined. Finally, stations with no nearby LULC change tend to be spatially clumped. For example, the Central Basin and Range and Mojave Basin and Range ecoregions, adjacent areas located primarily in Nevada and southern California and comprised largely of arid regions with little human influence, collectively contain nearly one quarter of the stations with no nearby LULC change. An average of all stations with no nearby LULC change thereby would not be representative of areas with differing regional climatology.

[9] Clearly, a different approach for establishing causation between the LULC changes and temperature trends reported by Hale et al. [2006] is needed. Kalnay and Cai [2003] established a unique method of isolating the effects of LULC on temperature through comparing surface observations with data from the NCEP-NCAR 50-year Reanalysis (NNR [Kistler et al., 2001]). They report that since the NNR does not assimilate surface observations directly, but rather estimates surface temperature from rawinsonde and satellite-derived data, the NNR estimates are insensitive to local LULC. In examining nearly 2000 stations, they found that the NNR estimates showed strong correlation with surface observations at stations with an elevation less than 500 m (high-elevation stations showed weaker correlation and were not used in the analysis) and synoptic and mesoscale patterns were well represented. The difference in mean temperature trends between the surface observations and NNR data was found to be 0.35°C century−1, which the authors believed to be primarily attributable to LULC influences.

[10] Although Trenberth [2004] argued that the use of the NNR data in this manner by Kalnay and Cai [2003] was flawed since direct observations of greenhouse gas (particularly CO2) concentrations were not assimilated into the NNR model, Cai and Kalnay [2004] subsequently showed that the NNR data indeed capture the warming caused by increasing greenhouse gases since atmospheric temperatures are utilized in the model. They state that, “the reanalysis can capture essentially the full strength of climate trends caused by the increase in greenhouse gases.” Another criticism of the Kalnay and Cai [2003] methodology was their use of surface observations that had not been adjusted for inhomogeneties arising from instrument changes, time of observation biases, or site moves [Vose et al., 2004]. Cai and Kalnay [2004] acknowledge this fact, but counter that inclusion of the inhomogeneity adjustments merely results in a larger trend difference attributable to LULC (0.147°C decade−1 rather than 0.035°C decade−1), which is comparable to that reported by Kukla et al. [1986].

[11] While introducing a novel approach to isolating LULC effects on near-surface temperature from other climatic effects, the Kalnay and Cai [2003] study examined the issue only in broad terms, estimating the total effect of all LULC changes over a 40-year period near the stations utilized. They were not able to discern the individual effects of urbanization, development or reversion of agricultural lands, deforestation and reforestation, etc. These individual effects can be estimated, however, through utilization of the Land Cover Trends Project analyses in conjunction with comparisons of surface temperature observations and corresponding NNR data.

[12] The objective of this study is to differentiate from other factors the influence of land cover change on observed temperature differences at U.S. Climatological Normals stations.

2. Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[13] To identify U.S. Climate Normals stations potentially affected by nearby LULC change, LULC images produced by the Land Cover Trends Project [Loveland et al., 2002] were utilized. Such images were thus far available for 31 of the 84 Level III ecoregions of the U.S. originally described by Omernik [1987] (Figure 1). In these images, LULC is classified into one of 11 categories at a 60-m resolution for the nominal years 1973, 1980, 1986, 1992, and 2000.

image

Figure 1. Ecoregions of the United States depicting the 31 thus far analyzed by the Land Cover Trends Project (shaded) and the 94 Normals stations utilized in this study (solid circles).

Download figure to PowerPoint

[14] Based on the findings of Gallo et al. [1996] that near-surface temperatures could be influenced by surrounding LULC up to 10 km distant, Normals stations within this range of Land Cover Trends Project sample blocks were selected. For each station, changes in LULC classifications between the above-described nominal years were examined to determine the period of greatest change, defined as the period in which the most change of a single type (e.g., cropland to urban) occurred, and the type of conversion that took place. The set of 526 stations within 10 km of Land Cover Trends Project sample blocks was then further limited to stations that: (1) had at least 2 years worth of data before and after the period of greatest LULC change, so as to facilitate statistically meaningful comparisons; and (2) were sited at less than 500 m elevation, since Kalnay and Cai [2003] found that NNR data for higher elevation stations suffered from errors due to model heights differing from actual station elevation. Ninety-four stations were found meeting these criteria, and were subsequently analyzed.

