1. Introduction
Infrared-derived land surface temperature (LST) is a key variable for understanding many surface processes, which makes the accurate determination of LST values from satellites a major goal. However, the accurate detection of LST with passive remote sensing is often modified by cloud contamination. The Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua satellites is less susceptible to cloud contamination than previous sensors because of the high frequency of coverage, 36 relatively narrow spectral bands and relatively high spatial resolution (250, 500, or 1000 m, depending on the spectral band). MODIS uses 14 spectral band radiance values to evaluate atmospheric contamination and determine whether scenes are affected by cloud shadow (Ackerman et al. 1998). In general, cloud contamination of MODIS imagery is largely related to cloud edges, optically thick aerosols, and difficult-to-identify cloud types, such as thin clouds (e.g., Ackerman et al. 1998; Platnick et al. 2003). The MODIS Collection-5 cloud detection algorithm has been assessed against lidar sensor returns by Ackerman et al. (2008) with a high (~85%) degree of success. Similar levels of success in cloud detection have been reported for daytime MODIS cloud mask when compared with Multiangle Imaging Spectroradiometer (MISR) data over the Arctic (Shi et al. 2007). The imperfect cloud detection indicates there is undetected cloud contamination in clear-sky MODIS imagery.
Validation for the MODIS LST product has been conducted on Collection-4 nighttime data (Wang et al. 2008). The use of nighttime data reduces scaling issues related to discrete thermal measurements made within the MODIS grid cell. The effects of relative humidity, wind speed, soil moisture, air temperature, and sensor view zenith angle on LST have been explored by Wang et al. (2008) for nighttime temperatures. Sensor view zenith angle showed a weak influence on error generation of swath data; the other variables showed no influence. Wan (2008) found that Collection-5 MODIS LST products have accuracy better than 1 K in 39 of 47 cases that were compared with in situ observations; the RMS of differences is reported to be 0.7 K for the full suite of cases. Langer et al. (2010) compared thermal infrared images collected in the field within 10 min of MODIS LST observations, over wet polygonal tundra located at 72°22′N in Siberia, using up to 18 daily swaths from Aqua and Terra for July–September 2008. They concluded that subpixel water bodies have a moderating impact on growing season daytime LST and must be accounted for in determining LST over tundra, and improved cloud-cover-masking and gap-filling techniques are required to improve the performance of MODIS LST. Strong negative deviation of MODIS LST values (5–15 K) compared with thermal camera images indicates that cloud top temperatures were contaminating some MODIS LST values.
Here we show that sampled air temperature, used in conjunction with Collection-5 MODIS tile LST, can reproduce daytime clear-sky land surface temperatures with sufficiently high accuracy to identify cloud-contaminated grid cells missed by the standard MODIS cloud-masking algorithms. We first define the relationship between MODIS LST and air temperature Tair recorded at four Environment Canada weather stations located between 542 and 807 m MSL in the southwest Yukon (Fig. 1). Air temperature is recorded on the hour between 2000 and 2008. These measurements were matched to daytime MODIS Terra LST values that were recorded within 12 min of the hour. We subsequently used ground-based sky cloud content information recorded on the hour at two of the four stations to determine how sky cloud content affects the LST–Tair relationship. The sky cloud content gathered at weather stations provides a means to examine how increased cloud content correlates with increased cloud contamination in MODIS LST. We refine the LST–Tair relationships for each of the sky cloud coverage using the MODIS Terra land surface temperature and emissivity daily level-3 global 1-km grid product (MOD11A1) quality flag. Last, we compare good-quality confirmed “clear” sky relationships with good-quality “mainly clear” and “mostly cloudy” sky relationships to identify cloud contamination missed by the MODIS quality flag in the latter two datasets.
