We present a new total column water vapor (TCWV) retrieval algorithm in the visible blue spectral band for the Global Ozone Monitoring Experience 2 (GOME-2) instruments on board the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Metop satellites. The blue band algorithm allows the retrieval of water vapor from sensors which do not cover longer wavelengths, such as the Ozone Monitoring Instrument (OMI) and the Copernicus atmospheric composition missions Sentinel-5 Precursor (S5P), Sentinel-4 (S4) and Sentinel-5 (S5). The blue band algorithm uses the differential optical absorption spectroscopic (DOAS) technique to retrieve water vapor slant columns. The measured water vapor slant columns are converted to vertical columns using air mass factors (AMFs). The new algorithm has an iterative optimization module to dynamically find the optimal a priori water vapor profile. This makes it better suited for climate studies than usual satellite retrievals with static a priori or vertical profile information from the chemistry transport model (CTM). The dynamic a priori algorithm makes use of the fact that the vertical distribution of water vapor is strongly correlated to the total column. The new algorithm is applied to GOME-2A and GOME-2B observations to retrieve TCWV. The data set is validated by comparing it to the operational product retrieved in the red spectral band, sun photometer and radiosonde measurements. Water vapor columns retrieved in the blue band are in good agreement with the other data sets, indicating that the new algorithm derives precise results and can be used for the current and forthcoming Copernicus Sentinel missions S4 and S5.
Atmospheric water vapor is the most important natural greenhouse gas in the troposphere, accounting for more than 60 % of the greenhouse effect
Satellite remote sensing observations are an effective way of monitoring the spatiotemporal variations of column amount water vapor on a global scale. High-quality water vapor data can be derived from a large number of satellite sensors operating in various wavelength regions
In this work, we focus on the development of a water vapor retrieval algorithm for spectroscopic satellite observations in the ultraviolet (UV) and visible (Vis) spectral range with nadir viewing geometry. This kind of observation has long been conducted since the Global Ozone Monitoring Experience (GOME) mission launched in 1995
TCWV is typically retrieved in the visible red and near-infrared (NIR) spectral range
Water vapor absorption cross section.
Parameters and settings of water retrieval in the blue band used in previous studies and this work.
The objective of this study is to develop a TCWV-retrieval algorithm for spectroscopic satellite observations which fulfills the following requirements. First, the algorithm should be feasible for the current and forthcoming satellite sensors such as OMI, S5P, S4 and S5. Second, the retrieval should not rely on input from the chemistry transport model (CTM) to avoid propagating model errors into the climatological measurement records. Last, the retrieval should provide a realistic error estimation as measurement uncertainty is an important parameter for data assimilation and future harmonization of satellite data. Based on the results from previous studies, we have further optimized the spectral analysis settings for the TCWV retrieval and developed a statistical analysis approach to optimize the a priori water vapor profile used in the retrieval. In addition, a comprehensive error estimation is also included in the new water vapor retrieval algorithm. The developed algorithm has been implemented to retrieve TCWV from GOME-2 observations; in the future, we will extend the application to other, similar satellite sensors. For validation, the new TCWV data set retrieved from GOME-2 observations is compared to the GOME-2 operational product, ground-based sun photometer and radiosonde measurements.
The paper is organized as follows. Section
In this section, the GOME-2 instruments and the level 1B products used in the retrieval are described. Brief descriptions of the operational GOME-2 TCWV product, sun photometer TCWV data set and the radiosonde measurements used to validate the new GOME-2 TCWV data are presented. In addition, the ERA-Interim data set used for the statistical analysis of water vapor vertical distribution is also presented.