[15] Monthly minimum and maximum temperature data were obtained for the above-described Normals stations from the National Climatic Data Center [NCDC, 2002]. These data have been corrected for time of observation biases [Karl et al., 1986] and discontinuities arising from station or instrumentation changes [Peterson and Easterling, 1994; Easterling and Peterson, 1995]. The data used for this study were not adjusted for nearby urbanization, as such correction might mask LULC effects of interest. Anomalies of the monthly maxima and minima were computed for each station by subtracting that station's 1979–2000 (to correspond to the NNR data below) monthly average maximum or minimum.

[16] NNR estimates of near-surface monthly maximum and minimum temperature were linearly interpolated to individual Normals station locations. Anomalies of the NNR data were then produced based on 1979–2000 averages of the NNR estimates. By comparing anomalies of the Normals and NNR data rather than the raw data, systematic biases in either data set relative to the other are substantially eliminated.

[17] Means of maximum and minimum temperature for both Normals and NNR then were computed for the intervals preceding and following the period of greatest LULC change for each station. These intervals were thus of varying length, extending from the beginning of 1979 to the beginning of the period of greatest LULC change (pre-change), or from the end of the period of greatest LULC change to the end of 2000 (post-change). Differences between pre-LULC change and post-change means in the Normals data may arise from the LULC change itself, but may also include other climatic influences (e.g., warming due to increased atmospheric greenhouse gas concentration). On the other hand, pre-change versus post-change differences in the corresponding NNR data should not include LULC change effects, and reflect only the other climatic influences. Thus a comparison of the Normals pre- versus post-change differences to the NNR pre- versus post-change differences isolates the LULC change effects experienced by the Normals stations from other factors. Such “difference of differences” were computed for each station, and then aggregated by LULC change type. Paired Student's t-tests were performed on each change type, as well as for all stations as a whole. In some cases, Normals stations were located near enough to each other that NNR data from a common grid point affected the bilinear interpolation of temperatures at each location. In these instances, the independence of the data was reduced. Such reductions were accounted for in determining the appropriate degrees of freedom utilized in calculating the statistical significance of differences, and also manifest themselves in noninteger sample size values in subsequent figures.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[18] Figure 2 shows the difference, averaged by LULC change type, between post-LULC change mean minimum temperatures from the Normals stations and their respective pre-LULC change means. Also displayed is the significance of those mean differences, through both 90% confidence intervals and shading of bars for which the difference is significant at a confidence level α = 0.10. All conversion types showed warmer minimum temperatures following the period of greatest LULC change compared to the period before change. For the majority of LULC conversion types for which the significance of the pre- versus post-change difference could be computed (i.e., when the number of stations n ≥ 2), the difference was significant at α = 0.10. The difference for all 94 stations as a whole was also highly significant. Normals maximum temperatures (Figure 3) show similar post-change warming, with the exception of three conversion types (each represented by only a single station) which had cooler post-change temperatures.

image

Figure 2. Postchange minus pre-change mean Normals minimum temperature anomalies for various LULC conversion types. Numbers below the bars express the effective number of stations undergoing each conversion type, and error bars represent 90% confidence intervals. Conversions for which the post-change minus pre-change difference is significant at α = 0.10 are highlighted with shaded bars.

Download figure to PowerPoint

image

Figure 3. As for Figure 2, but for Normals maximum temperature anomalies.