Southwest Yukon study area showing the location of Environment Canada meteorology air temperature recording stations (black triangles) and air temperature and sky observation recording stations (black circles). Base Landsat mosaic of the Yukon was obtained from the Yukon Government, Department of Geomatics.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
2. Study area
The southwest Yukon study area is bounded by the Yukon–Alaska border (141°W) on the west and southwest, the Yukon–British Columbia border on the south (60°N), 134°W on the east, and 62.5°N on the north. This section of the Yukon contains four Environment Canada meteorology monitoring stations (Fig. 1), all of which are situated near airfields. This region is affected by the St. Elias mountains’ rain shadow, and annual precipitation is 30–80 cm. Although not typically considered to be a semiarid or arid landscape [a type known to impede the function of the MODIS LST algorithm (Wan et al. 2002)], the effect that relatively dry portions of this landscape might have on emissivity determination has not been quantified. The Environment Canada weather station sites at Burwash Landing, Haines Junction, and Carmacks are largely covered by homogeneous boreal forest, with a small area of open water from the Yukon River contained in the Carmacks MODIS LST grid cell. The Whitehorse MODIS grid cell contains several land cover classes, including urban, roads, and forest. Cloud cover is frequent throughout the year in this part of the Yukon.
3. Data
The daily MODIS Terra land surface temperature and emissivity 5-min level-2 swath 1-km dataset (MOD11_L2) product is dependent on several parameters related to the satellite platform and sensor and is generated using the product generation executive (PGE16) code. The MOD11_L2 is in swath format and collected sequentially on a single overpass. The swath is composed of several products, including geolocation, sensor radiance, atmospheric temperature and water profiles, cloud mask, quarterly land cover, and snow cover. Several MODIS land surface temperature products are produced at daily, 8-day, or monthly intervals at a gridded resolution of 1 km and 6 km and on a 0.05° grid. The MOD11A1 product used here is produced by mapping daily single clear-sky observation MOD11_L2 swath data onto 1-km tiled grids in sinusoidal projection. The MODIS Terra satellite and its sister satellite, Aqua, are in sun-synchronous near-polar orbits, with Aqua in an ascending orbit and Terra in a descending orbit. The orbital structures dictate equatorial crossings at 1030 for Terra and 1330 for Aqua (local solar time). Because of the near-polar orbit, there is progressively more swath overlap at locations greater than 30° latitude, thus producing multiple daily observations in these regions.
a. MODIS Terra (MOD11A1, Collection 5) clear-sky daytime LST
The MODIS LST data used in this study are the MOD11A1 h11 tile data acquired daily under daytime clear-sky conditions, using reprocessing Collection 5 (henceforth LST). The MOD11A1 data contain single observations in each grid cell rather than averaged observations. MODIS data were downloaded from the Land Processes Distributed Active Archive Center (LPDAAC; https://lpdaac.usgs.gov/). The basic 1-km gridded MODIS land surface temperature is produced with a split-window technique that uses MODIS bands 31 and 32 (10.78–11.28 μm and 11.77–12.27 μm, respectively) and is detailed by Wan et al. (2002). This technique uses a global land surface classification-based emissivity lookup table (Snyder et al. 1998) to estimate emissivity values in these two bands. A split-window technique is used to produce the 1-km LST product, whereas the 5-km LST product uses the day/night algorithm. The split-window class of techniques uses the difference in water vapor absorption that exists between band 31 and band 32 to determine surface temperature. Cloud contamination is known to cause LST error in the split-window temperature extraction method (Jin and Dickinson 2000).
Nishida et al. (2003) employed the MOD11A1 quality control flag for cloud screening, an approach we also employ here. The quality flag information relating to Collection 5 of the MODIS land surface temperature products is found in the product users’ guide (http://www.icess.ucsb.edu/modis/LstUsrGuide/MODIS_LST_products_Users_guide_C5.pdf). The MODIS quality flag used in this study is the daytime LST and emissivity quality control layer. The first bit indicates whether the produced LST is of good quality and does not require further inspection of the quality flags or if the LST was produced but that the quality is unreliable or unquantifiable and further quality flag inspection is required. Inspection of all the MODIS quality flags produced for the unreliable or unquantifiable quality first bit provides differentiation of the average temperature quality flags as ≤3, ≤2, and ≤1 K, which is the nominal (or minimum) level; the >3-K-average temperature quality flag was not assigned to any of the data considered in this study. Grid cells with the MODIS quality flags raised related to emissivity and subpixel cirrus cloud presence were eliminated from this study; these affected less than 5% of the total dataset.
