An improved physical split-window algorithm for precipitable water vapor retrieval exploiting the water vapor channel observations

Remote Sensing of Environment 194 (2017) 366–378

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

An improved physical split-window algorithm for precipitable water vapor retrieval exploiting the water vapor channel observations Hailei Liu a,b, Shihao Tang b,⁎, Juyang Hu b, Shenglan Zhang a, Xiaobo Deng a a b

Key Laboratory of Atmospheric Sounding, Chengdu University of Information Technology, Chengdu 610225, China National Satellite Meteorology Center, China Meteorological Administration, Beijing 100081, China

a r t i c l e

i n f o

Article history: Received 5 May 2016 Received in revised form 3 February 2017 Accepted 25 March 2017 Available online xxxx Keywords: PWV Geostationary satellites SVISSR MODIS Emissivity

a b s t r a c t This paper presents a new atmospheric precipitable water vapor (PWV) retrieval method based on three thermal infrared band observations from geostationary satellites. The proposed method is similar to the traditional physical split-window (PSW) retrieval technique, but a water vapor channel observation near 6.7 μm was included. Sensitivity analyses and simulation retrievals were carried out respectively according to the instrument characteristics of the Stretched Visible and Infrared Spin Scan Radiometer onboard FengYun-2G (SVISSR/FY-2G) and the Moderate Resolution Imaging Spectroradiometer aboard Terra (MODIS/Terra). The results indicate that the proposed 3-band algorithm can significantly reduce PWV retrieval errors caused by surface emissivity uncertainty and observation errors, especially in dry atmospheric conditions (i.e., PWV b 2 cm). The proposed algorithm was validated using SVISSR/FY-2G and MODIS/Terra observations, and was compared with radiosonde and GPS PWV. The determination coefficient (R2), root mean square error (RMSE), and bias between the SVISSR retrieved PWV and the radiosonde PWV are 0.87, 0.43 cm and 0.14 cm, respectively. The R2, RMSE and bias of the MODIS retrieved PWV are 0.89, 0.10 cm and −0.042 cm, respectively, which are slightly better than the MODIS L2 thermal infrared and near-infrared PWV products. © 2017 Elsevier Inc. All rights reserved.

1. Introduction Water vapor is the most abundant greenhouse gas in the Earth's atmosphere. Water vapor and its variations are the main driving forces of weather and climate change (Solomon et al., 2007; Zveryaev and Allan, 2005). It plays an important role in the study of climate change, hydrological cycle, energy budget and biogeochemistry at global and regional scales (Dessler et al., 2008; Raval and Ramanathan, 1989). Precipitable water vapor (PWV), which is the total atmospheric water vapor contained in a vertical column of unit cross-sectional area extending from the Earth's surface to the top of the atmosphere, is an important parameter for the climate analysis of energy budgets, hydrological cycles and numerical weather prediction (Nakamura et al., 2004; Smith et al., 2000; Trenberth et al., 2009). Moreover, PWV is one of the main geophysical parameters that affects surface remote sensing applications, such as land surface temperature (LST) retrieval and atmosphere correction of satellite data (Li et al., 2013; Qin et al., 2001; Sobrino et al., 1993; Vermote et al., 2002). A number of techniques have been used to obtain the PWV such as radiosonde, GPS, ground-based sun photometer and microwave radiometer, as well as polar-orbiting and geostationary satellite ⁎ Corresponding author. E-mail address: [email protected] (S. Tang).

http://dx.doi.org/10.1016/j.rse.2017.03.031 0034-4257/© 2017 Elsevier Inc. All rights reserved.

observations (Alshawaf et al., 2015; Czajkowski et al., 2002; Firsov et al., 2013; Li et al., 2003; Wang et al., 2015). Satellite observations, given its unique temporal and spatial resolution advantages, can effectively provide global or regional PWV distributions (Justice et al., 2002). Geostationary satellites can provide continuous observations of certain areas on the Earth's surface generally every 15 to 30 min (Maini and Agrawal, 2010; Suggs et al., 1998). One geostationary satellite can cover almost 1/3 of the Earth's surface, and a constellation of three equally spaced satellites can provide full coverage of the Earth (except the polar regions). Consequently, using geostationary satellite observations is an effective way to obtain high temporal and spatial resolution PWV at regional or global scales. At present, PWV retrieval algorithms of geostationary satellites have been mainly based on thermal infrared data given the lack of observations in the near infrared and microwave bands (Cziczo et al., 2013; Julien et al., 2015; Schroedter-Homscheidt et al., 2008; Suggs et al., 1998). In past decades, a number of algorithms have been proposed to derive PWV from thermal infrared observations. In general, these algorithms are mainly based on the water vapor differential absorption within the split-window channels, and can be classified into the linear split-window algorithm, split-window covariance-variance ratio method, physical split-window (PSW) algorithm, and look-up table approach (Dalu, 1986; Guillory et al., 1993; Labbi and Mokhnache, 2015; Ottle et al., 1997; Schroedter-Homscheidt et al., 2008; Sobrino et al., 2002;

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Sobrino and Romaguera, 2008). The accuracy of these algorithms are easily affected by the surface emissivity uncertainty, first-guess field error, instrument noise and calibration error (Barton and Prata, 1999; Knabb and Fuelberg, 1997), especially for dry atmospheric conditions (i.e., PWV b 2 cm) (Hulley et al., 2012; Sun et al., 2013). Based on the PSW technique (Guillory et al., 1993), the present study aimed to develop an improved PWV retrieval algorithm by adding a water vapor channel observation which has been included in most of the imagers onboard the geostationary satellites such as the Visible-Infrared Spin-Scan Radiometer onboard GOES (VISSR/GOES), Spinning Enhanced Visible and Infrared Imager onboard MSG (SEVIRI/MSG), and Stretched Visible and Infrared Spin Scan Radiometer onboard FY2G (SVISSR/FY-2G). The water vapor channel observations mainly respond to middle and upper tropospheric water vapor, which is useful in determining locations of moisture and atmospheric circulations (Laurent, 1993; Roca et al., 1997). Furthermore, the water vapor channel can also provide atmospheric temperature information if there is enough moisture in the atmosphere which is helpful to improve the accuracy of PWV retrievals (Seemann et al., 2003; Tang and Li, 2008). To our knowledge, few studies have been reported on the retrieval of PWV by combining a water vapor channel and split-window channel measurements. We expect our proposed algorithm to reduce the PWV retrieval uncertainty due to the surface emissivity uncertainty, firstguess field uncertainty, and observation errors, especially under dry atmospheric conditions. Section 2 describes the dataset used in this study. Section 3 first introduces the PSW retrieval technique, and then presents the sensitivity analysis and proposed 3-band PWV algorithm. The new algorithm is evaluated using the simulated SVISSR/FY-2G and MODIS/Terra radiances in Section 4. Validation of the proposed algorithm for the actual satellite observations is shown in Section 5. Finally, the conclusion is given in Section 6. 2. Datasets SVISSR/FY-2G and MODIS/Terra data were used to evaluate the proposed algorithm. The reference data, including GPS and radiosonde PWV data were used for validation. In addition, the ECMWF reanalysis data were used for the numerical simulation and first-guess field for the proposed algorithm. 2.1. SVISSR/FY-2G and MODIS/Terra data FengYun-2 (FY-2) is the first generation of Chinese geostationary meteorological satellite series, and currently includes seven satellites (Hu et al., 2013). One of the FY-2 satellites, FY-2G, was launched on December 31, 2014 with a main payload SVISSR. SVISSR/FY-2G observes the Earth every 30 min in 5 spectral channels, including two split-window channels, one water vapor channel, one mid-infrared channel, and one visible channel (Table 1). As compared with previous instruments from the FY-2 satellite series, the SVISSR/FY-2G has been improved from the following three aspects: reduced stray infrared radiation, uplifted observation frequency for the blackbody, and improved telemetry resolution of optical components. These improvements are conducive in improving the accuracy of FY-2G quantitative products. Combining with previous FY-2 satellites