Download figure to PowerPoint

[19] As with the Normals observations, minimum temperatures from the NNR data set (Figure 4) exhibit post-LULC change warmth compared to pre-change values. However, the magnitudes of the differences are smaller for every LULC conversion type, and there are fewer conversion types for which the pre- versus post-change difference is statistically significant. NNR maximum temperatures exhibit warmer post-change values as well (Figure 5), but in this case the pre- versus post-change differences are more similar to those found in the Normals observations.

image

Figure 4. As for Figure 2, but for NNR minimum temperature anomalies.

Download figure to PowerPoint

image

Figure 5. As for Figure 2, but for NNR maximum temperature anomalies.

Download figure to PowerPoint

[20] Differences between Normals and NNR differences [i.e., (Normals post-change – Normals pre-change) – (NNR post-change – NNR pre-change)] in minimum temperatures are shown in Figure 6. Again, the differences in the Normals data are uniformly larger than those in the NNR data, and they are significantly larger for LULC change as a whole as well as two specific LULC conversion types: urbanization of crop/pastureland and urbanization of forested land. Figure 7 displays the more mixed results for differences between maximum temperature differences in the Normals and NNR data. In some cases, the Normals temperatures warmed more (or cooled less) than the NNR temperatures, but in other cases the opposite is true. Again, two land cover conversion types show statistically significant differences between Normals and NNR differences: urbanization of forested land and reforestation of mechanically disturbed land (classification as “mechanically disturbed” is usually indicative of forest clear-cutting).

image

Figure 6. Differences between Normals and NNR pre- versus post-change differences in minimum temperature anomalies. Numbers below the bars express the effective number of stations undergoing each conversion type, and error bars represent 90% confidence intervals. Conversions for which the Normals-NNR “difference of differences” is significant at α = 0.10 are highlighted with shaded bars.

Download figure to PowerPoint

image

Figure 7. As for Figure 6, but for maximum temperature anomalies.

Download figure to PowerPoint

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[21] The pre- versus post-LULC change temperature differences in the Normals data are consistent with analyses presented by Hale et al. [2006] and Pielke et al. [2007] that were based on a superset of the stations utilized in this study (i.e., stations for those studies were not limited by availability of NNR data or restricted to elevations less than 500 m). The previous studies found that post-LULC change trends in minimum temperature were much more likely to express significant warming compared to pre-LULC change trends. This also was found to be true for trends in maximum temperature.

[22] It is unclear from the Normals data alone, however, whether the post-change warmer temperatures are due solely to LULC change. In fact, the prevalence of post-change warmer temperatures in the NNR data indicate that at least some of the observed warming in the Normals data is due to large-scale influences that also affect the NNR data, such as warming resulting from increasing greenhouse gas concentrations. This hypothesis is supported by the relatively uniform warming seen across the various LULC change types in the NNR data compared to the Normals data. The standard deviation in pre- versus post-change differences as a function of LULC conversion type in the NNR data were 0.16°C and 0.19°C for minimum and maximum temperatures, respectively. These small standard deviations indicate that the amount of warming did not vary much regardless of the type of LULC change that took place, as would be expected in a data set insensitive to LULC. In contrast, the Normals data exhibited standard deviations of 0.45°C and 0.29°C for minimum and maximum temperature differences. These larger standard deviations reflect the inclusion of LULC effects in the Normals data.

[23] Averaged over all LULC conversion types, the NNR minimum temperatures were 0.298°C warmer following LULC change relative to before change. Maximum temperatures warmed by a similar amount, 0.282°C. These values are representative of the warming that would be expected to have occurred in the Normals data in the absence of LULC change, with changes significantly different from these values indicating LULC change influences. Averaged over all LULC conversion types, non-LULC warming (based on the NNR data) represented 54% of the total warming (0.552°C) of minimum temperatures based on the Normals data, and 87% of the total warming (0.324°C) of maximum temperatures.