b. MODIS data preprocessing
The MOD11A1 tile data were subset to the study area, as defined above. MOD11A1 LST and the coincident local solar view time from 2000 through 2008 were reprojected from sinusoidal projection to 1-km gridded geotiff Albers equal area projection, North American Datum 83 (NAD83), using nearest neighbor resampling. This processing step was batch run using the MODIS Reprojection Tool (MRT) (Dwyer and Schmidt 2006) available from the LPDAAC. These LST product layers were converted to local solar time and temperature using the conversion factors provided at https://lpdaac.usgs.gov/products/modis_products_table/mod11a1. The view times in the MOD11A1 product are in local solar time, which is defined as the MODIS observation time in UTC plus longitude in degrees at the 1-km grid cell divided by 15. MODIS view times are converted from local solar time to local time to enable comparison with meteorological station data.
c. Meteorology station air temperature and weather observations
Air temperature data used in this study are collected on the hour at four long-term Environment Canada meteorology monitoring stations. Hourly air temperature data and weather observations were downloaded from Environment Canada’s National Climate Data and Information Archive (http://www.climate.weatheroffice.gc.ca/Welcome_e.html). Ground-based cloud observations are only available from nonautomated observation weather stations, which include two of the four stations in the study region (Burwash and Whitehorse stations). The sky observation total cloud amount is recorded in tenths when no visibility obstruction (e.g., smoke) occurs. Cloud observations are grouped into four categories: clear (0 tenths obstructed), mainly clear (1–4 tenths), mostly cloudy (5–9 tenths), and cloudy (10 tenths). Table 1 provides information regarding the location, elevation, and other characteristics of the four meteorology stations.
Station information for the four Environment Canada locations in the southwest Yukon.
d. MODIS and meteorology data intersection
Daily MODIS layers containing LST values and each grid cells’ associated view time (time of LST capture) were matched with the four Environment Canada meteorology station locations (Whitehorse, Carmacks, Haines Junction, and Burwash Landing) using nearest neighbor sampling. The MODIS data used extend from March 2000 to the end of 2008. Only MODIS LST values that were recorded within 12 min of the hour (air temperature is recorded on the hour) were retained for analysis. The positions of each meteorology station within the 1 km × 1 km grid cell are different for each location.
e. Statistical analysis
At warmer temperatures (>0°C) the MODIS LST and air temperatures reported here displayed a slight departure from a linear relationship. We fitted the MODIS LST–air temperature relationship for all sites with a linear model and again with a quadratic model. We compared the values of the residuals generated from both statistical fits at each sampling period with a matched-pair t test to determine if the two models were significantly different from each other (Zar 2009).
We compared the departure from the predicted linear fit for the MODIS LST–air temperature relationship at all sites for three levels of average temperature quality flag (≤1, ≤2, and ≤3 K). The data were log (x + 1) transformed to a normal distribution, thus compensating for a positive skew. Preliminary analyses indicated that the within-group variances of the three quality groups were not equal, so we proceeded to analyze the data with Welch’s analysis of variance (ANOVA) (Zar 2009). The quality flags we used are associated with temperature uncertainty; all other quality flags, such as errors associated with cirrus clouds, occurred infrequently and were removed from analysis throughout.
We compared the mean temperature associated with each average temperature quality flag. If assignment of the quality flags occur randomly throughout the year and are based on MODIS spectral thresholds, the mean temperature associated with each error code should not be significantly different between groups. Data were square-root transformed and then reflected to correct a negative skew in distribution (Zar 2009). Preliminary analyses indicated that the within-group variances of the three groups were not equal, so we proceeded to analyze the data with Welch’s ANOVA.