Table 1 Instrument specifications for SVISSR/FY-2G. Band

Wavelength (μm)

SNR/NEΔT

IFGOV

Spatial resolution

VIS IR1 IR2 IR3 WV

0.55–0.75 10.3–11.3 11.5–12.5 3.50–4.00 6.30–7.60

N1.2 (ρ = 1%) 0.2–0.4 K@300 K 0.2–0.4 K@300 K 0.3–0.6 K@300 K 0.3–0.6 K@260 K

35 140 140 140 140

1.25 km 5 km 5 km 5 km 5 km

367

Fig. 1. Spectral response functions of SVISSR/FY-2G (in black) and MODIS/Terra (in gray) along with the calculated transmittance (blue line) of the standard mid-latitude winter atmosphere by MODTRAN 5.2. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

data, the FY-2G satellite observations can provide long-term climatologies of PWV information over the Eastern Hemisphere. The FY-2G products used in this study are mainly comprised of the SVISSR L1B radiance and L2 cloud mask data, which are provided by the Fengyun Satellite Data Center (http://satellite.nsmc.org.cn/PortalSite/Data/ Satellite.aspx). Similar to SVISSR/FY-2G, MODIS/Terra has split-window and water vapor bands, and with higher spatial resolution, lower instrument noise, and a narrower spectral response function (SRF) (Fig. 1) (Justice et al., 2002; Wan, 2008). MODIS has two water vapor bands centered at 6.7 μm (band 27) and 7.2 μm (band 28) (Table 2). Band 27 is almost centrally located in the water vapor absorption region and only monitors the radiation from higher level atmosphere. In contrary, band 28 is located closer to the wings of the absorption region, which allows the sensor to detect radiation from lower layers. Given its relatively weak absorption, MODIS band 28 observations were used in this study. The MODIS L1b radiance, L2 cloud mask, near-infrared and thermal infrared PWV data were used. 2.2. Radiosonde and GPS PWV data The radiosonde and GPS-derived PWV data were used as the reference data to evaluate the retrieved PWV. The radiosonde data used in this study was obtained by the L-band sounding system (1675 MHz) of the China Meteorological Administration (CMA), and was collected from the University of Wyoming website (http://weather.uwyo.edu/ upperair/sounding.html). The CMA sounding system is composed of a GTS1 digital electronic radiosonde, a secondary wind-finding radar and a ground-check set. It is widely used to measure the air pressure, temperature, relative humidity, and wind from the ground to about 30 km in radiosonde sites across China (CMA, 2010). The accuracy of the measured pressure, temperature, and relative humidity is 1–2 hPa, 0.2–0.3 °C and 4–5%, respectively (CMA, 2010). The radiosonde PWV was derived by integrating the specific humidity from the surface to the top of the sounding profile. SuomiNet hourly GPS PWV was used to validate the PWV retrievals across North America. SuomiNet is funded by the National Science

Table 2 Instrument specifications for MODIS/Terra. Band

Wavelength (μm)

NEΔT

IFGOV

Spatial resolution

27 28 31 32

6.53–6.89 7.17–7.47 10.78–11.28 11.7–12.27

0.25 K@240 K 0.25 K@250 K 0.05 K@300 K 0.05 K@300 K

55 55 55 55

1 km 1 km 1 km 1 km

368

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Foundation and the network is managed by the University Corporation for Atmospheric Research (UCAR) (Ware et al., 2000). SuomiNet includes N300 GPS stations, providing near-real time PWV data that is available every 30 min. The typical accuracy of the SuomiNet PWV is about 1–2 mm (Ware et al., 2000). The SuomiNet data were obtained via http://www.suominet.ucar.edu/data.html.

LST (δTs) from the first-guess field result in a perturbation of the at-sensor radiance which can be written as:

 ∂τλ;s ∂B T gλ ∂I↑ δT s τλ;s δr þ λ δr þ δI λ ¼ B T gλ ∂r ∂r ! ∂T s
 
 ∂τλ;s ∂I↑λ ∂B T gλ þ δT s τλ;s ¼ δr B T gλ þ ∂r ∂r ∂T s

ð4Þ

2.3. ERA-interim data ECMWF Re-analysis Interim (ERA-Interim) is a third generation global atmospheric reanalysis operated by the European Centre for Medium-Range Weather Forecasts (ECMWF) (Dee et al., 2011). ERA-Interim monthly and every 6 h (i.e., 00 h, 06 h, 12 h, and 18 h UTC) reanalysis data including temperature, pressure and humidity profiles at full resolution were used. The monthly average data were applied to the radiative simulations and sensitivity analyses, and the 6 h data were used as the first-guess field for the proposed algorithm. The ERA-Interim data were obtained via http://apps.ecmwf.int/datasets/data. 3. Methodology

Assuming

 ∂τλ;s ∂I↑λ ∂B T gλ þ ; Dλ ¼ τλ;s C λ ¼ B T gλ ∂r ∂r ∂T s

ð5Þ

Eq. (4) becomes linear (δIλ = δrCλ + δTsDλ) with unknown parameters δTs and δr. The solution for δTs and δr can be obtained by applying the linear equation to two or more channel measurements: 8 δI > > < λ1 δIλ2 > > : δIλn