The Global Ozone Monitoring Experiment 2 (GOME-2) are passive nadir-viewing, satellite-borne spectrometers on board the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Metop series of satellites. The Metop satellites orbit at an altitude of
The GOME-2 instruments are optical spectrometers equipped with scanning mirrors which enable across-track scanning in the nadir and sideways views for polar coverage
The first step in GOME-2 data processing is the conversion of the detector signal (level 0 data) to geolocation and radiometric-calibrated radiance and irradiance data (level 1B data). GOME-2 observations taken before 25 June 2015 were processed by the level 1B processor version 6.0, while GOME-2 data taken after 25 June 2015 were processed by the updated level 1B processor version 6.1. The processor update mainly resolved spectral artefacts in the GOME-2 on-ground calibration key data. The spectral artifact in the level 1B data is due to the incomplete removal of the xenon line in the GOME-2 calibration key data. The calibration key data were taken during the preflight on-ground calibration, and the calibration key data are used as input for the level 0 to level 1B data processing. The effect of the spectral contamination in level 1B data processed by the version 6.0 processor is significant at the blue band (Band 3) and more significant for wavelengths longer than 460 nm
The operational GOME-2 water vapor product is processed by the German Aerospace Center (DLR) within the framework of EUMETSAT's Satellite Application Facility on Atmospheric Composition Monitoring (AC SAF), using the GOME Data Processor (GDP) version 4.8. The product is used as reference to validate the TCWV retrieved in the blue band. The operational algorithm retrieves water vapor slant columns in the wavelength range of 614–683 nm. The conversion of slant columns to vertical columns uses air mass factors (AMFs) derived from oxygen slant columns measured in the same spectral band. Water vapor absorption in the red band is much stronger (more than an order of magnitude) than that in the blue spectral range (see Fig.
The CIMEL CE-318 sun photometers are used in the AERosol RObotic NETwork (AERONET) to measure direct sun and sky radiance at multiple wavelengths
Locations of sun photometer (red triangles) and radiosonde (blue circles) stations providing co-located TCWV measurements with GOME-2 satellite observations. The size of the markers is proportional to the number of valid observations available.
Radiosonde data are taken from the Integrated Global Radiosonde Archive version 2 (IGRA2) database. The database is managed by the National Centers for Environmental Information (NCEI) of the National Oceanic and Atmospheric Administration (NOAA). The IGRA2 database includes quality-assured radiosonde measurements from over 2700 globally distributed stations. The measurements consist of temperature, relative humidity, dew point depression, wind direction and wind speed at multiple pressure levels. The IGRA2 radiosonde data are publicly available on the website of NCEI (
ERA-Interim is a global atmospheric reanalysis data set produced by the European Centre for Medium-Range Weather Forecasts
The GOME-2 water vapor retrieval algorithm in the blue spectral range follows the classical differential optical absorption spectroscopy (DOAS) approach, which is a standard spectroscopic method for the retrieval of weakly absorbing trace gases
Typical absorption spectroscopy describes the attenuation properties of radiation along an optical path with the Beer–Lambert–Bouguer law. For satellite measurements, the equation can be written as Eq. (
In practice, Eq. (
Characteristic absorption features of different trace gases are then used to determine their concentrations
The spectral fit settings for the retrieval of water vapor slant column.
Slant column densities (SCDs) of water vapor are retrieved from GOME-2 spectra by applying the DOAS spectral fitting technique. The SCD is defined as the integrated concentration along the optical path from TOA through the atmosphere to the Earth's surface and reflected back to the satellite sensor (
An example of the DOAS retrieval of water vapor slant columns from a GOME-2A spectrum over the Pacific Ocean. The retrieved water vapor slant column is 30.6 kg m
The spectral-fitting window is optimized for water vapor retrieval, which includes a relatively strong water vapor absorption structure at about 442 nm. Including liquid water absorption in the analysis effectively eliminates the interference of liquid water and reduces the systematic error above surfaces covered by water
The next step in the TCWV retrieval is the conversion of water vapor SCDs to vertical column densities (VCDs). The VCD (or total column) is defined as the vertical integral of water vapor from the surface to the top of atmosphere. The SCD to VCD conversion is accomplished by using the concept of the air mass factor
Light traveling in the atmosphere can be scattered by air molecules, aerosols and clouds, resulting in a complex optical path. To resolve the optical path and the box air mass factor (
The
Parameters in the box air mass factor look-up table.
For retrieval, the
The vertical distribution of water vapor is important for the conversion of slant columns of water vapor to vertical columns as expressed in Eq. (
Statistical analysis of water vapor vertical profiles from ECMWF ERA-Interim reanalysis data over a small region of the Pacific Ocean (5
By making use of the characteristic that water vapor profile shapes are strongly correlated to their total columns, we have formulated a water vapor vertical profile shape look-up table for the entire globe with a spatial resolution of 0.75
Clouds are treated as opaque Lambertian surfaces in the retrieval algorithm. The treatment of partially cloudy pixels is based on the independent pixel approximation
The cloudy AMF (AMF
AMFs of partially cloudy pixels are calculated as the intensity-weighted average of the
The resulting AMFs are used to divide the measured slant columns and convert the water vapor slant columns into vertical columns. This AMF is used for the iterative optimization of a priori profile of partially cloudy pixels.