4. Results
The four sites studied here produce similar curves of LST versus Tair, which display a slight positive curvature. Strong quadratic correlations between MODIS LST and instrumental air temperatures (R2 = 0.95–0.96) are observed at the four Environment Canada meteorology stations (Fig. 2). The Whitehorse, Carmacks, Burwash Landing, and Haines Junction monitoring station locations are found at different elevations and are located at different positions within the LST grid cell. The results are similar to those of Comiso (2003) for matching air temperature to Advanced Very High Resolution Radiometer (AVHRR) LST data for air temperatures below 0°C. Furthermore, the relationships are consistent with findings by Karlsen and Elvebakk (2003), where the differences between subsurface temperature and air temperature start to appear above 0°C and are greatest for the warmest temperatures. When the pooled MODIS LST–Tair relationship was fitted with a quadratic equation (to compensate for the departure from linearity), the increase in fit from R2 = 0.942 (ANOVA, F1,1023 = 1666, p = 0.000 01) for the linear fit to R2 = 0.954 (ANOVA, F2,1022 = 10 652, p = 0.000 01) for the quadratic fit was strongly insignificant (matched pairs, t2291 = 0.021, p = 0.9826). View angle, relative humidity (Carmacks’ relative humidity data were unavailable), amount of time between recording of air temperature and LST (to a maximum of 12 min), and wind speed did not produce statistically significant results for the variation between LST and Tair for the polynomial fits. The magnitude in the difference from the predicted quadratic fit for MODIS LST–Tair was significantly different among the three groups of average temperature quality flags (Welch’s ANOVA, F2,137.56 = 7.7111, p = 0.0007). The mean departure (untransformed data) in kelvins from the predicted fit was lowest for an average temperature quality flag ≤3 K (n = 49, mean ± standard error = 1.75 ± 0.36 K) compared with quality flag ≤1 K (n = 771, mean ± standard error = 2.75 ± 0.087 K) and ≤2 K (n = 933, mean ± standard error = 2.87 ± 0.079 K).
The 2000–08 temporally matched MODIS clear-sky LST vs Environment Canada air temperature at Carmacks, Haines Junction, Whitehorse, and Burwash Landing stations. The solid line indicates the quadratic trend line and the dashed line indicates the 1:1 line.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
Langer et al. (2010) found that the largest differences in LST among tundra land cover classes, including water, were found for the highest values of net radiation. The MODIS LST values used here are collected under conditions of daytime clear sky when the land surface is experiencing the higher end of the daily range of net radiation. Thus, large water bodies found within the MODIS LST grid cell will be moderated by the relatively high heat capacity of water. The slightly greater linearity and higher R2 value found at Carmacks (Fig. 2) is likely related to the moderating effect of water from the Yukon River, which is contained within the southwest 10%–15% of the LST grid cell. The 5% uncertainty in the previously stated range of areal water coverage is estimated using the 50 m at nadir MODIS geolocation accuracy (Wolfe et al. 2002). None of the other three MODIS LST grid cells contain large water bodies; the Whitehorse grid cell containing the Environment Canada meteorology station did not cover the Yukon River.
The mean LST grouped by three temperature error codes differed significantly (Welch’s ANOVA, F2,285.16 = 1403.4077, p = 0.001). The mean temperature for quality flag ≤3 K was significantly lower (X = mean ± standard error = 242.89 ± 0.67 K) compared with both error code ≤1 K (X = mean ± standard error = 281.59 ± 0.56 K) and error code ≤2 K (X = mean ± standard error = 276.53 ± 0.45 K). This indicates the LST associated with the ≤3-K quality flag originated from the colder portion of the dataset (<0°C).