¼ δrC λ1 þ δT s Dλ1 ¼ δrC λ2 þ δT s Dλ2 ⋯ ¼ δrC λn þ δT s Dλn

ð6Þ

3.1. Theory of PSW retrieval technique The PSW retrieval technique was originally developed to retrieve PWV and LST from aircraft data, and has been widely used for PWV retrieval from geostationary satellite observations (Guillory et al., 1993; Suggs et al., 1998). The PSW technique is essentially derived from a perturbation of the radiative transfer equation. It uses at least two splitwindow channel observations to simultaneously solve the perturbations of PWV and LST from an initial guess value (Guillory et al., 1993; Suggs et al., 1998). A simple derivation of this technique will be described as follows. In clear sky conditions, the top of atmosphere (TOA) radiance observed at the satellite can be calculated by (Liou, 2002; Wendisch and Yang, 2012)  ∂τðλ; pÞ 0
dp I ðλÞ ¼ ε ðλÞτ s ðλÞBðλ; T s Þ þ ∫ ps B λ; T p ∂p
 ∂τ ð λ; p Þ p þ ð1−ε ðλÞÞτ s ðλÞ∫ 0s B λ; T p dp ∂p

ð1Þ

where I(λ) is the spectral radiance at wavelength λ, ε is the surface emissivity, τ is the atmospheric transmissivity, τs is the atmospheric transmittance from the surface to the top of the atmosphere, Ts is the surface temperature, Tp is the atmospheric temperature at pressure p, and B(λ,Ts) is the Planck function. By setting the atmospheric upwelling radiance at wavelength λ as I↑λ, Eq. (1) can be rewritten as:
 Iλ ¼ τ λ;s Bλ T gλ þ I ↑λ

ð2Þ

where τλ, s is the atmospheric transmittance and Tgλ is the at-surface brightness temperature (BT) at wavelength λ. Assuming the true water vapor profile has the same vertical structure as the first-guess water vapor profile, then it can be obtained from the first-guess water vapor profile multiplying by a scaling factor γ: 0

γ¼

w ðpÞ PW ¼ PW wðpÞ

0

ð3Þ

where w(p) is the first-guess water vapor mixing ratio profile, and w′(p) is the true water vapor profile. PW' and PW are the total precipitable water vapor values for the w′(p) and w(p) profiles, respectively, which can be derived by integrating the water vapor profile from the surface to the top of the atmosphere. The perturbations of γ (δγ) and

where n is the total number of channels. δIλn is the radiance difference between the at-sensor and calculated radiances for the different channels (λ1, λ2, …, λn) at a first-guess field. The coefficients of Cλn and Dλn can be calculated by Eq. (5) using a radiative transfer model (i.e., MODTRAN, RTTOV or CRTM) with a priori information. The first-guess of γ is usually set to 1. Then, δr and δTs can be obtained by solving the Eq. (6), where the retrieved PWV is equal to PW×(1+ δγ), while the retrieved LST is equal to Ts + δTs. At present, most of the PSW algorithms are based on split-window channels, which are easily affected by observation errors (i.e., instrument random noise and calibration error), surface emissivity uncertainty, first-guess field uncertainty and cloud contamination (Knabb and Fuelberg, 1997; Suggs et al., 1998). The PWV retrieval error for PSW algorithms based on two split-window channels is typically b 20% (Guillory et al., 1993; Suggs et al., 1998). In this study, a water vapor channel is added to the traditional PSW algorithm. This is expected to increase the information content for PWV and decrease the sensitivity of the solutions towards the emissivity uncertainty and observation error. The addition of a water vapor channel deems the equations (Eq. (6)) overdetermined, and the optimal approximation solutions can be obtained using the least squares method. 3.2. Sensitivity analysis To analyze the sensitivity of split-window and water vapor channel observations to PWV, radiative transfer simulations were performed using MODTRAN 5.2 (Berk et al., 2011). The simulation model consists of various combinations of atmosphere, LST, and surface types. The radiances measured at the SVIIRS/FY-2G and MODIS/Terra channels were calculated for each combination. The surface temperature and the atmospheric profiles (geopotential height, air temperature, and humidity) were derived from the ERA-interim monthly mean products between January and December 2012. The study area was within the range of 15–55° N, 70–140° E (Fig. 2). With a 3.5° interval in latitude and a 7.5° interval in longitude, 132 profiles were finally selected. The selected profiles can represent a variety of atmospheric conditions, with the PWV ranges between 0.03 and 7.5 cm and elevation ranges between 0 and 5.0 km (Fig. 2). Six surface types such as desert, grass, open shrubs, grassland, ocean water, and fresh snow provided by MODTRAN 5.2 were assigned to each simulation (Berk et al., 2011). The SVISSR and MODIS band-average emissivities are illustrated in Table 3. The view zenith angle was set at 0°. Finally, a total number of 9504 (12 months × 132 profiles × 6 surface types) radiative transfer calculations were performed.

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

369

Fig. 2. The topographic map of the study area. The elevation data were taken from the Shuttle Radar Topography Mission (SRTM).

In addition, we added Gaussian-distributed random noise to the simulated BTs to get better simulations of real satellite data. The standard deviations of SVISSR/FY2G noise equivalent delta temperature (NEΔT) in the split-window and water vapor channels were set at 0.2 K and 0.3 K, respectively, according to the design requirement for SVISSR infrared channels (http://www.cma.gov.cn/en2014/news/ Features/201503/t20150318_277087.html). The standard deviations of the MODIS NEΔT value in the split-window and water vapor channels were set at 0.05 K and 0.2 K, respectively. Fig. 3a illustrates the BT difference (BT11μm − BT12μm) between the SVISSR IR1 and IR2 channels as a function of the PWV for the six surface types. In general, the BT difference mostly ranges from 0 to 3 K, and increases with PWV for a specific surface type. The BT difference due to emissivity difference in split-window channels can reach up to 2 K, which is comparable with that caused by water vapor differential absorption. Therefore, the surface type (or emissivity) has an obvious effect on the BT difference in split-window channels. It is difficult to determine the PWV from such a small BT difference in split-window channels especially for dry atmosphere, when surface emissivity is unknown. Furthermore, the BT difference is b 1 K when PWV is b1 cm, which may be comparable with the observation errors. Therefore, PWV retrieval using only split-window measurements was more difficult given the presence of observation errors (i.e., random noise and calibration error), especially for dry atmospheres. In contrast, the dynamic range of the BT difference (BT11μm − BT6.7μm) between SVISSR IR1 and the water vapor channels is evidently larger by about 2–50 K as compared to that of split-window channels

Table 3 The band-average emissivities for the six surface types used in the simulations in the splitwindow and water vapor channels of SVISSR/FY-2G and MODIS/Terra. Sensor