The presence of aerosols affects the radiative transfer in the atmosphere and may influence the retrieval of surface properties, cloud and atmospheric water vapor
The error of the TCWV is composed of many sources. Major sources of error can be divided into two parts where one is related to the measurement itself, and the other is related to the uncertainties of assumptions in the retrieval. The uncertainty of the TCWV can be derived analytically through error propagation. As the retrieval of TCWV is separated into two major steps, namely slant column retrieval and AMF calculation, the error estimation also follows these two steps. The uncertainty of TCWV can be express as Eq. (
The uncertainties of water vapor slant column are mainly attributed to the instrument noise, instrument characteristics and the uncertainties related to the DOAS retrieval of the slant column. Instrument noise is expected to cause random errors, and this error can be quantified by analyzing the DOAS fit residual
The uncertainty of the AMF is mainly related to the uncertainties of each input parameter used in the AMF calculation. These input parameters include the solar and viewing geometries, surface albedo, surface pressure and water vapor vertical profile. The solar and viewing geometries are well calibrated, and their errors are mainly related to the interpolation of the box AMF look-up table. These uncertainties are negligible compared to other sources of error. The contribution to the AMF uncertainty from the remaining sources of error can be estimated by the AMF sensitivity (or Jacobian) with respect to each parameter
In this study, surface albedo is taken from the surface reflectance climatology at 440 nm, which is derived from GOME-2 measurements from 2007 to 2013. The uncertainty of surface albedo (
The error related to the a priori vertical distribution of water vapor is determined by using the a priori water vapor from the last iteration plus 1
The calculation of the uncertainty of the cloudy AMF is similar to the one used for the clear-sky AMF, with surface albedo and surface pressure uncertainties replaced by cloud albedo and cloud top pressure errors. In this study, the cloud top pressure error is assumed to be 50 hPa
Following Eq. (
Combining the slant column density error with the AMF error, the error of TCWV can then be calculated following Eq. (
Summary of the major sources of error in the water vapor retrieval.
The ground pixels of the satellite observations vary in size and shape, and often multiple pixels overlap in higher latitudes. To better reconstruct the spatial distribution of satellite observations and compare the results to different data sets, the retrieved GOME-2 water vapor columns are gridded onto a high-resolution latitude–longitude grid with a spatial resolution of
In this study, water vapor columns retrieved from GOME-2 observations in the blue band are compared to ground-based sun photometer and radiosonde measurements. As the satellite, sun photometer and radiosonde data are different in spatial and temporal resolution and coverage, only coinciding data are used in the comparison. The criteria used to select coinciding data are that the (1) satellite data are selected so that the center coordinate of the satellite pixel is within 50 km of the sun photometer or radiosonde site, and the (2) sun photometer or radiosonde data are selected around the satellite overpass time so that the time difference between the satellite and ground observations is less than 2 h. Subsequently, satellite, sun photometer and radiosonde measurements are averaged to daily data for comparison. As the sun photometer only provides data under clear-sky conditions, satellite data are filtered for intensity-weighted cloud fractions smaller than 0.5 for consistency. Daily averaged GOME-2 data are used for the comparison to sun photometer and radiosonde measurements.
In this section, we present validation studies of GOME-2 TCWV retrieved in the blue spectral range. Our retrieval results are compared to the GOME-2 operational water vapor product which is derived in the red spectral band. In addition, the new data set is validated against ground-based sun photometer observations and radiosonde measurements.
Figure
Monthly average TCWV derived from the GOME-2A observations. Panels
The water vapor columns derived from the blue band and the operational retrieval are sorted by their measurement latitudes and are plotted in Fig.
Comparison of TCWV from the blue retrieval and operational algorithm.
