Development and Analysis of a Long-Term, Global, Terrestrial Land Surface Temperature Dataset Based on HIRS Satellite Retrievals

Amanda L. Siemann Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

Search for other papers by Amanda L. Siemann in
Current site
Google Scholar
PubMed
Close
,
Gabriele Coccia Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

Search for other papers by Gabriele Coccia in
Current site
Google Scholar
PubMed
Close
,
Ming Pan Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

Search for other papers by Ming Pan in
Current site
Google Scholar
PubMed
Close
, and
Eric F. Wood Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey

Search for other papers by Eric F. Wood in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Land surface temperature (LST) is a critical state variable for surface energy exchanges as it is one of the controls on emitted radiation at Earth’s surface. LST also exerts an important control on turbulent fluxes through the temperature gradient between LST and air temperature. Although observations of surface energy balance components are widely accessible from in situ stations in most developed regions, these ground-based observations are not available in many underdeveloped regions. Satellite remote sensing measurements provide wider spatial coverage to derive LST over land and are used in this study to form a high-resolution, long-term LST data product. As selected by the Global Energy and Water Exchanges project (GEWEX) Data and Assessments Panel (GDAP) for development of internally consistent datasets, the High Resolution Infrared Radiation Sounder (HIRS) data are used for the primary satellite observations because of the data record length. The final HIRS-consistent, hourly, global, 0.5° resolution LST dataset for clear and cloudy conditions from 1979 to 2009 is developed through merging the National Centers for Environmental Prediction (NCEP) Climate Forecast System Reanalysis (CFSR) LST estimates with the HIRS retrievals using a Bayesian postprocessing procedure. The Baseline Surface Radiation Network (BSRN) observations are used to validate the HIRS retrievals, the CFSR LST estimates, and the final merged LST dataset. An intercomparison between the original retrievals and CFSR LST datasets, before and after merging, is also presented with an analysis of the datasets, including an error assessment of the final LST dataset.

Corresponding author address: Amanda L. Siemann, Department of Civil and Environmental Engineering, E-208 E-Quad, Princeton University, Princeton, NJ 08544. E-mail: siemann@princeton.edu

Abstract

Land surface temperature (LST) is a critical state variable for surface energy exchanges as it is one of the controls on emitted radiation at Earth’s surface. LST also exerts an important control on turbulent fluxes through the temperature gradient between LST and air temperature. Although observations of surface energy balance components are widely accessible from in situ stations in most developed regions, these ground-based observations are not available in many underdeveloped regions. Satellite remote sensing measurements provide wider spatial coverage to derive LST over land and are used in this study to form a high-resolution, long-term LST data product. As selected by the Global Energy and Water Exchanges project (GEWEX) Data and Assessments Panel (GDAP) for development of internally consistent datasets, the High Resolution Infrared Radiation Sounder (HIRS) data are used for the primary satellite observations because of the data record length. The final HIRS-consistent, hourly, global, 0.5° resolution LST dataset for clear and cloudy conditions from 1979 to 2009 is developed through merging the National Centers for Environmental Prediction (NCEP) Climate Forecast System Reanalysis (CFSR) LST estimates with the HIRS retrievals using a Bayesian postprocessing procedure. The Baseline Surface Radiation Network (BSRN) observations are used to validate the HIRS retrievals, the CFSR LST estimates, and the final merged LST dataset. An intercomparison between the original retrievals and CFSR LST datasets, before and after merging, is also presented with an analysis of the datasets, including an error assessment of the final LST dataset.

Corresponding author address: Amanda L. Siemann, Department of Civil and Environmental Engineering, E-208 E-Quad, Princeton University, Princeton, NJ 08544. E-mail: siemann@princeton.edu

1. Introduction

Land surface temperature (LST) is affected by the land–atmosphere energy fluxes and physical processes at global and regional spatial scales (Wan et al. 2002). As a critical state variable for surface energy exchanges, emitted radiation from Earth’s surface is a function of LST and thereby is an important component of the surface net radiation. As part of the surface radiative budget, LST also controls the surface turbulent fluxes, particularly through the temperature gradient between the LST and air temperature that drives the sensible heat turbulent flux involved in nighttime cooling and daytime heating of the surface (Sun and Pinker 2003). Additionally, the diurnal cycle of LST can be used for estimating drought conditions (Anderson et al. 2011) and surface soil moisture modeling (Wan et al. 2002), for weather prediction and climate modeling (Dash et al. 2002), and, when examined with emissivity, for land-cover analysis (Wan et al. 2002).

Although observations of surface energy budget variables, such as LST, are widely available at in situ stations or networks in developed, well-populated regions, many sparsely populated or less developed regions do not have available in situ observations because of sparse spatial coverage or inconsistent operation (Sahoo et al. 2011). Therefore, at the global scale, satellite remote sensing provides the wider spatial coverage necessary to derive LST globally, allowing for various spatial and temporal resolutions depending on the platform (Gao et al. 2010). Limitations of the satellite-observed LST include cloud-cover blockage of the surface for measurements in the thermal infrared spectrum, field-of-view limitations that, for polar-orbiting satellites, only a swath can be observed at a given time with full global coverage requiring a number of days depending on the satellite’s orbit characteristics or for geosynchronous satellites with a field of view limited to about ±60° of latitude. Additionally, sensors that offer higher resolution have narrower swaths and longer orbit repeat patterns (lower frequency of measurements at one location).

Satellite measurements used to derive LST have historically been made from various platforms with a variety of spatial resolutions and repeat patterns. The first generation of instruments, such as those on the Geostationary Operational Environmental Satellites (GOES) and the Advanced Very High Resolution Radiometer (AVHRR) and High Resolution Infrared Radiation Sounder (HIRS) on NOAA’s polar-orbiting environmental satellites, was used to make satellite-based observations as early as the late 1970s, with the Landsat series of high-resolution, 16-day repeat orbit starting in the 1980s. NASA’s Earth Observing System Terra platform (launched in 1997) included the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) and the Moderate Resolution Imaging Spectroradiometer (MODIS). Polar-orbiting satellite systems’ hosting instruments such as Landsat, ASTER, MODIS (Şahin et al. 2012), AVHRR (Sun and Pinker 2003), and HIRS provide return rates from 16 days (e.g., Landsat and ASTER; Li et al. 2004) to one or two times per day [e.g., MODIS (Şahin et al. 2012) and AVHRR (Sun and Pinker 2003)]. The range of the resolutions for the observations is approximately 60 m for the Landsat instruments, approximately 90 m for ASTER, approximately 1 km for MODIS and AVHRR (Li et al. 2004), and approximately 70 km for HIRS (Robel 2009). In contrast to some of the return rates of the polar-orbiting satellites, GOES observe the surface continually at a resolution of approximately 4 km, providing diurnal coverage (Sun and Pinker 2003) up to about ±60° of latitude.

Regardless of the resolution and return rate, gaps still exist in the measurements from each satellite for the reasons previously discussed. Efforts to fill gaps in remotely sensed satellite data have been commonly applied to land-cover classification products or other land surface properties. Strategies relating to gap-filling NDVI values include using an “effective” value from a representative time period that had similar climatic conditions to those present at the time of missing data or filling missing or inaccurate data due to high cloud levels with the highest NDVI value at that location within a given year (Sellers et al. 1996). Moody et al. (2005) filled in missing data in global surface albedo datasets derived from MODIS land-cover classification products using a temporal interpolation strategy dependent on ecosystem type and properties (Moody et al. 2005). Additionally, geostatistical techniques such as kriging and cokriging have been used to estimate missing data in scan line corrector-off Landsat Enhanced Thematic Mapper Plus (ETM+) images (Zhang et al. 2007).

While the previous gap-filling work has been successful in these cases, the nature of land surface classifications and images is that these variables are not changing at the hourly time scale, and more pixels of data may be available within a given area. Specific to LST, efforts have been made to form products at the daily scale that are time consistent (e.g., Duan et al. 2014) and that compare well across various satellite platforms with in situ data in temperature-based validation (e.g., Guillevic et al. 2014), but these products do not contain hourly estimates of the LST throughout the diurnal cycle. Previous work has been done to interpolate the satellite-based land surface skin temperature to a diurnal cycle by Jin and Dickinson (1999) using information from climate models to determine climatologies of model diurnal temperature patterns that were applied to GOES-8 and AVHRR satellite-observed data to obtain the skin temperature diurnal cycle for clear-sky conditions, but because of cloud cover and swath location, many pixels in one given area may be missing. Jin (2000) applied knowledge of neighboring clear-sky pixels and other parameters such as air temperature and net radiation to estimate skin temperature under cloudy conditions, but in many cases these other parameters and neighboring data are not available to be used for filling with interpolation techniques.