Band

Desert

Grass

Open shrubs

Grassland

Ocean water

Fresh snow

SVISSR

WV IR1 IR2

0.833 0.888 0.900

0.993 0.982 0.988

0.955 0.956 0.947

0.925 0.852 0.850

0.980 0.990 0.990

0.993 0.983 0.991

MODIS

Band 28 Band 31 Band 32

0.840 0.890 0.900

0.993 0.984 0.989

0.951 0.955 0.946

0.918 0.840 0.851

0.980 0.990 0.990

0.991 0.982 0.991

(Fig. 3b). The value of BT11μm − BT6.7μm is mainly determined by the water vapor differential absorption in the two channels, therefore is relatively less affected by the surface type and observation errors. The BT difference (BT11μm − BT12μm) of MODIS bands 31 and 32 mainly ranges from −0.2 to 2 K, which is relatively smaller than that of the SVISSR split-window channels (Fig. 4a). The BT difference (BT11μm − BT7.2μm) of MODIS bands 31 and 28 ranges from 4 to 60 K, which is relatively larger than the BT difference (BT11μm − BT6.7μm) of SVISSR (Fig. 4b). This is mainly caused by the water vapor absorption differences due to specification differences (e.g., central wavelength and bandwidth) between MODIS and SVISSR. In either the simulated SVISSR or MODIS BTs, the BT difference between the infrared channel (e.g., 11 μm) and the water vapor channel is significantly larger than between two split-window channels. To some degree, this indicates the helpfulness in improving the PWV retrieval accuracy and reducing the PWV retrieval uncertainty caused by the surface emissivity uncertainty and observation errors following the addition of a water vapor channel to the traditional PSW algorithm.

3.3. 3-band PSW algorithm According to the sensitivity analysis, the accuracy of PWV retrievals is expected to improve following the addition of the water vapor channel observations to the traditional PSW algorithms, especially in dry atmospheric conditions. For simplicity, the traditional PSW algorithm that exploits the two split-window channel measurements is named the “2-band” algorithm, while the proposed PSW algorithm that uses the measurements of two split-window and one water vapor channels is named the “3-band” algorithm in the following descriptions. Fig. 5 illustrates the PWV retrieving process of the 3-band algorithm including the forward calculation and inversion. Given the first-guess field and the viewing geometry for a specific sounding, the forward model MODTRAN 5.2 was used to generate the at-sensor radiance, as well as the atmospheric transmittance, path radiance and downward radiance. The 3-band algorithm then retrieved the PWV and LST simultaneously from the differences between the observed and simulated radiances in two split-window and one water vapor channels by solving the linear perturbation equations (Eq.(6)).

370

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Fig. 3. The BT difference in SVISSR channels as a function of PWV for different surface types: (a) IR1 and IR2 channels; (b) IR1 and WV channels.

4. PWV retrievals from simulated data Retrieval experiments were performed using the simulated SVIRRS and MODIS radiances to evaluate the performance of the 2- and 3band algorithms. The initial water vapor profile was set at 80% of the true profile. Given accurate information on pixel emissivity is still difficult to obtain (Coll et al., 2007; Goïta and Royer, 1997; Valor and Caselles, 1996; Watson, 1992), it is necessary to evaluate the influence of the emissivity uncertainty on PWV retrieval, therefore the retrieval experiments were divided into two groups: (1) surface emissivity was known; (2) surface emissivity was unknown. In case 1, the initial emissivity was set to the true value. In case 2, the emissivity in each band was set to 1. Three statistical indicators, including the determination coefficient (R2), root mean square error (RMSE), and bias were used to evaluate the retrieved PWV from the simulated SVISSR and MODIS measurements with the true PWV which was directly integrated from the water vapor profiles.

4.1. Retrievals with known emissivity The retrieval experiments were performed using the simulated SVIRRS and MODIS radiances with known surface emissivities. The

results are shown in Fig. 6, which illustrates a comparison of the retrieved and true PWV values based on the 2- and 3-band algorithms. In general, the PWV retrievals of the 3-band algorithm are significantly better than that of the 2-band algorithm, especially for the dry atmospheres (i.e., PWV b 2 cm), no matter from the simulated SVISSR or MODIS measurements. The R2, RMSE and bias of the retrieved PWV from simulated SVISSR BTs are 0.998, 0.07 cm and − 0.07 cm according to the 3-band algorithm, respectively, and are 0.909, 0.68 cm and − 0.12 cm based on the 2-band algorithm (Fig. 6 upper panel). The PWV retrievals from simulated MODIS radiances are shown in the bottom panel of Fig. 6. The R2, RMSE and bias of the retrieved PWV from the MODIS simulated measurements are 0.999, 0.05 cm and − 0.07 cm using the 3-band algorithm, respectively, and are 0.983, 0.32 cm and − 0.14 cm for the 2-band algorithm. These results agree well with the sensitive analysis in Section 3.2. As analyzed in Section 3.2, the BT difference in two split-window channels caused by water vapor differential absorption is relatively small, especially for the dry atmospheres, which may be comparable to the observation error, thereby making it difficult to obtain accurate PWV using the 2-band algorithm even if the surface emissivity is known. However, by adding a water vapor channel, the dynamic range of BT differences between the water vapor channel and the split-window channels

Fig. 4. The BT difference in MODIS channels as a function of PWV for different surface types: (a) bands 31 and 32; (b) bands 31 and 28.

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

371

Fig. 5. Flowchart for the 3-band PWV retrieval algorithm.

(e.g., BT11μm-BT6.7μm) becomes evidently larger, which is helpful in improving the PWV retrieval accuracy and reducing the retrieval uncertainty caused by observation errors. It can also be seen that the accuracy of MODIS PWV retrievals are both slightly better than the SVISSR results using two algorithms. This is mainly due to a better instrument performance of MODIS such as NEΔT and calibration accuracy. In addition, retrievals from simulated

MODIS split-window channels and channel 27 (centered at 6.7 μm) measurements were also carried out. There is no obvious improvement compared to the results of 2-band algorithm. The proper explanation is that water vapor absorption in band 27 is too strong to provide total column water vapor information. Therefore, the channel located at the wings of the water vapor absorption band is more effective to improve the accuracy of PWV retrievals.

Fig. 6. PWV retrievals based on the 2-band (left column) and 3-band algorithm (right column) from the simulated SVISSR (upper panel) and MODIS (bottom panel) radiances assuming the emissivity was known. The blue line has a 1:1 ratio and the red line represents the initial value with respect to the true PWV. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

372

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Fig. 7. PWV retrievals based on the 2- (left column) and 3-band algorithm (right column) from simulated SVISSR (upper panel) and MODIS (bottom panel) radiances assuming emissivity was unknown. The blue line has a 1:1 ratio, and the red line represents the initial value with respect to the true PWV. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

4.2. Retrievals with unknown emissivity Similar to Section 4.1, the retrieval experiments in this section were performed using the simulated SVIRRS and MODIS radiances but with unknown surface emissivities. The surface emissivity in all channels was set to 1. The results are shown in Fig. 7, which illustrates a comparison of the retrieved and true PWV values based on the 2- and 3-band algorithms. In this case, the PWV inversion accuracy of the 2-band algorithm decreases significantly, especially for dry atmospheric conditions. For example, the PWV inversion error can reach up to 8 cm when the PWV is b1 cm. As mentioned in Section 3.2, the BT difference in split-window channels due to water vapor absorption is usually small, especially for a dry atmosphere, and it may be comparable to that caused by the emissivity difference in split-window channels and the observation errors. Therefore, it is difficult to obtain reliable PWV information from the BT difference in split-window channels when emissivity is unknown, especially for a dry atmosphere. In contrast, the accuracy of the 3-band algorithm does not show obvious decline in the case when emissivity was unknown. The R2, RMSE and bias of the retrieved PWV are 0.998, 0.07 cm and −0.05 cm respectively from simulated SVISSR data, and are 0.999, 0.05 cm and −0.07 cm respectively from simulated MODIS data, indicating the effectiveness and accuracy of the 3-band algorithm in retrieving the PWV. The 3band algorithm can effectively reduce the PWV retrieval uncertainty caused by the surface emissivity uncertainty and observation errors, which agrees well with sensitive analysis in Section 3.2.