Constraining the LST–Tair relationship at Whitehorse and Burwash with ground-based sky observations of clear (zero cloud cover), mainly clear, and mostly cloudy sky conditions decreases the correlation coefficient for a quadratic model and causes a progressively more negative y intercept (Fig. 3). The use of the quadratic model for this section of analysis is to show that some of the curvature in the LST–Tair relationships reported here is related to cloud cover. Figure 4 shows quadratic relationships displayed in Fig. 3 for the Whitehorse meteorological station but separated by the MODIS quality flag of good quality (left) and unreliable or unquantifiable quantity (right). Figure 5 shows quadratic relationships displayed in Fig. 3 for the Burwash meteorological station but separated by the MODIS quality flag of good quality (left) and unreliable or unquantifiable quantity (right). The quadratic equations of fit for Figs. 4 and 5 show that the y intercept decreases as sky cloud content increases (from top to bottom panels). Plotting the good-quality confirmed clear-sky trend on ensemble good-quality mainly clear and mostly cloudy conditions for each location shows that 13% (Whitehorse) and 17% (Burwash) of LST values appear below the lower bound of the trend lines as defined by the RMS value for the confirmed uncontaminated trend (Fig. 6). These values are consistent with the estimates for unidentified cloud contamination of MODIS data (Ackerman et al. 2008; Shi et al. 2007). Four data points in the Burwash ensemble good-quality mainly clear and mostly cloudy conditions were removed from this analysis because they were colder than any data point in the good-quality confirmed clear-sky data.
The 2000–08 temporally matched MODIS clear-sky LST vs Environment Canada air temperature at Whitehorse and Burwash for ground-based cloud content assessment, including all MODIS temperature quality flag information. (top) Clear sky, (middle) mainly clear sky, and (bottom) mostly cloudy sky conditions. The solid line indicates a quadratic trend; the dashed line indicates a 1:1 line.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
The 2000–08 temporally matched MODIS clear-sky LST vs Environment Canada air temperature at Whitehorse for ground-based cloud content assessment. (top) Clear sky, (middle) mainly clear sky, and (bottom) mostly cloudy sky conditions. The left column is MODIS LST with good-quality flag values and the right column contains LST values of unreliable or unquantifiable quality. The solid line indicates a quadratic trend; the dashed line indicates a 1:1 line.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
The 2000–08 temporally matched MODIS clear-sky LST vs Environment Canada air temperature at Burwash for ground-based cloud content assessment. (top) Clear sky, (middle) mainly clear sky conditions, and (bottom) mostly cloudy sky conditions. The left column is MODIS LST with good-quality flag values and the right column contains LST values of unreliable or unquantifiable quality. The solid line indicates a quadratic trend; the dashed line indicates a 1:1 line.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
The 2000–08 temporally matched MODIS clear-sky LST vs Environment Canada air temperature at Whitehorse and Burwash for good-quality mainly clear and mostly cloudy sky data from Figs. 4 and 5. The plotted quadratic trends found for clear-sky and good-quality data for each site (solid line) are shown. The dashed line is the RMS generated from the clear-sky good-quality data specific to each site. Any points falling below the bottom dashed line are likely cloud contaminated.
Citation: Journal of Climate 26, 5; 10.1175/JCLI-D-12-00250.1
5. Discussion
Approximately 15% of cloud-contaminated grid cells are not being detected by the MODIS cloud mask (Ackerman et al. 2008). For the AVHRR sensor, an increase in cloud coverage as viewed from the ground correlates with a diminished confidence that the sensor is viewing clear sky (Ackerman et al. 1998). Here, we extend this principle to include surface air temperature as a way to identify undetected cloud contamination in MODIS LST data. Thus, an increasing amount of sky cloud content observed from the ground (a portion of which contains a MODIS grid cell that encompasses an air temperature monitoring station) correlates with colder LST values when compared with temporally matched air temperature recordings. The confirmed clear-sky good-quality relationships (top left panels in Figs. 4 and 5) displayed the highest correlation coefficients using a quadratic fit. If good-quality flagged MODIS grid cells were truly free of contamination, we would expect that the relationships found for clear, mainly clear, and mostly cloudy would be similar. The consistent reduction of the y intercept from the LST–Tair relationships, progressively constrained by ground-based sky observations of clear to mostly cloudy, indicates the relationship in the good-quality flagged MODIS grid cells are increasingly contaminated with clouds (left panels in Figs. 4 and 5). Indeed, if the right panels in Figs. 4 and 5 (unreliable or unquantifiable quality) are considered, a clear trend to more cloud contamination from clear to mostly cloudy is observed as expected. We conclude that this technique provides a semiquantitative method to determine daytime MODIS LST cloud contamination by plotting any daytime LST–Tair intersection against the quadratic fit for confirmed clear-sky good-quality data. LST values found below the RMS of the trend line are considered contaminated.