In this paper, a new strategy is used to form a long-term, temporally and spatially consistent global LST data product based on a remotely sensed dataset. The Global Energy and Water Exchanges project (GEWEX) Data and Assessments Panel (GDAP), among other research programs [e.g., NASA Energy and Water Cycle Study (NEWS) program; Yang et al. 2006], is focused on developing global satellite-based datasets for the global water and energy cycles that can be used as climate data records (see National Research Council 2004) for analyses of these cycles and their possible trends. In addition, such datasets can be used to improve global and regional hydrological modeling and thereby substantially improve global modeling of evaporation, precipitation, and sensitivities of the atmosphere and radiation balances to changes in climate. These modeling improvements involve the use of temporally and spatially consistent and continuous global data products for climate model assessment (i.e., “benchmarking”; Taylor et al. 2012) or for improving offline global land surface model simulations (Nijssen et al. 2001; Liang et al. 1998; Lohmann et al. 1998; Wood et al. 1998) for which the global datasets are used as model forcings. One such example of the need for observation-based datasets for climate model assessment is the World Climate Research Programme (WCRP) Data Advisory Council’s (WDAC) observations for model evaluation task team’s call for observation datasets for model evaluation of output (including LST) from phases 5 and 6 of the Coupled Model Intercomparison Project (CMIP5 and CMIP6) as part of Observations for Model Intercomparisons (Obs4MIPs) (see Earth System Grid Federation 2015).

A primary goal of GDAP is the development of long-term (multidecadal) datasets that are internally consistent in that, to the extent possible, all derived water and energy datasets should use the same primary satellite measurements. Because the HIRS instrument has flown on 11 NOAA operational satellites since July 1979, GDAP has selected the HIRS instrument as the basis of a long-term LST dataset and subsequently derived products such as Earth-emitted longwave radiation. Therefore, we have developed this LST product using the dataset from the HIRS instrument to provide GDAP with a product consistent with their other LST-based data products.

Thus, the focus of this paper is on the development of an hourly, global, 1979–2009 LST dataset for clear- and cloudy-sky conditions that is consistent with the HIRS LST data that is retrieved under clear-sky conditions from the 11 NOAA operational polar-orbiting satellites. Except at very high latitudes, there will be less than two potential observations per day, so the HIRS dataset is sparse relative to the complete dataset. The challenge is increased by the NOAA satellites having different equatorial crossing times, such that the retrieved LSTs are obtained at varying times during the diurnal cycle, resulting in an inconsistent time series from an observational perspective. Therefore, the HIRS LST dataset needs to be merged with a different global LST dataset with a continuous time series of estimates in all conditions to fill these unobserved gaps in the HIRS retrievals. The HIRS dataset is described in section 2a.

The overall strategy is to “in fill” the unobserved HIRS LST with surface temperature estimates from simulations based on the National Centers for Environmental Prediction (NCEP) Climate Forecast System Reanalysis (CFSR; Saha et al. 2010) (see section 2b) and to merge these estimates with the HIRS retrievals through a Bayesian postprocessing procedure that is summarized in section 4a and more fully described in Coccia et al. (2015). In the Bayesian postprocessing, the HIRS LST is the “target” so the CFSR estimates are adjusted to be consistent with the HIRS estimate, creating a final data product consistent with the HIRS LST retrievals. Results are provided in section 3 and include in section 3a the validation of the HIRS retrievals and CFSR LST estimates using the Baseline Surface Radiation Network (BSRN) station data (Ohmura et al. 1998) and an intercomparison between the HIRS and CFSR LST datasets prior to and after merging. Section 4 carries out an analysis on the datasets, which includes an error assessment for the final (merged) time series.

2. Methods

a. HIRS LST dataset

1) HIRS dataset formation

The instantaneous LST dataset from the series of HIRS sensors has recently been reprocessed at the National Climatic Data Center (NCDC) (Shi 2011, 2013). The reprocessed data encompass the period from 1979 to 2009, from the NOAA-6 to NOAA-17 operational satellites. Table 1 displays the duration of data and the nominal equatorial crossing time for each satellite for the purpose of categorizing the satellites into afternoon or morning satellites. The actual time stamps associated with each retrieval are used in this study, and retrievals are binned into hourly time steps. The data are provided in swath format and retrieved as point values for clear-sky conditions, assuming a surface emissivity of 1.0. The LST point retrievals were aggregated into 0.5° pixels as part of this study. The 0.5° pixel size was selected in part to be consistent with the resolution of the CFSR data resolution. Each grid cell is associated with a time stamp of the HIRS observation.

Table 1.

The duration of data used in this study from each satellite (acronyms in parentheses are used in the text) and the nominal equatorial crossing time (ECT) in local time at launch for each satellite (Ignatov et al. 2004). These ECT are nominal times used to categorize the satellites as afternoon and morning crossing times.

Table 1.

For the purposes of developing a HIRS-consistent, all-weather LST product based upon land surface model (LSM) estimates, the optimal temporal scale of intercomparison and analysis needs to be determined. Producing a high-temporal-resolution LST product is necessary because the HIRS retrievals are instantaneous. Additionally, the NOAA satellite carrying each HIRS sensor has a specific orbit and equatorial crossing time, so temporal averaging during a day is infeasible. This information, coupled with the temporal dynamics of LSTs and the desire to provide more data points for intercomparison between model-derived and HIRS-retrieved LSTs, prompted the choice of a time step no longer than hourly. For these reasons, the HIRS data were assigned (binned) to a year-day hourly time if the observation fell within that hour. A data file for each hour of a year-day was created for the 1979–2009 period.

2) Cloud albedo

For assessment of the HIRS retrievals for clear-sky conditions and for later analysis, the cloud albedo was computed following the procedure of Betts (2009). For consistency with the CFSR LST, the CFSR incoming shortwave (SW) radiation, which was bias corrected using the SW radiation of the Princeton measures dataset (Ma and Pinker 2012), was selected as the observed incoming SW radiation for the cloud albedo calculations. The theoretical, clear-sky, incoming SW radiation was then computed by summing the beam and diffuse components of the incoming SW radiation. This theoretical, clear-sky SW radiation accounts for the latitude, time of day, and sun-incidence angle through the calculation of the zenith angle for each grid. The calculations for the beam and diffuse components are based on Campbell and Norman (1998) in which SC is the solar constant of 1366 W m−2, T is the transmissivity of the atmosphere and taken as 0.75 (Campbell and Norman 1998), ψ is the zenith angle, and m is the air mass number P/[P cos(ψ)], where P is the atmospheric pressure at sea level and assumed to be 101.3 kPa:
e1
e2
e3
e4

The cloud albedo was then calculated by dividing the observed incoming surface SW radiation by the theoretical incoming SW radiation from Eq. (4) such that a cloud albedo close to 1 corresponds to a clear-sky grid cell with minimal atmospheric interference, and a cloud albedo close to 0 corresponds to a cloudy-sky grid cell with near-maximum atmospheric interference. While we refer to this estimate as cloud albedo and the reduction of surface SW radiation is primarily affected by clouds, we recognize that this also includes other processes that reduce surface SW radiation, such as aerosols, dust, black carbon, atmospheric gas absorption, and so forth.

Table 2 displays how many HIRS data points and what percentage of all HIRS data points are classified as clear sky using as a threshold specified cloud albedo values that range from 0.2 to 0.75. This provides an assessment of the consistency of the HIRS cloud screening with regard to the bias-corrected CFSR radiation dataset—bias corrected at the monthly time scale (Coccia et al. 2015) using the measures of Princeton’s SW radiation dataset (Ma and Pinker 2012). This assessment shows that 58.3% of the HIRS (clear sky) retrievals occurred when the cloud albedo was greater than 0.75 and increasing to 83.4% when the cloud albedo threshold was lowered to 0.4. This agreement between clear-sky conditions at the time of the HIRS retrievals with clear-sky conditions assessed using the CFSR radiation demonstrates consistency between the conditions in the models producing the CFSR-based LST with the conditions at the time of the HIRS retrievals. The differences between the lower cloud albedo values and the apparent clear-sky conditions for the HIRS retrievals can be attributed to errors in the bias-corrected CFSR radiation (model errors) and temporally averaged radiation (timing errors), among other factors.

Table 2.

The percentage of total HIRS observations that are clear sky using given cloud albedo (CA) thresholds.

Table 2.

b. CFSR LST dataset

The choice of an hourly time step for the final dataset and subsequent analysis requires an hourly dataset for the LST model estimates. Since the estimated model LSTs are to match temporally the HIRS retrievals, accurate time fidelity is critical. This requirement resulted in choosing the reanalysis from NCEP’s CFSR because they are available at an hourly time step. The original spatial resolution of the CFSR LST is spectral T362 (~0.33° Gaussian grid), so the estimates were regridded to 0.5° resolution to match the resolution of the gridded HIRS retrievals. Therefore, the reanalysis dataset used for the analysis is an hourly averaged LST time series dataset at 0.5° resolution, globally from January 1979 through 2009. The LST is based on assuming a surface emissivity of 1.0, which provides appropriate consistency with the retrieved HIRS dataset.