4.3. Effects of water vapor vertical distribution on PWV retrievals Vertical distribution of water vapor affects the atmospheric upwelling and downwelling radiances, as well as the satellite observed BTs. In the PSW algorithms, the priori water vapor profiles are generally assumed

to have the same vertical distribution structure as the true water vapor profile. However, the priori information of water vapor profile taken from the numerical prediction model or the radiosonde measurements typically exhibits discrepancies with the true profile. Furthermore, water vapor profiles with different vertical distributions can have identical PWV values. Therefore, it is necessary to investigate the PWV retrieval errors produced by the difference between the assumed vertical distribution of water vapor and the true profile. For this purpose, the radiative transfer calculations and retrieval experiments were performed for different model atmospheres at different water vapor profiles. The PWV retrieval experiments were performed using MODTRAN 5.2 for the six standard model atmospheres, including tropical (TRO), mid-latitude summer (MLS), mid-latitude winter (MLW), sub-arctic summer (SAS), sub-arctic winter (SAW), and U.S. standard (USS) atmosphere. In the six model atmospheres, the PWV ranges from 0.42 to 4.20 cm, and the vertical distributions of water vapor are different. To investigate PWV retrieval dependence on the water vapor vertical distribution, 20 water vapor profiles were generated for each model atmosphere with a fixed PWV using the Gaussian-distributed random scaling factors at each layer. Twenty newly generated water vapor profile sets for MLS and MLW are shown in Fig. 8. Consequently, the SVISSR/FY2G BTs were simulated for the six model atmospheres (20 profiles for each) under the six surface types (Table 3).The satellite viewing zenith angle was set to 0, and the LST was set to the respective surface air temperature of each model atmosphere in the simulations. To comprehensively evaluate the PWV retrieval errors caused by the vertical distribution of water vapor as well as observation and surface emissivity errors, the retrieval experiments were performed under two different situations, namely test 1 and test 2. Test 1 was performed assuming that the surface emissivity was known and no noise was added to the simulated BTs, by which the influence of the water vapor vertical distribution on the PWV retrievals can be investigated. Test 2 was performed assuming an unknown surface emissivity, with measurement

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

373

Fig. 8. The water vapor mixing ratio profile sets (color dotted lines) produced from the MODTRAN model atmosphere by adding random errors at each layer: (a) mid-latitude summer atmosphere; (b) mid-latitude winter atmosphere. The black solid line represents the MODTRAN model atmospheric profile for water vapor.

errors added to the simulated BTs. Therefore, the comprehensive influence of three factors on the PWV retrievals, namely the observation error, surface emissivity uncertainty, and water vapor vertical distribution uncertainty can be analyzed. As a result, 1440 retrievals (6 model atmosphere × 20 water vapor profile sets × 6 surface types × 2 situations) from the simulated SVISSR/FY-2G BTs were performed (Fig. 9). In test 1, the PWV errors using the 2- and 3-band algorithms are both within 0.25 cm for most cases and increase with PWV values (Fig. 9a

and b). The PWV errors for the 2-band algorithm are b 0.15 cm for dry atmospheres (MLW, SAS, SAW and USS), and are within 0.25 cm for moist atmospheres (TRO and MLS), indicating that the retrieval error caused by the water vapor vertical distribution uncertainties increases with the total column water vapor. While the PWV errors of the 3band algorithm are slightly higher than that of 2-band algorithm for the TRO atmosphere and are comparable to that of the 2-band algorithm for other model atmospheres, indicating the 3-band algorithm

Fig. 9. PWV retrieval errors of 2- (left panel) and 3-band (right panel) algorithms performed on SVISSR simulations for different model atmospheres and surface types. (a) and (b) present the retrievals in test 1; (c) and (d) present the retrievals in test 2.

374

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Table 4 Bias and RMSE of the retrieved PWV using 2- and 3-band algorithms from BTs simulated using MODITRAN for the different background atmospheres in test 2. For each model atmosphere, the bias and RMSE were computed from 120 retrievals (20 priori profiles × 6 surface types). Model atmosphere

Tropical Mid-latitude summer Mid-latitude winter Sub-arctic summer Sub-arctic winter U.S. standard

PWV (cm)

4.20 2.98 0.87 2.12 0.42 1.44

2-band method

3-band method

Bias (cm)

Relative bias (%)

RMSE (cm)

Bias (cm)

Relative bias (%)

RMSE (cm)

0.11 0.46 0.84 0.37 1.67 0.45

2.57 15.30 96.89 17.42 398.31 31.06

0.35 0.73 1.26 0.67 2.93 0.75

0.08 0.12 0.03 0.07 0.02 0.04

2.02 4.16 2.99 3.48 4.73 3.00

0.22 0.17 0.04 0.12 0.03 0.08

is more sensitive to the water vapor profile especially at high PWV values (e.g., TRO). In test 2, the emissivity in all channels was set to 1, and the PWV retrieval results are shown in Fig. 9(c) and (d). In general, the PWV errors of the 2- and 3-band algorithms for test 2 are both larger than that of test 1. The PWV errors of the 2-band algorithm become evidently larger in test 2, especially for dry atmospheres such as MLW and SAW (Fig. 9c), while the PWV errors of 3-band algorithm show slight increase. The results indicate the 2-band algorithm is more easily affected by the surface types than the 3-band algorithm, which is consistent with the results presented in Section 4.2. Furthermore, the PWV retrieval errors of the 2-band algorithm increase with the decrease of PWV, for example, the RMSE of PWV can reach up to 2.9 cm for SAW (Table 4). In contrast, the PWV errors of the 3-band algorithm increases with PWV. The RMSE of the retrieved PWV using the 3-band algorithm is within 0.22 cm for all the six model atmospheres and is b 0.1 cm for the dry atmospheres (MLW, SAW and USS) (Table 4). The results illustrate lower sensitivity to emissivity uncertainty and measurement errors in the 3-band algorithm than in the 2band algorithm, especially for dry atmospheres. According to the above results, the performance of the 3-band algorithm is better than that of the 2-band algorithm in most cases. The vertical distribution of the water vapor is one of the important factors that affect the performance of the 3-band algorithm. However, the PWV retrieval error caused by the water vapor profile with an uncertainty of ±15% is within 0.2 cm in most cases, which is usually acceptable. 5. Validation and results The SVISSR/FY-2G and MODIS/Terra observations were used to further evaluate the performance of the 3-band PWV retrieval algorithm.