Some variation in the LST–Tair relationship trends might be related to the ≤12 min discrepancy between air temperature recording and LST capture. However, we showed that this variation was statistically nonsignificant, likely because air temperature reacts slowly compared with LST following short-lived perturbations (i.e., cloud shadow) over a 1-km grid cell, especially in comparison with cloud contamination in the LST grid cell. Findings by Wang et al. (2008) indicated no seasonal bias in nighttime field LST measurements and MODIS LST. Furthermore, Bartlett et al. (2006) found that snow cover exerts only a minor influence on the annual tracking of subsurface and air temperature. We found that the ≤3-K quality flag originated from the below 0°C portion of the LST dataset exclusively, suggesting that the assignment of this quality flag is sensitive to a specific set of atmospheric and land cover conditions.
The LST versus Tair equations of fit for Whitehorse are bounded by the equations of fit for Burwash Landing and Haines Junction, which have simpler and more homogeneous forested land cover class compared with multiple land cover classes found at the Whitehorse airport, where air temperature is recorded (Fig. 2). This suggests that the MODIS LST at 1 km adequately integrates landscape variability (and thus emissivity variation) within the grid cell, regardless of the placement of the monitoring station, well enough to identify the influence of cloud contamination missed by the MODIS cloud mask. Furthermore, there is likely an influence on the relationships found here from cloud shadow, but this was assumed to be small compared with that from cloud contamination, because view angle did not show a statistically significant influence on the LST–Tair relationship. The view capture angles of MODIS LST (<55°) likely make the cloud shadow effect small compared with the effect of cloud contamination, especially in relation to air temperature.
The degree of cloud contamination in the MODIS grid cells appears to be responsible for the major part of observed variation in LST, manifested as the systematic bias toward colder-than-true values. This phenomenon will likely have a moderating effect on apparent LST trends if contamination is not identified or cloud content is not fully considered (e.g., Liu et al. 2008). In the larger context of infrared-derived LST, the findings of this study indicate that more accurate trends will be produced if cloud contamination is further reduced. Therefore, we argue that all available means for the identification of cloud contamination, including using meteorological stations in a coordinated approach to produce validated satellite data, should be employed.
Further work is required to identify the type and degree of cloud contamination in MODIS land surface temperature. A reevaluation of the MODIS cloud mask could be completed using the technique outlined in this paper in conjunction with more precise observations, such as lidar, to further refine spectral thresholds for cloud contamination detection.
6. Conclusions
In this study, we compared MODIS Terra (MOD11A1, Collection 5) clear-sky daytime LST with temporally matched Environment Canada meteorological station air temperatures in the southwest Yukon to identify cloud contamination missed by the MODIS cloud mask. Using qualitative ground-based sky condition observations and coincident MODIS quality flag information available at two stations, we further refined the MODIS LST–air temperature relationships. The relationships constrained by ground observations of clear-sky conditions showed less variability than those found under mainly clear and mostly cloudy sky conditions. Furthermore, there was a systematic decrease in the y intercept of the quadratic equations, indicating a systematic bias of unidentified cloud contamination (not detected by the MODIS cloud-screening methods) on the MODIS good-quality LST product. The trend of an increasing cloudy sky decreasing the LST–air temperature y intercept is largely consistent with the MODIS LST quality flag information. However, the assignment of the ≤3-K quality flag was biased to temperatures below 0°C, suggesting the degree of temperature contrast between cloud and target needed for discrimination of cloud contamination is greater than 2 K. The amount of unidentified cloud was determined to be 13% at Whitehorse and 17% at Burwash, values that are consistent with other published studies on unidentified cloud contamination in MODIS data. The clear-sky good-quality LST–air temperature relationship provided here could be used to provide further cloud discrimination of MODIS data, especially if used in concert with other cloud detection methods.