3. Assessment of the HIRS and CFSR LST datasets

a. Comparisons with BSRN-derived LST

Comparisons of land model output, remotely sensed data, and in situ data have previously been done to assess how well the datasets agree (e.g., Wang et al. 2014), and validation of LST has previously been done using various methods including temperature-based validation (e.g., Guillevic et al. 2014) and radiance-based validation (e.g., Wan and Li 2008). The temperature-based validation in this study of both the retrieved HIRS LST and estimated LST is based upon data from 12 BSRN sites (Ohmura et al. 1998), shown in Fig. 1 and listed in Table 3. The measured upward LW data were averaged to an hourly time step and converted into LST, assuming 1.0 for the IR emissivity to provide consistency with the other datasets. The BSRN observations were then used to evaluate the HIRS LST retrievals for coincident times. This allowed for a separate assessment of HIRS retrieval errors for the day- or nighttime retrievals and from different NOAA satellites. In a similar manner, the BSRN observations were used to validate the CFSR LST, providing confidence in CFSR LST estimates under all conditions, and the joint intercomparisons offered a quantification of the errors between HIRS and the CFSR estimates. Finally, the CFSR LST and HIRS LST were compared over the BSRN sites, and a regional analysis and seasonal analysis were completed to assess the HIRS–CFSR differences by land-cover type and by season.

Fig. 1.
Fig. 1.

The locations of the BSRN sites with upward longwave radiation data used in this study.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

Table 3.

Station codes, lon and lat coordinates, names, and duration of record for the 12 BSRN stations used in this analysis.

Table 3.

It was deemed important to validate the CFSR LST at each of the BSRN station locations, using the BSRN-derived LST, to provide confidence that the reanalysis-based LST forms a viable basis for estimating LST globally under both clear-sky and cloudy conditions. For this, all available BSRN data at each station were used. In this validation, after subtracting three standard deviations from the mean of each time series for each station separately, the lowest threshold BSRN LST of 209 K was chosen as the cutoff threshold for valid data to remove unrealistic outliers. Figure 2 displays a scatterplot of the daytime CFSR and BSRN-derived LST at each of the stations. Including all stations, the total daytime average difference is −1.9 K, when the BSRN-derived mean is subtracted from the CFSR mean. We will define this difference as a bias, thus implying that the BSRN estimates are the true values. This negative daytime bias of the CFSR LST is consistent with negative daytime biases found in Zheng et al. (2012). The bias, root-mean-square error (RMSE), and Pearson correlation coefficient are displayed for each station for both daytime and nighttime comparisons in Table 4. For Tables 4 through 8, we completed a one-sample Student’s t test on each bias, and we calculated the confidence interval on the Pearson correlation coefficients based on Milton and Arnold (1986). We use an asterisk to denote statistically significant biases at the 95% confidence level, and we provide the 95% confidence interval for the correlation coefficients in brackets. The largest daytime biases occur at the sites of Desert Rock, Nevada; Boulder, Colorado; and Tateno, Japan, and the largest nighttime biases occur at Fort Peck, Montana, and Sioux Falls, South Dakota. The total nighttime bias is less than the daytime bias, but both are within 2 K. Additionally, the total correlation coefficients are 0.95 or above, and the total RMSE values are less than 4.8 K, demonstrating the high quality of the CFSR-based LST compared to the BSRN-derived LST.

Fig. 2.
Fig. 2.

The daytime validation of CFSR LST at the 12 BSRN sites using the BSRN-derived LST.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

Table 4.

The daytime and nighttime statistics between CFSR LST and BSRN-derived LST including average difference (bias), RMSE, correlation coefficient (CC), and number of data points (NDP). An asterisk indicates that a bias is statistically different from zero at the 95% confidence level, and the range between brackets is the 95% confidence interval for each CC. Bias is calculated as the difference CFSR − BSRN.

Table 4.
Table 5.

As in Table 4, but for the daytime statistics between HIRS LST and CFSR LST as well as between HIRS LST and BSRN-derived LST. Bias is calculated as the differences (left) HIRS − CFSR and (right) HIRS − BSRN.

Table 5.
Table 6.

As in Table 4, but for the nighttime statistics between HIRS LST and CFSR LST as well as between HIRS LST and BSRN-derived LST. Bias is calculated as the differences (left) HIRS − CFSR and (right) HIRS − BSRN.

Table 6.
Table 7.

As in Table 4, but for the daytime and nighttime statistics between HIRS LST retrievals and the final LST. Bias is calculated as the difference HIRS − final.

Table 7.
Table 8.

As in Table 4, but for the daytime and nighttime statistics between BSRN LST and the final LST. Bias is calculated as the difference final − BSRN.

Table 8.

The same statistics are detailed for each NOAA satellite for the HIRS LST in Tables 5 and 6, averaged across all the BSRN stations, for daytime and nighttime, respectively, to demonstrate that the HIRS LST retrievals also have a strong relationship with the BSRN-derived LST as well as with the CFSR LST. The strong relationship between HIRS and the BSRN stations is particularly seen during daytime with a total bias of 0.42 K, a total correlation coefficient of 0.94, and a total RMSE value of around 5.5 K. Between satellites, the highest correlation coefficient during daytime and nighttime, as well as some of the lowest RMSE values, occurred for satellite N15. The lowest correlations (during both daytime and nighttime retrievals) were N11, N12, and N14. For this validation and any other comparisons using HIRS LST, only HIRS LSTs below 350 and above 220 K were considered as valid data (Li et al. 2013).

b. Intercomparison between HIRS LST retrievals and CFSR LST over BSRN stations

The HIRS-retrieved LST time series from each satellite, over each BSRN station location, was compared with the matching time series of CFSR LST to assess the relationship between the two datasets, as shown in Fig. 3. The average difference, RMSE, and correlation coefficient for each satellite are also displayed in Tables 5 and 6. Although the total daytime difference between the HIRS and CFSR LST is 2.09 K, which is among the largest daytime differences in Tables 46, the total daytime correlation is still at 0.95, which is higher than the correlation for the HIRS and BSRN LST. The total nighttime difference between HIRS and CFSR is very small, 0.72 K. As shown in Table 6, the nighttime difference is smaller than the daytime difference for each satellite except N06, which we attribute to the lower spatial variability in nighttime LST, which is accentuated by the differences in scale between the BSRN measurements and CFSR or HIRS. The total nighttime RMSE, 4.6 K, is also lower than the total daytime RMSE. Overall, for both daytime and nighttime, the relationship between HIRS and CFSR LST is strong, with a difference no greater than 4 K for any satellite, a correlation coefficient no less than 0.93, and an RMSE no greater than 7.6 K. The statistics vary between satellites, but the largest difference of 3.9 K occurs for N09 during the daytime.

Fig. 3.
Fig. 3.

The HIRS LST compared with the CFSR LST at all 12 BSRN sites for each of the 11 satellites.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

c. Variability of HIRS LST and CFSR LST with respect to BSRN-derived LST

The variability within the HIRS retrievals from each satellite and the corresponding CFSR LST time series is examined using a Taylor diagram, shown in Fig. 4. The BSRN-derived LST is taken as truth, to which the HIRS and CFSR time series are compared. To display the variability between statistics of satellite time series relative to one another on the same plot, the average variance was calculated from all the BSRN-derived LST time series corresponding with each satellite time period, and the square root of this average was used as the mean standard deviation. The HIRS and CFSR standard deviations for each satellite were adjusted by the corresponding amount for each satellite separately to normalize them to the mean standard deviation. The data corresponding to a high majority of the satellites fall within 0–6 K for root-mean-square difference and 0.9–0.97 for correlation coefficient. The variability of the LST time series, as measured by their standard deviations, ranging from 10 to 15 K, but the data corresponding to retrievals from each satellite and the corresponding CFSR time series are slightly different among satellites. The time series of HIRS and CFSR LST that show the largest differences in variability from the HIRS and CFSR time series for the majority of satellites are the time series of LST corresponding to N11 and N14.

Fig. 4.
Fig. 4.

The Taylor diagram for the HIRS LST (H) for each satellite and corresponding time series of CFSR LST (C) each with respect to the corresponding BSRN-derived LST time series (B) (measured in kelvin).

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

d. Variability of HIRS LST and CFSR LST by land-cover type and season

As for regional variation between HIRS and CFSR LST, Fig. 6 displays the average differences for specific satellites at grid cells that are greater than or equal to 60% covered by a given dominant land-cover type shown in Fig. 5. Figure 6a shows the average difference for N06, which provides an example of one pattern of differences that is similar for N06, N08, N10, and N12. This pattern generally shows the highest differences for the Amazon region of up to 5 or 6 K; large differences over India, parts of eastern Africa, and parts of Australia; and differences of up to, or exceeding, 8 K over the Tibetan Plateau and parts of northern Africa. The difference pattern for N15 is shown in Fig. 6c specifically because it has the most unique regional patterns of all the satellites. It shows lower differences than other satellite groups in that there are few areas showing average differences above 4 or 5 K, with higher differences in small areas of North Africa and very low differences in small areas of north-central Africa and central Eurasia.