For the SVISSR/FY-2G data, the study area was mainly over China and the radiosonde PWV was used as the reference data. For the MODIS/ Terra data, an area of higher elevation (N 1.0 km) over America was selected, and the SuomiNet GPS hourly PWV data were used as the reference data. For PWV retrieval, the surface emissivity was set to 1 for all the selected SVISSR and MODIS channels. The first-guess atmospheric profiles were derived from the ERA-interim reanalysis data. The ERA-Interim provided LST information at 0.75° × 0.75° grid every 6 h. Considering the huge variation in the LST spatial and temporal distribution, the SVISSR and MODIS BT at 11 μm was used as the priori LST, which is similar to the approach presented in the ASTER Temperature–Emissivity Separation (TES) algorithm (Gillespie et al., 1998). The SVISSR and MODIS L2 cloud mask products were directly used to select clear sky conditions.

5.1. Retrievals from SVISSR data Using the proposed 3-band algorithm, the PWV was retrieved from the SVISSR/FY-2G L1B radiance, and compared to the radiosonde PWV at 83 stations across China. Fig. 10 shows the location and elevation information of the 83 stations. The station elevations range from 0 to 5 km, and the LST and PWV cover a wide representative range. Only the FY-2G data that fell into the 0.05° × 0.05° boxes near the radiosonde sites were used to validate the algorithm. Given that sounding data are acquired two times a day (0000 UTC and 1200 UTC), the FY-2G data at 0000 UTC from June 4 to November 30, 2015 were used. Furthermore, to evaluate the influence of the first guess water vapor profile on the PWV retrievals from the real satellite observations, the retrievals were

Fig. 10. Geographical location and elevation of the radiosonde stations used in this study over China.

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

performed with the first guess water vapor profile setting to 80% and 100% of the ERA-Interim water vapor profile, respectively. Fig. 11 illustrates comparisons of the SVISSR/FY-2G PWV retrievals with the radiosonde PWV at different first-guess water vapor profiles under clear-sky conditions. The R2, RMSE, and bias of the retrieved PWV are 0.87, 0.43 cm and 0.14 cm respectively when the first guess water vapor profile was set as the ERA-Interim profile (Fig. 11a), and are 0.87, 0.48 cm, and − 0.18 cm respectively when the first guess water vapor profile was set at 80% of the ERA-interim profile (Fig. 11c). The inversion accuracy is not significantly reduced when the first guess water vapor profile is set to 80% of the ERA-Interim profile, which is consistent with the results of the retrieval experiments using the simulated data presented in Sections 4.1 and 4.2. It indicates that the 3-band algorithm can effectively correct the systematic deviation of the first guess water vapor profiles. The high determination coefficient (R2) and low RMSE indicate the effectiveness and accuracy of the 3-band algorithm in obtaining the PWV from the SVISSR observations. Furthermore, the histogram of the PWV retrieval error (PWVretrieved − PWVradiosonde) is presented in Fig. 11(b) and (d). In general, the PWV retrieval error lies within a range of ± 1 cm, and N78% of the retrieval error is within a range of ± 0.5 cm. The results indicate that a robust 3-band algorithm can accurately provide PWV information even with an unknown emissivity. It is worth mentioning that the FY-2G L2 cloud mask products are mainly based on the threshold method from the visible, thermal infrared and water vapor channel observations (e.g., VIS, IR1, IR2 and WV channels). Unlike MODIS, SVISSR/FY-2G does not have a channel

375

centered at 1.375 μm that has been widely used for cirrus cloud detection (Gao and Kaufman, 1995). Therefore, it is difficult for SVISSR/FY2G to detect thin cirrus clouds. The presence of thin cirrus clouds may decrease the accuracy of PWV retrievals.

5.2. Retrievals from MODIS data MODIS data over the North America (35°–45° N, 105°–115° W) between January 1 and June 30, 2015 were selected to validate the 3band algorithm. MODIS/Terra cloud cover products (MOD35), radiance products (MOD021KM), near infrared (NIR), thermal infrared (TIR) water vapor products (MOD05_L2 and MOD07_L2), and geolocation products (MOD03) were used for PWV retrieval. SuomiNet GPS-derived hourly PWV data were used as the reference data. Fig. 12 shows the distribution map of SuomiNet GPS sites across the United States, with the color indicating the elevation information of each site. The elevation of the study area (rectangular box in Fig. 12) is relatively high (N 1.0 km), while the PWV is relatively low (b1.25 cm). In addition, only the MODIS observations that fell into the 0.05° × 0.05° boxes near the GPS sites were collected to validate the algorithm. Fig. 13 shows the comparisons of the retrieved PWV from MODIS data and the GPS PWV. The R2, RMSE, and bias of the retrieved PWV are 0.89, 0.10 cm and −0.042 cm, respectively (Fig. 13a), which indicates the applicability of the 3-band algorithm in obtaining high accuracy PWV data in dry atmospheric conditions. The distribution of the error between the MODIS retrievals and the GPS PWV is shown in Fig. 13b.

Fig. 11. (a) Scatter plot of SVISSR retrieved PWV and radiosonde PWV across China between June and November 2015; (b) Histogram of the difference between the SVISSR retrieved and radiosonde PWVs; (c) and (d) are same as (a) and (b) but with a first-guess water vapor profile set at 80% of the ERA-interim profile.

376

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Fig. 12. The geolocation and elevation of the SuomiNet GPS stations. The colors indicate the elevation information of each site. The study area falls across the rectangular box (35°–45° N, 105°–115° W).

Fig. 13. (a) Scatter plot of the MODIS retrieved PWV and GPS PWV; (b) histogram of the difference between the MODIS retrieved and radiosonde PWVs.

The error lies mainly within the range of ±0.2 cm, and N90% of the retrieval errors lie within a range of ±0.15 cm. To compare the performance of the proposed 3-band algorithm with the MODIS PWV products, MODIS L2 thermal infrared (TIR) and near-

infrared (NIR) PWV products were also evaluated within the study region between January and June 2015 using the SuomiNet GPS-derived hourly PWV data (Fig. 14). The R2, RMSE and bias between the MODIS TIR PWV and the GPS PWV are 0.79, 0.14 cm and − 0.066 cm,

Fig. 14. (a) Scatter plot of MODIS L2 TIR PWV and GPS PWV; (b) scatter plot of MODIS L2 NIR PWV and GPS PWV.