Acknowledgments
Financial support for this project was provided by the Canada Foundation for Innovation, the Canada Research Chairs Program, the Government of Canada International Polar Year Program (PPS Arctic Canada Project), and the Natural Sciences and Engineering Research Council of Canada (Discovery and International Polar Year Programs). Air temperature and weather observation data for the southwest Yukon were obtained through the Environment Canada’s National Climate Data and Information Archive (www.climate.weatheroffice.gc.ca).
REFERENCES
Ackerman, S. A., K. I. Strabala, W. P. Menzel, R. A. Frey, C. C. Moeller, and L. E. Gumley, 1998: Discriminating clear sky from clouds with MODIS. J. Geophys. Res., 103, 32 141–32 157.
Ackerman, S. A., R. E. Holz, R. Frey, E. W. Eloranta, B. C. Maddux, and M. McGill, 2008: Cloud detection with MODIS. Part II: Validation. J. Atmos. Oceanic Technol., 25, 1073–1086.
Bartlett, M. G., D. S. Chapman, and R. N. Harris, 2006: A decade of ground–air temperature tracking at Emigrant Pass Observatory, Utah. J. Climate, 19, 3722–3731.
Comiso, J. C., 2003: Warming trends in the Arctic from clear sky satellite observations. J. Climate, 16, 3498–3510.
Dwyer, J., and G. Schmidt, 2006: The MODIS reprojection tool. Earth Science Satellite Remote Sensing, J. J. Qu et al., Eds., Springer, 162–177.
Jin, M., and R. E. Dickinson, 2000: A generalized algorithm for retrieving cloudy sky skin temperature from satellite thermal infrared radiances. J. Geophys. Res., 105, 27 037–27 047.
Karlsen, S. R., and A. Elvebakk, 2003: A method using indicator plants to map local climactic variation in the Kagerlussuaq/Scoresby Sund area, east Greenland. J. Biogeogr., 30, 1469–1491.
Langer, M., S. Westermann, and J. Boike, 2010: Spatial and temporal variations of summer surface temperatures of wet polygonal tundra in Siberia—Implications for MODIS LST base permafrost monitoring. Remote Sens. Environ., 114, 2059–2069.
Liu, Y., J. R. Key, and X. Wang, 2008: The influence of changes in cloud cover on recent surface temperature trends in the arctic. J. Climate, 21, 705–715.
Nishida, K., R. R. Nemani, J. M. Glassy, and S. W. Running, 2003: Development of an evapotranspiration index from Aqua/MODIS for monitoring surface moisture status. IEEE Trans. Geosci. Remote Sens., 41, 493–501.
Platnick, S., M. D. King, S. A. Ackerman, W. P. Menzel, B. A. Baum, J. C. Riedi, and R. A. Frey, 2003: The MODIS cloud products: Algorithms and examples from Terra. IEEE Trans. Geosci. Remote Sens., 41, 459–473.
Shi, T., E. E. Clothiaux, B. Yu, A. J. Braverman, and D. N. Groff, 2007: Detection of daytime arctic clouds using MISR and MODIS data. Remote Sens. Environ., 107, 172–184.
Snyder, W. C., Z. Wan, Y. Zhang, and Y. Z. Feng, 1998: Classification-based emissivity for land surface temperature measurement from space. Int. J. Remote Sens., 19, 2753–2774.
Wan, Z., 2008: New refinements and validation of MODIS land-surface temperature/emissivity products. Remote Sens. Environ., 112, 59–74.
Wan, Z., Y. Zhang, Q. Zhang, and Z. Li, 2002: Validation of the land-surface temperature product retrieved from Terra Moderate Resolution Imaging Spectroradiometer data. Remote Sens. Environ., 83, 163–180.
Wang, W., S. Liang, and T. Meyers, 2008: Validating MODIS land surface temperature products using long-term nighttime ground measurements. Remote Sens. Environ., 112, 623–635.
Wolfe, R., and Coauthors, 2002: Achieving sub-pixel geolocation accuracy in support of MODIS land science. Remote Sens. Environ., 83, 31–49.
Zar, J. H., 2009: Biostatistical Analysis. 5th ed. Prentice Hall, 944 pp.
