Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 Alterations in land use/land cover (LULC) in areas near meteorological observation stations can influence the measurement of climatological variables such as temperature. Urbanization near climate stations has been the focus of considerable research attention, however conversions between non-urban LULC classes may also have an impact. In this study, trends of minimum, maximum, and average temperature at 366 U.S. Climate Normals stations are analyzed based on changes in LULC defined by the U.S. Land Cover Trends Project. Results indicate relatively few significant temperature trends before periods of greatest LULC change, and these are generally evenly divided between warming and cooling trends. In contrast, after the period of greatest LULC change was observed, 95% of the stations that exhibited significant trends (minimum, maximum, or mean temperature) displayed warming trends.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 While the impacts of increasing amounts of greenhouse gases in the atmosphere on climate change have been the focus of numerous studies, analyses of altered land cover and land use on climate change and its detection have been limited. Much of the research on the climatic consequences of land use/land cover (LULC) alterations has concentrated primarily on urbanization [e.g., Gallo et al., 1999; Gallo et al., 2002; Hansen et al., 2001; Kalnay and Cai, 2003], which in many regions is not the predominant type of conversion. While some studies have included estimates of changes in other land cover types [e.g., Adegoke et al., 2003; Marshall et al., 2003; Marshall et al., 2004; Pielke et al., 1999], these are usually regional-scale analyses due to the lack of historical land use/land cover data.
 The Land Cover Trends project undertaken by the U.S. Geological Survey [Loveland et al., 2002] is designed to describe the rates and types of land cover change across the United States for four time periods spanning a total of approximately 30 years. This analysis includes manual interpretation of Landsat satellite imagery of randomly sampled blocks in each of the 84 ecoregions originally demarcated by Omernik . 20 km by 20 km sample blocks were utilized for the initial nine ecoregions analyzed, while subsequent ecoregions are being evaluated with 10 km by 10 km sample blocks to improve ecoregion representativeness. Land cover then is classified as one of 11 types at a resolution of 60 m within each sample block [Loveland et al., 2002]. The 23 (of 84) ecoregions completed by the Land Cover Trends Project thus far were used in this study (Figure 1).
 The U.S. Climate Normals of Temperature and Precipitation [National Climatic Data Center, 2002] is the definitive data set of climatological values of temperature and precipitation for the most recent (1971–2000) thirty-year interval. In addition to defining “normal” temperatures for stations included in the data set, the temperature data are critical to the production of several derivative data sets, including climatographies of frost/freeze probabilities and heating and cooling degrees. Furthermore, the active stations of the U.S. Historical Climatology Network [Easterling et al., 1996] represent a subset of the Normals stations. Data from the Normals stations have been adjusted for time of observation biases [Karl et al., 1986] and undergone quality control [Peterson et al., 1998]. In addition, inhomogeneities in the data arising from changes in station location or instrumentation have been addressed [Peterson and Easterling, 1994; Easterling and Peterson, 1995]. Climate normals are derived from arithmetic means of the adjusted and quality controlled data.
 Data available from the Land Cover Trends project offer unique opportunities for identification of LULC change associated with U.S. Climate Normals stations and the study of how such changes may influence observed climate trends. The objectives of this study are to identify Normals stations with changes in local LULC and document any climate trends potentially associated with these changes.
2. Data and Methods
 Land cover-classified images were obtained for 580 Trends Project sample blocks, totaling 85,600 km2 in area, for each of five nominal years: 1973, 1980, 1986, 1992, and 2000. Actual dates of sample block images varied due to the availability of Landsat images for classification. Difference images were created from each subsequent pair (e.g., 1973 and 1980) of classified land cover images to identify changes occurring in four periods, each approximately 6 years in length.
Gallo et al.  examined diurnal temperature ranges as measured at over 1200 U.S. Historical Climatology Network stations and compared them to local LULC data gathered by surveying weather station observers about surrounding conditions. They found that diurnal temperature range could be significantly affected by LULC within a 10-km radius around the stations. Thus, in this study, Normals stations located within a 10-km radius of a Land Cover Trends sample block were identified. For each of the 366 stations meeting this criterion, the intersected area of the Normals station 10-km buffer zone and the Land Cover Trends sample block was examined to determine both the dominant type of LULC and any changes in LULC occurring during the 1973–2000 interval of the Land Cover Trends data. Furthermore, the period in which the greatest aerial extent of LULC change of a single type (e.g., forest to urban) occurred was identified as the period of dominant change.
 In addition to evaluation of LULC types and changes in the vicinity of Normals stations, temperature data from these stations also were investigated to determine if relationships existed between trends in the temperature data and observed changes in nearby LULC. In statistically analyzing the monthly minimum, maximum, and average temperature data for the various periods of interest, two factors needed to be accounted for: the large annual cycle in temperature data and the high degree of temporal autocorrelation that might be present. In fitting a least squares linear regression line to observed temperature data, large residuals and high variance of the residuals result from the annual cycle of temperature, especially at higher latitudes. This source of variance about the least squares line was removed through computation of monthly anomalies from the 30-year monthly averages of temperature values. These anomalies were utilized for all subsequent analyses.
 Time series of temperature data often exhibit temporal autocorrelation (serial correlation), where an individual observation is not independent from the previous observation, and the degrees of freedom is reduced. This can lead to unrealistically small confidence intervals for various statistical parameters. While removal of the annual cycle typically reduces the degree of autocorrelation present, some stations still exhibit considerable autocorrelation in data even after the annual cycle has been removed. To initially determine if autocorrelation was present in the anomalies of minimum, maximum, and average temperature, the Durbin-Watson statistic [Durbin and Watson, 1951] was computed for the 1971–2000 Climate Normals period and the intervals before and after the period of dominant LULC change. A conservative approach was utilized in which Durbin-Watson statistic values that would normally be considered indeterminate (i.e., between threshold values associated with significant or nonsignificant autocorrelation) were considered significant.