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

respectively, whereas the corresponding values between MODIS NIR PWV and GPS PWV are 0.83, 0.22 cm and 0.13 cm. The MODIS NIR PWV shows a higher correlation with the GPS data, but with a positive deviation of 0.13 cm. This appears to indicate an overestimation in MODIS NIR PWV, which is consistent with the analysis of the previous studies (Liu et al., 2015). Overall, in comparing the MODIS NIR and TIR PWV products, the retrieved PWV using the 3-band algorithm is in better agreement with the GPS PWV data given its higher determination coefficient, lower RMSE and bias. 6. Conclusions At present, the geostationary satellite is the only feasible way to obtain high temporal and spatial resolution water vapor information at large scales. However, imagers onboard geostationary satellites in orbit are not able to provide observations in the near infrared or microwave region, which are typically used for PWV retrieval. The PWV inversion methods of these imagers are mainly based on split-window channel observations. The accuracy of these algorithms is easily affected by the surface emissivity uncertainty and observation errors. The present study proposed a 3-band algorithm based on the traditional PSW retrieval technique, by adding a water vapor channel. Radiative transfer simulations and numerical experiments were performed for SVISSR/FY-2G and MODIS/Terra to analyze the sensitivity of splitwindow and water vapor channel observations to the PWV. The results show that the 3-band algorithm can effectively improve the PWV retrieval accuracy and reduce the PWV retrieval uncertainty caused by the surface emissivity uncertainty and observation errors. Furthermore, the algorithm was applied to SVISSR/FY-2G and MODIS/Terra observations, and the accuracy of the PWV retrievals was evaluated using the radiosonde and GPS derived PWV, respectively. The results demonstrate the applicability of the 3-band algorithm in obtaining highly accurate PWV values, which can be applied to dry atmospheric conditions, even in the absence of known surface emissivity values. This research mainly focused on the development of the retrieval algorithm, where the MODTRAN was used as the forward radiative transfer model. For the operational use of the algorithm, a fast radiative transfer model such as RTTOV or CRTM is recommended. The influence of high aerosol loadings (such as dust) as well as thin cirrus clouds on PWV retrieval requires further investigation. Theoretically, the algorithm can be applied to the other satellite imagers such as SEVIRI/MSG and VISSR/GOES, which have one or more water vapor channels and split-window channels. Acknowledgements This work was supported by the China Special Fund for Meteorological Research in the Public Interest (No. GYHY201406001) and the National Natural Science Foundation of China (NSFC) (Nos. 41475032, 41375042 and 41527806). The authors thank the NASA Earth Observing System Data and Information System (EOSDIS) for sharing the MODIS data via the Simple Subset Wizard, the University of Wyoming for providing radiosonde data, the ECMWF for distributing the ERA-Interim data, and the National Satellite Meteorological Center (NSMC) of China Meteorological Administration for the FY/2G data. We thank the editor and anonymous reviewers for providing constructive comments on the manuscript. References Alshawaf, F., Fuhrmann, T., Knopfler, A., Luo, X., 2015. Accurate estimation of atmospheric water vapor using GNSS observations and surface meteorological data. IEEE Trans. Geosci. Remote Sens. 53, 3764–3771. Barton, I.J., Prata, A.J., 1999. Difficulties associated with the application of covariance–variance techniques to retrieval of atmospheric water vapor from satellite imagery. Remote Sens. Environ. 69, 76–83.