Figure
Monthly zonal average of TCWV from GOME-2A observations in the
Figure
Time series of a Pearson correlation coefficient between water vapor columns from the blue band retrieval and operational algorithm is shown in
In addition, we have compared the TCWV measured by both GOME-2A and GOME-2B to investigate the cross-sensors consistency. The mean water vapor column retrieved from GOME-2A observations from 2013 to 2014 is 20.72 kg m
Figure
Time series of annual average of TCWV retrieved from
Figure
Comparison of the TCWV, measured by the sun photometer, to GOME-2A is shown in
Figure
Comparison between sun photometer and GOME-2A observations. Panels
The scatterplots of the radiosonde TCWV measurements compared to the GOME-2A and GOME-2B measurements are shown in Fig.
Comparison of radiosonde measurements of TCWV to GOME-2A is shown in
Figure
Comparison between radiosonde and GOME-2A observations. Panels
The spatial distribution of water vapor from the blue band and operational retrieval shows good consistency. However, the blue band retrieval shows significantly higher values over west Africa, India and Southeast Asia Peninsula and slightly lower values over oceans in the tropics in July. A previous study reported that the operational GOME-2 product is underestimating water vapor columns over land and overestimating them over oceans in the tropics
Overestimation of TCWV can also be observed over South America in both summer and winter. The discrepancies are likely related to the uncertainties of Lambertian assumption of surface albedo over vegetation. The bidirectional reflectance distribution function (BRDF) effect has been reported to have significant impacts on the retrieval of cloud and trace gas over forested scenes
Compared to the operational product, the blue retrieval is overestimating the water vapor columns at upper latitudes and underestimating in the tropics and resulting a small overestimation (
A previous comparison of the GOME-2 operational water vapor product to radiosonde measurements shows that the operational product is, on average, underestimating water vapor columns over land by 1.0 kg m
The small positive offsets between GOME-2 and sun photometer measurements indicate that the blue band retrieval slightly overestimates the TCWV. On the other hand, the sun photometer data have been reported to underestimate TCWV by 6 %–9 % compared to GPS data
The analysis of bias between GOME-2 and sun photometer measurements shows a larger variation during summer months of the Northern Hemisphere. This is partly related to the geolocation distribution of the sun photometer stations. Most of the stations are situated in the Northern Hemisphere and result in a larger number of valid measurements and variations in summer. In addition, both GOME-2A and GOME-2B are slightly overestimating the water vapor columns by
The small overestimation of water vapor columns by GOME-2 compared to radiosonde measurements is partly related to the level 1B data issued before 2015. If we only consider data taken after 2015 in the comparison, the overestimation of GOME-2A is reduced from 1.20 to 0.36 kg m
In this work, we have developed a water vapor retrieval algorithm in the visible blue band of 427.7–455 nm, providing an alternative solution for satellite sensors that do not cover the red band where TCWV is typically retrieved. The major advantage of the new water vapor retrieval algorithm is that it does not rely on a priori information from a chemistry transport model. This improvement makes the satellite product independent from model simulations and avoids model errors propagating to the measurement, making the data more suitable for climate studies.
The developed TCWV retrieval has been successfully applied to GOME-2. Water vapor columns retrieved in the blue band show very good spatiotemporal consistency with the operation product, sun photometer and radiosonde measurements. However, reprocessing of GOME-2 level 1B data before 2015 is necessary to produce a reliable climate record. The blue band retrieval results are consistent between GOME-2A and GOME-2B, with discrepancies of less than 2 %. The retrieval is feasible enough to be applied to former, current and forthcoming UV and Vis satellite sensors to create an independent water vapor climate data record starting from 1995 and continuing for the next two decades.
We are planning to make the GOME-2 TCWV data publicly available through the Earth Observation Center (EOC) of the German Aerospace Center (DLR). However, it takes time to set up the data server. For the time being, the data are available on request from the corresponding author (ka.chan@dlr.de).
KLC conceptualized the paper with DL. KLC curated and analyzed the data, led the investigation, devised the methodology, developed the software and validated the data. CK assisted with the data validation. KLC prepared the manuscript with contributions from all the co-authors. PV and SS assisted in reviewing and editing of the paper.
The authors declare that they have no conflict of interest.
The work described in this paper was supported by the German Aerospace Center (DLR) programmatic research.
The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association.
This paper was edited by Cheng Liu and reviewed by three anonymous referees.