Fig. 5.
Fig. 5.

Predominant vegetation types used for grid cells whose dominant vegetation equals or exceeds 60% of the gridcell area. The types are 1–12 (corresponding color to left of number) as follows: from left to right, evergreen needleleaf, evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, mixed cover, woodland, wooded grasslands, closed shrublands, open shrublands, grasslands, croplands, and bare soil [see Liang et al. (1994) or Hansen et al. (2000) for detailed description of each type].

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

Fig. 6.
Fig. 6.

The LST bias (K) for (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

The other six satellites all have differences that are very similar to the difference patterns shown in Figs. 6b and 6d for N09 and N16, respectively. The primary differences for this satellite group from the first are that the Amazon region shows differences as low as −2 or −3 K and lower differences are seen for central Africa between −3 and 1 K. This satellite group also shows far higher average differences in the regions of northern and southern Africa as well as western Australia with differences ranging from 2 to 8 K. Figure 7 displays the global patterns of the correlation coefficient for specific satellites, using the same land-cover types as shown in Fig. 5. The correlation among satellites has a smaller range than the range among the average differences, but differences can be seen between the two groups of satellites as well. The satellite group correlation coefficients represented by N06 in Fig. 7a shows very high correlations ranging from 0.9 to 1 for most of North America and northern Eurasia and high correlations between 0.7 and 1 in all other areas except the tropics. In the tropics, the correlation coefficients are below 0.6 in the Amazon region, central Africa, and most of Indonesia, and even below 0.3 in many of those areas. Figures 7b and 7d display the correlation coefficients for N09 and N16, respectively, which are similar to N06 in North America and northern Eurasia but are as low as 0.4 or 0.5 in the tropics. In this way, the group represented by N09 and N16 has higher correlations globally than the group represented by N06. Figure 7c also displays the correlation coefficients for N15. The correlation coefficients for this satellite are similar to those of the group including N09 and N16, but the coefficients are closer to 0.7 and 0.8 and extend farther throughout the tropical regions.

Fig. 7.
Fig. 7.

The Pearson correlation coefficient for (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

The RMSE for N12, N09, N15, and N16 is shown in Figs. 8a–d, respectively. The satellites that are grouped with N12 have RMSE values between 3 and 6 K over most of the world, except for higher RMSE values at high latitudes and very high RMSE values in specific areas of northern Africa and Eurasia, as seen in Fig. 8a. As with the bias and correlation coefficient, the RMSE for satellite N15 is more similar to the RMSE for N06 than N09 and N16 but has even lower RMSE values below 3 K, which can be seen in South America and Africa. In contrast, overall for the group of satellites represented by N09 and N16, the RMSE values are higher, reaching above 6 K in much of Africa, Eurasia, and Australia. Finally, the number of satellite retrievals is displayed for the duration of four example satellites, which are N06, N09, N15, and N16, shown in Figs 9a–d, respectively. In all cases, the fewest data points are retrieved in the tropics, and the most data points are retrieved in the arid regions of the world. In Figs. 9a,b,d, the satellite records are all approximately four years, and the number of data values is generally similar at about 1200 in the Northern Hemisphere and areas south of the tropics and as few as 200 data values in the tropics. In Fig. 9c, many more total data values (retrievals) are used for N15 (up to around 2800), particularly in North Africa, the Arabian Peninsula, Australia, and the very high northern latitudes. The example satellites in Figs. 9a,b,d all represent groups of other satellites, each of which includes satellites of longer records, such as approximately 6–7 years, which are not shown in this figure, while the satellite shown in Fig. 9c is the satellite with the longest record of approximately 10 years.

Fig. 8.
Fig. 8.

The LST RMSE (K) for (a) N12, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

Fig. 9.
Fig. 9.

The number of data points for the duration of each of (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

For evaluation of seasonal variability, Fig. 10 displays the April through September and October through March average differences using all seasons from all satellites for all grid cells that have at least five data values per season (the white grid cells in the tropics have less than five data values for a season). Larger interseasonal differences occur in the Northern Hemisphere overall, as parts of central North America have a negative difference between October and March but a positive difference between April and September, Southeast Asia has a higher difference between October and March, and most of Eurasia has a lower difference between October and March. Additionally, the Tibetan Plateau and areas of central Asia have low differences in April through September, but significantly the Tibetan Plateau has a much higher difference in October through March when frozen ground and snow cover are extensive.

Fig. 10.
Fig. 10.

The average (a) April–September and (b) October–March LST biases (K) from all satellites. The white grid cells over land occur when one or more seasons for a given satellite have less than five data points, and therefore the bias is not calculated for that grid cell.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

4. Merging the HIRS and CFSR LST datasets

a. Bayesian model conditional processor

The HIRS and CFSR LST datasets have been merged using the Bayesian methodology proposed by Coccia et al. (2015), which is based on the assessment of the posterior distribution of HIRS LST conditional on the estimates provided by CFSR. Since HIRS LST is retrieved from different satellites that, as shown in the previous sections, are not completely consistent among one another, a posterior distribution has been assessed separately for each satellite in order to get 11 different ensembles, each one consistent with one specific satellite. It should be noted again that the GDAP has specified that the HIRS retrievals will be the target dataset in that other LST information (i.e., CFSR) will be made consistent with the HIRS retrievals. To achieve this the CFSR LST time series is first adjusted to be consistent with each HIRS satellite retrieval dataset resulting in the 11 ensembles—the 11 Bayesian posterior distributions. These posterior distributions are then merged through a distribution averaging method based on the linear opinion pool (LOP) approach (Stone 1961; Clemen and Winkler 1999).

All the posterior distributions have been estimated using the model conditional processor (MCP) approach (Coccia and Todini 2011; Todini 2008). The MCP is a Bayesian postprocessor that takes into account the heteroscedasticity of the errors while maintaining a parsimonious parameterization. In the MCP approach, the a priori distributions of both the predictand (HIRS LST) and the predictor (CFSR LST) are assumed to be equal to their climatological distributions, and they are transformed to a normal probability distribution space using the normal quantile transform (NQT) (Van der Waerden 1952, 1953a,b). To take into account the heteroscedasticity of the errors, in the normal space the joint distribution of predictand and predictor is assumed to be comprised of two truncated normal distributions. These two distributions split the joint distribution into two parts, and for each part the Gaussian assumption can be established. Given the a priori distributions of predictand and predictor and their joint distribution, Bayes’s theorem can be easily applied to derive the conditional distribution of the predictand given the predictor’s estimate. Finally, the conditional distribution is converted back from the normal space to the original probability space using the inverse of the NQT.

As said earlier, each satellite is processed individually resulting in 11 ensembles (posterior distributions) with each being conditioned on a specific satellite. Then these posterior distributions are merged using the LOP approach. Often, other approaches of intersensor calibration between satellites do not remove all differences between sensors resulting from the limited overlap of the sensors, and selecting different weights for each sensor can be a difficult challenge. Therefore, the LOP approach is the most practical solution in this case. The result is a dataset consistent with the HIRS LST retrievals provided by all the satellites and composed of a reliable probability distribution function for each grid and each time step. Full details of the applied methodology are in Coccia et al. (2015).

b. Assessment of the merged dataset

The final merged HIRS-consistent LST dataset is compared with the individual HIRS retrievals as well as the BSRN-derived LST at the 12 BSRN stations in Tables 7 and 8 and in Fig. 11. Across all stations and all satellites, the daytime average differences have been decreased by approximately 1–2 K compared with the daytime average differences between CFSR and HIRS LST (Tables 5 and 6). One example of this reduction is the reduction in total daytime average difference from 2.09 to 0.69 K. The nighttime average differences have also been lowered by approximately 1–2 K in most cases, with a total nighttime difference of 0.72 K reduced to 0.01 K. Daytime and nighttime RMSE values in Table 7 are between 3.4 and 5.5 K, which have been reduced from an RMSE larger than 6–7 K for CFSR when compared with HIRS. The correlation coefficients have increased by 0.01 (0.02 for nighttime) for each satellite such that the values are above 0.96 for both daytime and nighttime overpasses (see Table 7).

Fig. 11.
Fig. 11.

The comparison of the final LST estimates at the 12 BSRN sites with the BSRN-derived LST.