377

Berk, A., Anderson, G., Acharya, P., Shettle, E., 2011. MODTRAN 5.2. 1 User's Manual. Spectral Sciences, INC., Burlington, MA, p. 69. CMA (China Meteorological Administration), 2010. Specification of Routine Upper-Air Observation(in Chinese). China Meteorological Press. Coll, C., Caselles, V., Valor, E., Niclòs, R., Sánchez, J.M., Galve, J.M., Mira, M., 2007. Temperature and emissivity separation from ASTER data for low spectral contrast surfaces. Remote Sens. Environ. 110, 162–175. Czajkowski, K.P., Goward, S.N., Shirey, D., Walz, A., 2002. Thermal remote sensing of nearsurface water vapor. Remote Sens. Environ. 79, 253–265. Cziczo, D.J., Froyd, K.D., Hoose, C., Jensen, E.J., Diao, M., Zondlo, M.A., Smith, J.B., Twohy, C.H., Murphy, D.M., 2013. Clarifying the dominant sources and mechanisms of cirrus cloud formation. Science 340, 1320–1324. Dalu, G., 1986. Satellite remote sensing of atmospheric water vapour. Int. J. Remote Sens. 7, 1089–1097. Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U., Balmaseda, M.A., Balsamo, G., Bauer, P., 2011. The ERA-interim reanalysis: configuration and performance of the data assimilation system. Q. J. R. Meteorol. Soc. 137, 553–597. Dessler, A.E., Zhang, Z., Yang, P., 2008. Water-vapor climate feedback inferred from climate fluctuations, 2003–2008. Geophys. Res. Lett. 35, 293–310. Firsov, K.M., Chesnokova, T.Y., Bobrov, E.V., Klitochenko, I.I., 2013. Total water vapor content retrieval from sun photometer data. Atmos. Oceanic Opt. 26, 281–284. Gao, B.-C., Kaufman, Y.J., 1995. Selection of the 1.375-um MODIS channel for remote sensing of cirrus clouds and stratospheric aerosols from space. J. Atmos. Sci. 52, 4231–4237. Gillespie, A., Rokugawa, S., Matsunaga, T., Cothern, J.S., Hook, S., Kahle, A.B., 1998. A temperature and emissivity separation algorithm for advanced spaceborne thermal emission and reflection radiometer (ASTER) images. IEEE Trans. Geosci. Remote Sens. 36 (4), 1113–1126. Goïta, K., Royer, A., 1997. Surface temperature and emissivity separability over land surface from combined TIR and SWIR AVHRR data. IEEE Trans. Geosci. Remote Sens. 35, 718–733. Guillory, A.R., Jedlovec, G.J., Fuelberg, H.E., 1993. A technique for deriving column-integrated water content using VAS split-window data. J. Appl. Meteorol. 32, 1226–1241. Hu, X., Xu, N., Weng, F., Zhang, Y., Chen, L., Zhang, P., 2013. Long-term monitoring and correction of FY-2 infrared channel calibration using AIRS and IASI. IEEE Trans. Geosci. Remote Sens. 51, 5008–5018. Hulley, G.C., Hughes, C.G., Hook, S.J., 2012. Quantifying uncertainties in land surface temperature and emissivity retrievals from ASTER and MODIS thermal infrared data. J. Geophys. Res.-Atmos. 117, 1702–1711. Julien, Y., Sobrino, J.A., Mattar, C., Jimenez-Munoz, J.C., 2015. Near-real-time estimation of water vapor column from MSG-SEVIRI thermal infrared bands: implications for land surface temperature retrieval. IEEE Trans. Geosci. Remote Sens. 53, 1–7. Justice, C.O., Townshend, J.R.G., Vermote, E.F., Masuoka, E., Wolfe, R.E., Saleous, N., Roy, D.P., Morisette, J.T., 2002. An overview of MODIS land data processing and product status. Remote Sens. Environ. 83, 3–15. Knabb, R.D., Fuelberg, H.E., 1997. A comparison of the first-guess dependence of precipitable water estimates from three techniques using GOES data. J. Appl. Meteorol. 36, 417–427. Labbi, A., Mokhnache, A., 2015. Estimating of total atmospheric water vapor content from MSG1-SEVIRI observations. Atmos. Meas. Tech. Discuss. 8, 8903–8923. Laurent, H., 1993. Wind extraction from Meteosat water vapor channel image data. J. Appl. Meteorol. 32, 1124–1133. Li, Z., Muller, J.P., Cross, P., 2003. Comparison of precipitable water vapor derived from radiosonde, GPS, and moderate-resolution imaging spectroradiometer measurements. J. Geophys. Res. Atmos. 108, 87–107. Li, Z.L., Tang, B.H., Wu, H., Ren, H., Yan, G., Wan, Z., Trigo, I.F., Sobrino, J.A., 2013. Satellitederived land surface temperature: current status and perspectives. Remote Sens. Environ. 131, 14–37. Liou, K.-N., 2002. An Introduction to Atmospheric Radiation. second ed. Boston: Academic Press, Amsterdam. Liu, H., Tang, S., Zhang, S., Hu, J., 2015. Evaluation of MODIS water vapour products over China using radiosonde data. Int. J. Remote Sens. 36, 680–690. Maini, A.K., Agrawal, V., 2010. Satellite Technology: Principles and Applications. Wiley. Nakamura, H., Koizumi, K., Mannoji, N., 2004. Data assimilation of GPS precipitable water vapor to the JMA mesoscale numerical weather prediction model and its impact on rainfall forecasts. J. Meteorol. Soc. Jpn. 82, 441–452. Ottle, C., Outalha, S., Francois, C., Maguer, S.L., 1997. Estimation of total atmospheric water vapor content from split-window radiance measurements. Remote Sens. Environ. 61, 410–418. Qin, Z., Karnieli, A., Berliner, P., 2001. A mono-window algorithm for retrieving land surface temperature from Landsat TM data and its application to the Israel-Egypt border region. Int. J. Remote Sens. 22, 3719–3746. Raval, A., Ramanathan, V., 1989. Observational determination of the greenhouse effect. Nature 65, 1–25. Roca, R., Picon, L., Desbois, M., Treut, H.L., Morcrette, J.-J., 1997. Direct comparison of METEOSAT water vapor channel and general circulation model results. Geophys. Res. Lett. 24, 147–150. Schroedter-Homscheidt, M., Drews, A., Heise, S., 2008. Total water vapor column retrieval from MSG-SEVIRI split window measurements exploiting the daily cycle of land surface temperatures. Remote Sens. Environ. 112, 249–258. Seemann, S.W., Li, J., Strabala, K.I., Menzel, W.P., 2003. Operational retrieval of atmospheric temperature, moisture, and ozone from MODIS infrared radiances. J. Appl. Meteorol. 42, 1072–1091. Smith, T.L., Benjamin, S.G., Schwartz, B.E., Gutman, S.I., 2000. Using GPS-IPW in a 4-D data assimilation system. Earth Planets Space 52, 921–926.

378

H. Liu et al. / Remote Sensing of Environment 194 (2017) 366–378

Sobrino, J.A., Romaguera, M., 2008. Water-vapour retrieval from Meteosat 8/SEVIRI observations. Int. J. Remote Sens. 29, 741–754. Sobrino, J.A., Li, Z.L., Stoll, M.P., 1993. Impact of the atmospheric transmittance and total water vapor content in the algorithms for estimating satellite sea surface temperatures. IEEE Trans. Geosci. Remote Sens. 31, 946–952. Sobrino, J.A., Jimenez, J.C., Raissouni, N., Soria, G., 2002. A simplified method for estimating the total water vapor content over sea surfaces using NOAA-AVHRR channels 4 and 5. IEEE Trans. Geosci. Remote Sens. 40, 357–361. Solomon, S., Intergovernmental Panel on Climate Change, Intergovernmental Panel on Climate Change, Working Group I, 2007. Climate Change 2007: The Physical Science Basis: Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge; New York. Suggs, R.J., Jedlovec, G.J., Guillory, A.R., 1998. Retrieval of geophysical parameters from GOES: evaluation of a split-window technique. J. Appl. Meteorol. 37, 1205–1227. Sun, D., Yu, Y., Fang, L., Liu, Y., 2013. Toward an operational land surface temperature algorithm for GOES. J. Appl. Meteorol. Climatol. 52, 1974–1986. Tang, B., Li, Z.L., 2008. Estimation of instantaneous net surface longwave radiation from MODIS cloud-free data. Remote Sens. Environ. 112, 3482–3492. Trenberth, K.E., Fasullo, J.T., Kiehl, J., 2009. Earth's global energy budget. Bull. Am. Meteorol. Soc. 90.

Valor, E., Caselles, V., 1996. Mapping land surface emissivity from NDVI: application to European, African, and South American areas. Remote Sens. Environ. 57, 167–184. Vermote, E.F., Saleous, N.Z.E., Justice, C.O., 2002. Atmospheric correction of MODIS data in the visible to middle infrared: first results. Remote Sens. Environ. 83, 97–111. Wan, Z., 2008. New refinements and validation of the MODIS land-surface temperature/ emissivity products. Remote Sens. Environ. 112, 59–74. Wang, Y.Q., Shi, J.C., Wang, H., Feng, W.L., Wang, Y.J., 2015. Physical statistical algorithm for precipitable water vapor inversion on land surface based on multi-source remotely sensed data. Sci. China Earth Sci. 1–13. Ware, R.H., Fulker, D.W., Stein, S.A., Anderson, D.N., Avery, S.K., Clark, R.D., Droegemeier, K.K., Kuettner, J.P., Minster, J.B., Sorooshian, S., 2000. SuomiNet: a real-time national GPS network for atmospheric research and education. Bull. Am. Meteorol. Soc. 81, 677–694. Watson, K., 1992. Spectral ratio method for measuring emissivity. Remote Sens. Environ. 42, 113–116. Wendisch, M., Yang, P., 2012. Theory of Atmospheric Radiative Transfer: A Comprehensive Introduction. Wiley-VCH, Weinheim. Zveryaev, I.I., Allan, R.P., 2005. Water vapor variability in the tropics and its links to dynamics and precipitation. J. Geophys. Res. Atmos. 110, 3033–3043.