[24] For all types of LULC change as a whole, minimum temperatures warmed significantly more than would be expected from the above NNR values had no LULC change taken place. This was not the case for maximum temperatures, for which the overall mean warming in excess of NNR was small (0.042°C) and not statistically significant. Urbanization seems to have a particularly strong warming influence on both minimum and maximum temperatures. Averaged over both types of urbanization observed (conversion of crop/pasture and forested lands), minimum temperatures warmed 0.379°C beyond what would have been expected in the absence of LULC change, and maximum temperatures warmed 0.197°C. All types of urbanization present near the Normals stations resulted in excess warming, and in all but one case (maximum temperatures near previously forested land) this urbanization-induced warming was statistically significant. These findings are consistent with the well-documented “urban heat island” effect, as are the larger magnitudes of warming observed in minimum temperatures relative to maximum temperatures.

[25] The only non-urbanization LULC change that resulted in temperature changes significantly different from those seen in the NNR was warming of maximum temperatures following reforestation of mechanically disturbed land. Although the magnitude of the difference was quite large, and thus statistically significant, the small sample size (two stations) makes this finding suspect.

[26] Perhaps as interesting as the LULC conversion types that resulted in significant deviations from the background warming seen in the NNR data are the conversion types for which the deviations were not significant. For example, clear-cutting of forests resulted in only slight warming of minimum temperatures (0.150°C) and slight cooling of maximum temperatures (0.074°C), neither of which was significant. This lack of significance was in spite of a fairly large sample size (n = 19.9). Thus, the combined effects of albedo changes and differences in the partitioning of net radiation into sensible, latent, and ground heat fluxes following clear-cutting of a forest seem to result in negligible near-surface temperature alterations. This is also the case for reversion of crop/pastureland to forests, another conversion type where significant differences might have been expected.

[27] The ubiquitous warming seen in minimum temperatures for all LULC change types warrants discussion in light of the fact that some change-reversal pairs were examined. For example, conversion of forest to mechanically disturbed land resulted in warming beyond what would be expected from the NNR data by 0.150°C. Nevertheless, reforestation of mechanically disturbed land, the opposite conversion, also resulted in warming (0.578°C). It should be noted, however, that only two stations were present that underwent the latter conversion, and with such a small sample size, the warming found is not statistically significant. In no case does a change-reversal pair exhibit statistically significant warming (or cooling) for both conversion types. It appears that these seemingly contradictory cases arise purely by chance as a result of the small sample sizes representing some conversion types.

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[28] The USGS Land Cover Trends Project provides high-accuracy LULC data that afford a unique opportunity for examining the effects of LULC on near-surface air temperatures. In examining minimum and maximum temperatures from U.S. Climate Normals stations near areas of LULC change, it was found that post-change minimum temperatures were warmer than those before the LULC change, and often significantly so. This was also true for maximum temperatures, although the magnitude of the temperature difference was smaller and not all LULC conversion types resulted in warming.

[29] Examination of NNR data corresponding to these stations, however, revealed that about half the warming seen in the Normals minimum temperatures and the majority of warming in the maximum temperatures was attributable to effects other than LULC change. Despite this, minimum temperatures were found to be significantly warmer as a result of nearby LULC changes in general, and for two specific types of change, urbanization of forested land and urbanization of crop/pastureland. The latter of these conversion types also resulted in significant warming of maximum temperatures.

[30] The results presented here are based on LULC data thus far completed by the USGS Land Cover Trends Project, representing less than half of all U.S. ecoregions. As more ecoregions are completed, additional Normals stations can be identified and analyzed that are near sample blocks for which historical LULC has been determined. This represents an important opportunity for increasing sample sizes within each LULC conversion type such that effects of these conversions on near-surface temperature can be quantified with greater statistical confidence.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References

[31] The authors would like to acknowledge the assistance of Tim Owen (NOAA/NCDC) with the U.S. Climate Normals data and Kristi Sayler (USGS) in providing U.S. Land Cover Trends Project data. The authors also thank Ming Cai (Florida State University) and the anonymous reviewers for providing valuable comments to improve this manuscript. This study was partially supported by the NOAA Office of Global Programs Climate Change Data and Detection element.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References