Citation: Journal of Climate 29, 10; 10.1175/JCLI-D-15-0378.1

The relationship of the final LST is seen with the BSRN LST in Table 8, and improvements from the same comparison with CFSR (Table 4) are seen in the reduction of the bias. Decreases in the bias with BSRN LST are seen for all stations except for daytime at stations BON, BOS, FPE, NYA, PAY, and SXF. While reductions in bias are not seen for all stations, the daytime bias was reduced from −1.93 to −0.66 K. This reduction of bias can be seen by comparing Fig. 11 with Fig. 2. Although nighttime biases follow a different pattern from the comparison with CFSR and with the final LST, these biases all remain below 4 K. The correlation coefficients for the final LST and BSRN LST remain approximately similar to the correlation between CFSR and BSRN, with the exception of daytime at BOS, which reduces to 0.91 from 0.95. Additionally, RMSE values for daytime either reduce or remain similar between Tables 4 and 8, with the exception of daytime at BOS. Figure 11 also shows the exception of the BOS station as the outlier station that does not visually show improvement or remain similar between Figs. 2 and 11. We attribute this to the station location and significant heterogeneity within the 0.5° HIRS pixel (CFSR grid).

5. Discussion

a. Validation of CFSR and HIRS LST using BSRN-derived LST

When validating the CFSR LST against the BSRN-derived LST for each of the available station locations, the statistics of the temperature differences are very similar overall across all stations, as seen in Table 4. In total, during the daytime, the bias shows that overall the CFSR LST underestimates the BSRN-derived LST on average by 1.93 K. Large biases in individual sites relative to the average bias may be explained by sites at which the environment of the tower site is not fully representative of the land cover for the CFSR grid cell corresponding to the tower site location. The smaller bias found at night at many sites, and in total, could be attributed to lower spatial heterogeneity in nighttime LST so that the tower is more representative of the grid-average LST captured by the reanalysis model.

A strong relationship is also seen between the HIRS LST and BSRN-derived LST in Tables 5 and 6. The total daytime bias indicates that for all sites and satellites combined, the HIRS LST is slightly higher than the BSRN-derived LST, on average. The highest correlation coefficients and some of the lowest RMSE values are seen for N15, which could be because this satellite has the longest record (10 yr), and therefore most of the anomalous data will be smoothed out when the means are calculated over the total time period of the satellite duration at each site. There are larger differences between nighttime and daytime LST, which could be caused by a decreased ability to accurately filter out for cloudy conditions in HIRS nighttime retrievals. The bias at night is also larger for this comparison than for the CFSR versus BSRN-derived LST comparison, which could also be related to this potential problem of hard-to-detect cloudy conditions at nighttime. Similar to the previous case, some of the variability between the two datasets can be attributed to the difference between the local environment of the tower site and the pixel-average environment, which is the environment observed from the satellite.

b. Relationship between HIRS retrievals and CFSR LST

The relationship between HIRS LST and CFSR LST, shown in Fig. 3 and Tables 5 and 6, is strong, with a correlation coefficient of 0.95. The total daytime difference of 2.09 K results from the HIRS LST overestimating the BSRN-derived LST during the daytime by a little less than 0.5 K and the CFSR land surface model–based LST underestimating the BSRN-derived LST during the daytime by an average of a little less than 2 K. In this case, the total nighttime difference is much smaller than the daytime difference between HIRS and CFSR LST, and this observation could occur as a result of increased variability during the daytime for HIRS retrievals because of shadows from vegetation in some areas and variable land cover and topography within the pixel, which increase the spatial heterogeneity in LST. These sources of variability are not included in the CFSR land model component.

c. Variability of HIRS LST and CFSR LST with respect to BSRN-derived LST

Figure 4 shows that the variability of each time series of CFSR and HIRS LST is different among different satellites. The time series that have the largest difference in variability from the main group are the HIRS LST and the corresponding CFSR LST from N11 and N14. One possible reason for the shift in CFSR and HIRS LST variability for these satellites is that these satellites were launched with a 1330 equatorial crossing time, which is different from the morning launches of other satellites and the other afternoon launch included in this plot (1400 for N16). This difference in time of day could result in the lower relative standard deviations for the HIRS and CFSR time series for N11 or the lower correlations of the HIRS and CFSR time series for the both satellites.

d. Variability of HIRS LST and CFSR LST by land-cover type and season

Regionally, none of the statistics shown in Figs. 68 vary significantly by land-cover type. The variation within each statistic is organized into two main groups with one outlier in each case. N06, N08, N10, and N12 are grouped together for their regional patterns of each statistic. The one outlier is N15, which is most similar to this first group in each case but is unique. The remaining satellites make up the majority group, which is represented by N09 and N16 in Figs. 68. The feature that all of the satellites in the first group have in common is their early morning equatorial crossing time (see Table 1). Although the time of day for measurement appears to be the dividing factor between satellite statistics, N15 is an early morning satellite, but it is the outlier with its unique pattern that is quite different, as can be seen in Figs. 68. One reason this satellite could be showing statistics that are different from the others is that it has a far longer record (~10 yr) compared to the other early morning satellites. This longer time period could include more orbital drift, which may set it apart from the rest of the early morning satellites. Other sources of variation in the statistic patterns among each of the satellites can include individual instrument bias. Overall, the differences in the patterns of each statistic among satellites are relatively minimal. Additionally, no single region consistently shows a different behavior compared to the others except for correlations (Fig. 7) in the tropics. This drastic difference in correlation in the tropics from the rest of the world could occur because of the consistently low number of retrievals in the tropics, as shown in Fig. 9, resulting from cloudiness.

Seasonally, parts of North America and Eurasia show the largest changes in the difference in October through March compared with April through September. The areas of North America and Eurasia that show lower differences in the winter months (October through March) compared with the summer (April through September) could be affected by frozen ground, snow cover, and the loss of vegetation in the winter—this latter effect would bring down the variability in observed HIRS LST because of the lack of shadowing from vegetation. In contrast, locations in central Asia, such as the Tibetan Plateau, show much lower differences in the summer months than in the winter months. This change could occur because of heterogeneity in snow cover over the Tibetan Plateau during the winter months, which is not modeled accurately in CFSR but would significantly affect the HIRS LST observations. Overall, the changes in average difference over most of the world at the seasonal scale are only around 2–4 K.

Finally, an additional note should be made about the inherent uncertainty in the comparison of the HIRS LST and the CFSR LST. There is additional uncertainty because the HIRS LSTs are instantaneous temperatures, which are assigned to the hourly files based upon the hour in which they were retrieved, while the CFSR LST is a time series of hourly averaged values. A strong enough relationship is still shown despite this inherent variability and uncertainty from this difference.

e. Final HIRS-consistent LST dataset

As indicated in section 4b, the statistics in Table 7 demonstrate a general reduction in the average difference of the final LST dataset with HIRS retrievals from the original comparisons of CFSR LST with HIRS retrievals in Tables 5 and 6. Not only have the average differences been reduced, but the correlation coefficients have generally increased and most of the RMSE values have decreased. The changes in each of these statistics at each of the satellites indicate that overall, the final, merged LST dataset is consistent with the HIRS retrievals from each of the satellites across the 12 BSRN stations.

In comparing the final LST data product with the BSRN-derived LST in Table 8, many of the biases at the stations in the daytime are reduced, indicating a final data product that aligns more closely with the observed tower data in many cases. Because the final data product was only made consistent with the HIRS retrievals, differences are still expected with the BSRN-derived LST, as the HIRS retrievals themselves are not fully consistent with the BSRN data. With most daytime correlation coefficients remaining above 0.92 and most biases remaining within 3 K, the final LST dataset is still overall well validated by the BSRN-derived LST. One exception to these conclusions is the BOS site, which shows an increased bias, increased RMSE, and decreased correlation coefficient in Table 8 compared with the statistics in Table 4. The results for the final LST at this site could occur as an effect of the MCP and distribution averaging such that this particular location shows a larger daytime bias with BSRN data than the other site locations (see Fig. 11). Further investigation of this particular site, including its placement and changes in the land cover within the overlying grid cell (e.g., urbanization around Boulder, Colorado), is needed but is beyond the scope of this study.

6. Conclusions

A new high-resolution (0.5°), long-term (1979–2009), temporally and spatially consistent, HIRS-consistent LST global dataset has been formed at an hourly time scale using remote sensing and land surface modeling. Validation of the original, retrieved dataset temperatures using BSRN tower data has been demonstrated. Variation between satellites with respect to BSRN-derived LST as well as at the regional and spatial scale is present but has been attributed to differences in equatorial crossing times as well as instrument bias, orbital drift, and seasonal changes in vegetation or land cover. Regional and seasonal statistics between the original HIRS LST and CFSR LST were examined and, while differences were present, these differences were small in magnitude. This new dataset will help build a foundation of consistent global datasets to be used for environmental and meteorological modeling. Future work should extend this dataset up to the present if new data become available and should focus on developing other consistent global datasets so the global water and energy cycles can be better understood using more accurate and robust model simulations.

Acknowledgments

This research was supported by NASA Grants NNX09AK35G (development and diagnostic analysis of a multi-decadal global evaporation product for NEWS), NNX08AN40A (developing consistent Earth system data records for the global terrestrial water cycle), and NOAA Grant NA11OAR4310175 [improving land evaporative processes and land–atmosphere interactions in the NCEP Global Forecast System (GFS) and Climate Forecast System (CFS)]. The dataset contributes to the GEWEX Data and Assessments Panel LandFlux activities as well as the development of updated surface radiation products. We thank Lei Shi at NOAA/National Climatic Data Center for providing the reprocessed HIRS satellite data. Computational resources were made available through the Princeton Institute for Computational Science and Engineering (PICSciE).