 In cases where autocorrelation was suspected, the effective degrees of freedom [Mitchell et al., 1966] was used in computing the standard deviation, significance of temperature anomaly trends, and Student's t-values. For computing the t-value to determine if two periods have significantly different means, pooled values of the lag-1 correlation coefficient and sample variance must be used, as described by Zwiers and von Storch .
 Once the annual cycle had been removed and autocorrelation accounted for, trends in the minimum, maximum, and mean temperature anomalies were determined using a least squares approach.
 Minimum, maximum, and average temperatures at Normals stations had generally positive trends for the 1971–2000 period, consistent with the values observed for the Northern Hemisphere over a similar interval by Vose et al. . However, the LULC data provided by the Land Cover Trends Project allow these trends to be examined in the context of nearby LULC changes through analysis of separate temperature trends before and after the period of dominant LULC change. To facilitate such analyses, only stations with at least two years of temperature data available both before and after the period of dominant change were utilized to allow for a long enough time series for meaningful trends to be calculated. That is, stations with periods of dominant change beginning before December 31, 1972 or ending after January 1, 1999 were not used.
 Sixty percent of the 366 Normals stations located within a 10-km radius of a Land Cover Trends sample block had periods of dominant change corresponding to either the first (nominally 1973–1980) or last (1992–2000) analysis periods. The majority of these were excluded from detailed trend analysis by the above 2-year criterion, leaving 183 stations, with at least three stations in each ecoregion, that were further analyzed. The 30-year record of minimum temperature data from one such station, the Memphis, Tennessee Weather Service Forecast Office, is presented as an example (Figure 2). The time series has been divided into intervals before, during, and after its period of dominant change (May, 1986 to October, 1991), and the least squares trends before and after the period of dominant change are shown.
 Significant warming trends in minimum temperature for time series proceeding periods of dominant change were found for 22 stations. This number was eclipsed by the 27 stations exhibiting significant cooling trends in minimum temperature for these periods. Overall, these 49 stations showed an average pre-change trend of −0.20°C dec−1. Conversely, for time series after their period of dominant change, 78 stations had significant warming trends while only 2 had significant cooling trends, and the overall average post-change trend was 1.35°C dec−1. The most common type of dominant LULC conversion was from forest to urban, occurring at 35% of the stations with significantly positive trends in minimum temperature after their periods of dominant change in LULC. This is nearly double the 18% observed conversion of these LULC types (forest to urban) among all 366 stations. A dominant conversion type of cropland/pasture to grassland/shrubland was seen at 13% of the stations, compared to its overall prevalence of 6.0%. The shift toward more positive minimum temperature trends following periods of dominant change is clearly evident in a histogram of the trends (Figure 3a).
 Trends in maximum temperature before periods of dominant change were similar to those of minimum temperature such that there was roughly equal division between significantly positive and significantly negative trends. However, considerably fewer stations overall exhibited significant trends in maximum temperature prior to their period of dominant LULC change, with 7 warming and 5 cooling trends (overall pre-change trend of −0.33°C dec−1). Significant trends were far more numerous after periods of dominant LULC change, with positive trends clearly prevailing (76 positive versus 4 negative; overall post-change trend of 2.13°C dec−1). Figure 3b demonstrates the shift in trends from before periods of dominant change to after dominant LULC change. The most common type of dominant LULC change among the positive trend stations was again from forest to urban (24% of stations). It should be noted, though, that these are not all distinct from the 27 stations exhibiting significant warming trends in minimum temperature and this same LULC conversion type. Overall, 39 of the stations with significant warming trends in maximum temperature following periods of dominant change also had significant warming trends in minimum temperature. The four stations demonstrating cooling trends in maximum temperature following the periods of dominant change each underwent different dominant LULC conversions.
 Least squares trends in average temperature (not shown) followed the same patterns as those of minimum and maximum temperature. A total of 18 stations showed significant trends before their respective periods of dominant change, of which 10 were positive and 8 were negative. After the periods of dominant change, 82 stations had significant warming trends in average temperature, and only 1 station showed a significant cooling trend. It is important to note, however, that monthly average temperatures reported by Normals stations do not represent a true average over the course of a month, but are simply the mean of the recorded monthly minimum and maximum temperatures. As such, trends in average temperature do not provide additional information beyond what may be gleaned from records of minimum and maximum temperature, but are included here for completeness.
 Analysis of trends in minimum, maximum, and average temperature at 366 Normals stations located within a 10-km radius of a Land Cover Trends sample block revealed significant warming at the majority of stations for the 30-year period 1971–2000.
 Prior to the period during which the LULC around Normals stations underwent the greatest single type of LULC conversion, temperature trends were mostly insignificant. Furthermore, those trends that were significant were roughly equally divided between warming and cooling trends. This contrasted sharply with trends in temperature after periods of dominant LULC change, when 95% or more of the stations that exhibited significant trends in minimum, maximum, or mean temperature exhibited a warming trend. Significant warming in minimum temperatures was associated with a dominant LULC conversion of forest to urban at nearly twice the rate expected from chance alone. This conversion type also was strongly associated with significant warming in maximum temperatures.
 While there is strong correlation between increases in temperature trends at Normals stations and nearby LULC changes, this does not necessarily imply that the LULC changes are the causative factor. Further analyses are currently being undertaken to establish or refute causality.
 We would like to acknowledge the assistance of Kristi Sayler and Roger Auch of the Land Cover Trends Project at the USGS Center for Earth Resources Observation and Science. This study was partially supported by the NOAA Office of Global Programs Climate Change Data and Detection element.