REFERENCES

  • Anderson, M. C., C. Hain, B. Wardlow, A. Pimstein, J. R. Mecikalski, and W. P. Kustas, 2011: Evaluation of drought indices based on thermal remote sensing of evapotranspiration over the continental United States. J. Climate, 24, 20252044, doi:10.1175/2010JCLI3812.1.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., 2009: Land-surface-atmosphere coupling in observations and models. J. Adv. Model Earth Syst., 1, 4, doi:10.3894/JAMES.2009.1.4.

    • Search Google Scholar
    • Export Citation
  • Campbell, G. S., and J. M. Norman, 1998: Radiation fluxes in natural environments. An Introduction to Environmental Biophysics, G. S. Campbell and J. M. Norman, Eds., Springer, 167–184.

  • Clemen, R. T., and R. L. Winkler, 1999: Combining probability distributions from experts in risk analysis. Risk Anal., 19, 187203, doi:10.1111/j.1539-6924.1999.tb00399.x.

    • Search Google Scholar
    • Export Citation
  • Coccia, G., and E. Todini, 2011: Recent developments in predictive uncertainty assessment based on the model conditional processor approach. Hydrol. Earth Syst. Sci., 15, 32533274, doi:10.5194/hess-15-3253-2011.

    • Search Google Scholar
    • Export Citation
  • Coccia, G., A. Siemann, M. Pan, and E. F. Wood, 2015: Creating consistent datasets combining remotely-sensed data and land surface model estimates through a Bayesian uncertainty post-processing: The case of land surface temperature from HIRS. Remote Sens. Environ., 170, 290305, doi:10.1016/j.rse.2015.09.010.

    • Search Google Scholar
    • Export Citation
  • Dash, P., F.-M. Göttsche, F.-S. Olesen, and H. Fischer, 2002: Land surface temperature and emissivity estimation from passive sensor data: Theory and practice-current trends. Int. J. Remote Sens., 23, 25632594, doi:10.1080/01431160110115041.

    • Search Google Scholar
    • Export Citation
  • Duan, S.-B., Z.-L. Li, B.-H. Tang, H. Wu, and R. Tang, 2014: Generation of a time-consistent land surface temperature product from MODIS data. Remote Sens. Environ., 140, 339349, doi:10.1016/j.rse.2013.09.003.

    • Search Google Scholar
    • Export Citation
  • Earth System Grid Federation, 2015: Criteria, Process and Resources for contributing data to Obs4MIPs. World Climate Research Programme, accessed 11 January 2016. [Available online at https://www.earthsystemcog.org/projects/obs4mips/how_to_contribute.]

  • Gao, H., Q. Tang, C. R. Ferguson, E. F. Wood, and D. P. Lettenmaier, 2010: Estimating the water budget of major US river basins via remote sensing. Int. J. Remote Sens., 31, 39553978, doi:10.1080/01431161.2010.483488.

    • Search Google Scholar
    • Export Citation
  • Guillevic, P. C., and Coauthors, 2014: Validation of land surface temperature products derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) using ground-based and heritage satellite measurements. Remote Sens. Environ., 154, 1937, doi:10.1016/j.rse.2014.08.013.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., R. S. Defries, J. R. G. Townshend, and R. Sohlberg, 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 13311364, doi:10.1080/014311600210209.

    • Search Google Scholar
    • Export Citation
  • Ignatov, A., I. Laszlo, E. D. Harrod, K. B. Kidwell, and G. P. Goodrum, 2004: Equator crossing times for NOAA, ERS and EOS sun-synchronous satellites. Int. J. Remote Sens., 25, 52555266, doi:10.1080/01431160410001712981.

    • Search Google Scholar
    • Export Citation
  • Jin, M., 2000: Interpolation of surface radiation temperature measured from polar orbiting satellites to a diurnal cycle. Part 2: Cloudy-pixel treatment. J. Geophys. Res., 105, 40614076, doi:10.1029/1999JD901088.

    • Search Google Scholar
    • Export Citation
  • Jin, M., and R. E. Dickinson, 1999: Interpolation of surface radiation temperature measured from polar orbiting satellites to a diurnal cycle. Part 1: Without clouds. J. Geophys. Res., 104, 21052116, doi:10.1029/1998JD200005.

    • Search Google Scholar
    • Export Citation
  • Li, F., T. J. Jackson, W. P. Kustas, T. J. Schmugge, A. N. French, M. H. Cosh, and R. Bindlish, 2004: Deriving land surface temperature from Landsat 5 and 7 during SMEX02/SMACEX. Remote Sens. Environ., 92, 521534, doi:10.1016/j.rse.2004.02.018.

    • Search Google Scholar
    • Export Citation
  • Li, Z.-L., B.-H. Tang, H. Wu, H. Ren, G. Yan, Z. Wan, I. F. Trigo, and J. A. Sobrino, 2013: Satellite-derived land surface temperature: Current status and perspectives. Remote Sens. Environ., 131, 1437, doi:10.1016/j.rse.2012.12.008.

    • Search Google Scholar
    • Export Citation
  • Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 41514 428, doi:10.1029/94JD00483.

    • Search Google Scholar
    • Export Citation
  • Liang, X., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) phase 2(c) Red-Arkansas River Basin Experiment: 2. Spatial and temporal analysis of energy fluxes. Global Planet. Change, 19, 137159, doi:10.1016/S0921-8181(98)00045-9.

    • Search Google Scholar
    • Export Citation
  • Lohmann, D., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) phase 2(c) Red-Arkansas River Basin Experiment. 3. Spatial and temporal analysis of water fluxes. Global Planet. Change, 19, 161179, doi:10.1016/S0921-8181(98)00046-0.

    • Search Google Scholar
    • Export Citation
  • Ma, Y., and R. T. Pinker, 2012: Modeling shortwave radiative fluxes from satellites. J. Geophys. Res., 117, D23202, doi:10.1029/2012JD018332.

    • Search Google Scholar
    • Export Citation
  • Milton, J. S., and J. C. Arnold, 1986: Probability and Statistics in the Engineering and Computing Sciences. McGraw-Hill, 643 pp.

  • Moody, E. G., M. D. King, S. Platnick, C. B. Schaaf, and F. Gao, 2005: Spatially complete global spectral surface albedos: Value-added datasets derived from Terra MODIS land products. IEEE Trans. Geosci. Remote Sens., 43, 144158, doi:10.1109/TGRS.2004.838359.

    • Search Google Scholar
    • Export Citation
  • National Research Council, 2004: Climate Data Records from Environmental Satellites. National Academies Press, 150 pp., doi:10.17226/10944.

  • Nijssen, B., R. Schnur, and D. P. Lettenmaier, 2001: Global retrospective estimation of soil moisture using the Variable Infiltration Capacity land surface model, 1980–93. J. Climate, 14, 17901808, doi:10.1175/1520-0442(2001)014<1790:GREOSM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ohmura, A., and Coauthors, 1998: Baseline Surface Radiation Network (BSRN/WCRP): New precision radiometry for climate research. Bull. Amer. Meteor. Soc., 79, 21152136, doi:10.1175/1520-0477(1998)079<2115:BSRNBW>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Robel, J., Ed., 2009: NOAA KLM user’s guide. NOAA Climatic Data Center. [Available online at http://www.ncdc.noaa.gov/oa/pod-guide/ncdc/docs/klm/html/c8/sec831-5.htm.]

  • Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 10151057, doi:10.1175/2010BAMS3001.1.

    • Search Google Scholar
    • Export Citation
  • Şahin, M., B. Y. Yıldız, O. Şenkal, and V. Peştemalcı, 2012: Modelling and remote sensing of land surface temperature in Turkey. J. Indian Soc. Remote Sens., 40, 399409, doi:10.1007/s12524-011-0158-3.

    • Search Google Scholar
    • Export Citation
  • Sahoo, A. K., M. Pan, T. J. Troy, R. K. Vinukollu, J. Sheffield, and E. F. Wood, 2011: Reconciling the global terrestrial water budget using satellite remote sensing. Remote Sens. Environ., 115, 18501865, doi:10.1016/j.rse.2011.03.009.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J., C. J. Tucker, G. J. Collatz, S. O. Los, C. O. Justice, D. A. Dazlich, and D. A. Randall, 1996: A revised land surface parameterization (SiB2) for atmospheric GCMS. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate, 9, 706737, doi:10.1175/1520-0442(1996)009<0706:ARLSPF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Shi, L., 2011: Global atmospheric temperature and humidity profiles based on intersatellite calibrated HIRS measurement, computational intelligence methods for remote sensing and analysis. Proc. Ninth Conf. on Artificial Intelligence and Its Applications to the Environmental Sciences, Seattle, WA, Amer. Meteor. Soc., J6.3.

  • Shi, L., 2013: Intersatellite differences of HIRS longwave channels between NOAA-14 and NOAA-15 and between NOAA-17 and METOP-A. IEEE Trans. Geosci. Remote Sens., 51, 14141424, doi:10.1109/TGRS.2012.2216886.

    • Search Google Scholar
    • Export Citation
  • Stone, M., 1961: The opinion pool. Ann. Math. Stat., 32, 13391342, doi:10.1214/aoms/1177704873.

  • Sun, D., and R. T. Pinker, 2003: Estimation of land surface temperature from a Geostationary Operational Environmental Satellite (GOES-8). J. Geophys. Res., 108, 4326, doi:10.1029/2002JD002422.

    • Search Google Scholar
    • Export Citation
  • Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Todini, E., 2008: A model conditional processor to assess predictive uncertainty in flood forecasting. Int. J. River Basin Manage., 6, 123137, doi:10.1080/15715124.2008.9635342.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1952: Order tests for two-sample problem and their power I. Indag. Math., 55, 453458, doi:10.1016/S1385-7258(52)50063-5.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1953a: Order tests for two-sample problem (second communication). Indag. Math., 56, 303310, doi:10.1016/S1385-7258(53)50039-3.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1953b: Order tests for two-sample problem (third communication). Indag. Math., 56, 311316, doi:10.1016/S1385-7258(53)50040-X.

    • Search Google Scholar
    • Export Citation
  • Wan, Z., and Z.-L. Li, 2008: Radiance-based validation of the V5 MODIS land‐surface temperature product. Int. J. Remote Sens., 29, 53735395, doi:10.1080/01431160802036565.

    • Search Google Scholar
    • Export Citation
  • Wan, Z., Y. Zhang, Q. Zhang, and Z.-L. Li, 2002: Validation of the land-surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data. Remote Sens. Environ., 83, 163180, doi:10.1016/S0034-4257(02)00093-7.

    • Search Google Scholar
    • Export Citation
  • Wang, A., M. Barlage, X. Zeng, and C. S. Draper, 2014: Comparison of land skin temperature from a land model, remote sensing, and in situ measurement. J. Geophys. Res. Atmos., 119, 30933106, doi:10.1002/2013JD021026.

    • Search Google Scholar
    • Export Citation
  • Wood, E. F., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) Phase 2(c) Red-Arkansas River Basin Experiment. 1. Experiment description and summary intercomparisons. Global. Planet. Change, 19, 115135, doi:10.1016/S0921-8181(98)00044-7.

    • Search Google Scholar
    • Export Citation
  • Yang, S., W. S. Olson, J.-J. Wang, T. L. Bell, E. A. Smith, and C. D. Kummerow, 2006: Precipitation and latent heating distributions from satellite passive microwave radiometry. Part II: Evaluation of estimates using independent data. J. Appl. Meteor. Climatol., 45, 721739, doi:10.1175/JAM2370.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, C., W. Li, and D. Travis, 2007: Gaps-fill of SLC-off Landsat ETM+ satellite image using a geostatistical approach. Int. J. Remote Sens., 28, 51035122, doi:10.1080/01431160701250416.

    • Search Google Scholar
    • Export Citation
  • Zheng, W., H. Wei, Z. Wang, X. Zeng, J. Meng, M. Ek, K. Mitchell, and J. Derber, 2012: Improvement of daytime land surface skin temperature over arid regions in the NCEP GFS model and its impact on satellite data assimilation. J. Geophys. Res., 117, D06117, doi:10.1029/2011JD015901.

    • Search Google Scholar
    • Export Citation
Save
  • Anderson, M. C., C. Hain, B. Wardlow, A. Pimstein, J. R. Mecikalski, and W. P. Kustas, 2011: Evaluation of drought indices based on thermal remote sensing of evapotranspiration over the continental United States. J. Climate, 24, 20252044, doi:10.1175/2010JCLI3812.1.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., 2009: Land-surface-atmosphere coupling in observations and models. J. Adv. Model Earth Syst., 1, 4, doi:10.3894/JAMES.2009.1.4.

    • Search Google Scholar
    • Export Citation
  • Campbell, G. S., and J. M. Norman, 1998: Radiation fluxes in natural environments. An Introduction to Environmental Biophysics, G. S. Campbell and J. M. Norman, Eds., Springer, 167–184.

  • Clemen, R. T., and R. L. Winkler, 1999: Combining probability distributions from experts in risk analysis. Risk Anal., 19, 187203, doi:10.1111/j.1539-6924.1999.tb00399.x.

    • Search Google Scholar
    • Export Citation
  • Coccia, G., and E. Todini, 2011: Recent developments in predictive uncertainty assessment based on the model conditional processor approach. Hydrol. Earth Syst. Sci., 15, 32533274, doi:10.5194/hess-15-3253-2011.

    • Search Google Scholar
    • Export Citation
  • Coccia, G., A. Siemann, M. Pan, and E. F. Wood, 2015: Creating consistent datasets combining remotely-sensed data and land surface model estimates through a Bayesian uncertainty post-processing: The case of land surface temperature from HIRS. Remote Sens. Environ., 170, 290305, doi:10.1016/j.rse.2015.09.010.

    • Search Google Scholar
    • Export Citation
  • Dash, P., F.-M. Göttsche, F.-S. Olesen, and H. Fischer, 2002: Land surface temperature and emissivity estimation from passive sensor data: Theory and practice-current trends. Int. J. Remote Sens., 23, 25632594, doi:10.1080/01431160110115041.

    • Search Google Scholar
    • Export Citation
  • Duan, S.-B., Z.-L. Li, B.-H. Tang, H. Wu, and R. Tang, 2014: Generation of a time-consistent land surface temperature product from MODIS data. Remote Sens. Environ., 140, 339349, doi:10.1016/j.rse.2013.09.003.

    • Search Google Scholar
    • Export Citation
  • Earth System Grid Federation, 2015: Criteria, Process and Resources for contributing data to Obs4MIPs. World Climate Research Programme, accessed 11 January 2016. [Available online at https://www.earthsystemcog.org/projects/obs4mips/how_to_contribute.]

  • Gao, H., Q. Tang, C. R. Ferguson, E. F. Wood, and D. P. Lettenmaier, 2010: Estimating the water budget of major US river basins via remote sensing. Int. J. Remote Sens., 31, 39553978, doi:10.1080/01431161.2010.483488.

    • Search Google Scholar
    • Export Citation
  • Guillevic, P. C., and Coauthors, 2014: Validation of land surface temperature products derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) using ground-based and heritage satellite measurements. Remote Sens. Environ., 154, 1937, doi:10.1016/j.rse.2014.08.013.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., R. S. Defries, J. R. G. Townshend, and R. Sohlberg, 2000: Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens., 21, 13311364, doi:10.1080/014311600210209.

    • Search Google Scholar
    • Export Citation
  • Ignatov, A., I. Laszlo, E. D. Harrod, K. B. Kidwell, and G. P. Goodrum, 2004: Equator crossing times for NOAA, ERS and EOS sun-synchronous satellites. Int. J. Remote Sens., 25, 52555266, doi:10.1080/01431160410001712981.

    • Search Google Scholar
    • Export Citation
  • Jin, M., 2000: Interpolation of surface radiation temperature measured from polar orbiting satellites to a diurnal cycle. Part 2: Cloudy-pixel treatment. J. Geophys. Res., 105, 40614076, doi:10.1029/1999JD901088.

    • Search Google Scholar
    • Export Citation
  • Jin, M., and R. E. Dickinson, 1999: Interpolation of surface radiation temperature measured from polar orbiting satellites to a diurnal cycle. Part 1: Without clouds. J. Geophys. Res., 104, 21052116, doi:10.1029/1998JD200005.

    • Search Google Scholar
    • Export Citation
  • Li, F., T. J. Jackson, W. P. Kustas, T. J. Schmugge, A. N. French, M. H. Cosh, and R. Bindlish, 2004: Deriving land surface temperature from Landsat 5 and 7 during SMEX02/SMACEX. Remote Sens. Environ., 92, 521534, doi:10.1016/j.rse.2004.02.018.

    • Search Google Scholar
    • Export Citation
  • Li, Z.-L., B.-H. Tang, H. Wu, H. Ren, G. Yan, Z. Wan, I. F. Trigo, and J. A. Sobrino, 2013: Satellite-derived land surface temperature: Current status and perspectives. Remote Sens. Environ., 131, 1437, doi:10.1016/j.rse.2012.12.008.

    • Search Google Scholar
    • Export Citation
  • Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 41514 428, doi:10.1029/94JD00483.

    • Search Google Scholar
    • Export Citation
  • Liang, X., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) phase 2(c) Red-Arkansas River Basin Experiment: 2. Spatial and temporal analysis of energy fluxes. Global Planet. Change, 19, 137159, doi:10.1016/S0921-8181(98)00045-9.

    • Search Google Scholar
    • Export Citation
  • Lohmann, D., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) phase 2(c) Red-Arkansas River Basin Experiment. 3. Spatial and temporal analysis of water fluxes. Global Planet. Change, 19, 161179, doi:10.1016/S0921-8181(98)00046-0.

    • Search Google Scholar
    • Export Citation
  • Ma, Y., and R. T. Pinker, 2012: Modeling shortwave radiative fluxes from satellites. J. Geophys. Res., 117, D23202, doi:10.1029/2012JD018332.

    • Search Google Scholar
    • Export Citation
  • Milton, J. S., and J. C. Arnold, 1986: Probability and Statistics in the Engineering and Computing Sciences. McGraw-Hill, 643 pp.

  • Moody, E. G., M. D. King, S. Platnick, C. B. Schaaf, and F. Gao, 2005: Spatially complete global spectral surface albedos: Value-added datasets derived from Terra MODIS land products. IEEE Trans. Geosci. Remote Sens., 43, 144158, doi:10.1109/TGRS.2004.838359.

    • Search Google Scholar
    • Export Citation
  • National Research Council, 2004: Climate Data Records from Environmental Satellites. National Academies Press, 150 pp., doi:10.17226/10944.

  • Nijssen, B., R. Schnur, and D. P. Lettenmaier, 2001: Global retrospective estimation of soil moisture using the Variable Infiltration Capacity land surface model, 1980–93. J. Climate, 14, 17901808, doi:10.1175/1520-0442(2001)014<1790:GREOSM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ohmura, A., and Coauthors, 1998: Baseline Surface Radiation Network (BSRN/WCRP): New precision radiometry for climate research. Bull. Amer. Meteor. Soc., 79, 21152136, doi:10.1175/1520-0477(1998)079<2115:BSRNBW>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Robel, J., Ed., 2009: NOAA KLM user’s guide. NOAA Climatic Data Center. [Available online at http://www.ncdc.noaa.gov/oa/pod-guide/ncdc/docs/klm/html/c8/sec831-5.htm.]

  • Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 10151057, doi:10.1175/2010BAMS3001.1.

    • Search Google Scholar
    • Export Citation
  • Şahin, M., B. Y. Yıldız, O. Şenkal, and V. Peştemalcı, 2012: Modelling and remote sensing of land surface temperature in Turkey. J. Indian Soc. Remote Sens., 40, 399409, doi:10.1007/s12524-011-0158-3.

    • Search Google Scholar
    • Export Citation
  • Sahoo, A. K., M. Pan, T. J. Troy, R. K. Vinukollu, J. Sheffield, and E. F. Wood, 2011: Reconciling the global terrestrial water budget using satellite remote sensing. Remote Sens. Environ., 115, 18501865, doi:10.1016/j.rse.2011.03.009.

    • Search Google Scholar
    • Export Citation
  • Sellers, P. J., C. J. Tucker, G. J. Collatz, S. O. Los, C. O. Justice, D. A. Dazlich, and D. A. Randall, 1996: A revised land surface parameterization (SiB2) for atmospheric GCMS. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data. J. Climate, 9, 706737, doi:10.1175/1520-0442(1996)009<0706:ARLSPF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Shi, L., 2011: Global atmospheric temperature and humidity profiles based on intersatellite calibrated HIRS measurement, computational intelligence methods for remote sensing and analysis. Proc. Ninth Conf. on Artificial Intelligence and Its Applications to the Environmental Sciences, Seattle, WA, Amer. Meteor. Soc., J6.3.

  • Shi, L., 2013: Intersatellite differences of HIRS longwave channels between NOAA-14 and NOAA-15 and between NOAA-17 and METOP-A. IEEE Trans. Geosci. Remote Sens., 51, 14141424, doi:10.1109/TGRS.2012.2216886.

    • Search Google Scholar
    • Export Citation
  • Stone, M., 1961: The opinion pool. Ann. Math. Stat., 32, 13391342, doi:10.1214/aoms/1177704873.

  • Sun, D., and R. T. Pinker, 2003: Estimation of land surface temperature from a Geostationary Operational Environmental Satellite (GOES-8). J. Geophys. Res., 108, 4326, doi:10.1029/2002JD002422.

    • Search Google Scholar
    • Export Citation
  • Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Todini, E., 2008: A model conditional processor to assess predictive uncertainty in flood forecasting. Int. J. River Basin Manage., 6, 123137, doi:10.1080/15715124.2008.9635342.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1952: Order tests for two-sample problem and their power I. Indag. Math., 55, 453458, doi:10.1016/S1385-7258(52)50063-5.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1953a: Order tests for two-sample problem (second communication). Indag. Math., 56, 303310, doi:10.1016/S1385-7258(53)50039-3.

    • Search Google Scholar
    • Export Citation
  • Van der Waerden, B. L., 1953b: Order tests for two-sample problem (third communication). Indag. Math., 56, 311316, doi:10.1016/S1385-7258(53)50040-X.

    • Search Google Scholar
    • Export Citation
  • Wan, Z., and Z.-L. Li, 2008: Radiance-based validation of the V5 MODIS land‐surface temperature product. Int. J. Remote Sens., 29, 53735395, doi:10.1080/01431160802036565.

    • Search Google Scholar
    • Export Citation
  • Wan, Z., Y. Zhang, Q. Zhang, and Z.-L. Li, 2002: Validation of the land-surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data. Remote Sens. Environ., 83, 163180, doi:10.1016/S0034-4257(02)00093-7.

    • Search Google Scholar
    • Export Citation
  • Wang, A., M. Barlage, X. Zeng, and C. S. Draper, 2014: Comparison of land skin temperature from a land model, remote sensing, and in situ measurement. J. Geophys. Res. Atmos., 119, 30933106, doi:10.1002/2013JD021026.

    • Search Google Scholar
    • Export Citation
  • Wood, E. F., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) Phase 2(c) Red-Arkansas River Basin Experiment. 1. Experiment description and summary intercomparisons. Global. Planet. Change, 19, 115135, doi:10.1016/S0921-8181(98)00044-7.

    • Search Google Scholar
    • Export Citation
  • Yang, S., W. S. Olson, J.-J. Wang, T. L. Bell, E. A. Smith, and C. D. Kummerow, 2006: Precipitation and latent heating distributions from satellite passive microwave radiometry. Part II: Evaluation of estimates using independent data. J. Appl. Meteor. Climatol., 45, 721739, doi:10.1175/JAM2370.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, C., W. Li, and D. Travis, 2007: Gaps-fill of SLC-off Landsat ETM+ satellite image using a geostatistical approach. Int. J. Remote Sens., 28, 51035122, doi:10.1080/01431160701250416.

    • Search Google Scholar
    • Export Citation
  • Zheng, W., H. Wei, Z. Wang, X. Zeng, J. Meng, M. Ek, K. Mitchell, and J. Derber, 2012: Improvement of daytime land surface skin temperature over arid regions in the NCEP GFS model and its impact on satellite data assimilation. J. Geophys. Res., 117, D06117, doi:10.1029/2011JD015901.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    The locations of the BSRN sites with upward longwave radiation data used in this study.

  • Fig. 2.

    The daytime validation of CFSR LST at the 12 BSRN sites using the BSRN-derived LST.

  • Fig. 3.

    The HIRS LST compared with the CFSR LST at all 12 BSRN sites for each of the 11 satellites.

  • Fig. 4.

    The Taylor diagram for the HIRS LST (H) for each satellite and corresponding time series of CFSR LST (C) each with respect to the corresponding BSRN-derived LST time series (B) (measured in kelvin).

  • Fig. 5.

    Predominant vegetation types used for grid cells whose dominant vegetation equals or exceeds 60% of the gridcell area. The types are 1–12 (corresponding color to left of number) as follows: from left to right, evergreen needleleaf, evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, mixed cover, woodland, wooded grasslands, closed shrublands, open shrublands, grasslands, croplands, and bare soil [see Liang et al. (1994) or Hansen et al. (2000) for detailed description of each type].

  • Fig. 6.

    The LST bias (K) for (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

  • Fig. 7.

    The Pearson correlation coefficient for (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

  • Fig. 8.

    The LST RMSE (K) for (a) N12, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

  • Fig. 9.

    The number of data points for the duration of each of (a) N06, (b) N09, (c) N15, and (d) N16 for all grid cells having a dominant vegetation type equal to or exceeding 60%. White areas are grid cells without a dominant vegetation type.

  • Fig. 10.

    The average (a) April–September and (b) October–March LST biases (K) from all satellites. The white grid cells over land occur when one or more seasons for a given satellite have less than five data points, and therefore the bias is not calculated for that grid cell.

  • Fig. 11.

    The comparison of the final LST estimates at the 12 BSRN sites with the BSRN-derived LST.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 360 98 9
PDF Downloads 229 66 9