1. Introduction
Water vapor in the atmosphere varies considerably in time and space. These variations have considerable influence on weather and climate systems at local and global scales (American Geophysical Union 1995; Gradinarsky et al. 2002; Minschwaner and Dessler 2004). As the primary greenhouse gas, it plays an important role in the control of temperature in the troposphere as well as in cloud formation (Held and Soden 2000). The interactions of water vapor with the other atmospheric and climate components are complex and global (Inamdar and Ramanathan 1998; Soden et al. 2002). The spatial and temporal characterization of water vapor variation remains a global challenge in spite of the motivational impetus provided by the obvious need to better understand this key atmospheric constituent. Significant progress was made over the last few decades concerning water vapor observations and the modeling of these observations using general circulation models (GCMs.) The current satellite-based monitoring systems (Johnsen and Kidder 2002; Dubuisson et al. 2004) can be used to track weather systems ranging in size from a few kilometers to several hundred kilometers in dimension. Ground-based observations of water vapor are nonetheless still needed to monitor local phenomena and climate trends at the local scale. There are several ground-based measurement systems such as radiosondes, passive microwaves, GPS meteorology, and sun radiometry for water vapor monitoring (Bevis et al. 1994; Bokoye et al. 2003).
The research presented in this paper is performed in the context of worldwide sun radiometer networks such as the Aerosol Robotic Network (AERONET; Holben et al. 1998) and associated networks such as AEROCAN (Bokoye et al. 2001). To enhance the monitoring capabilities of these primarily aerosol monitoring networks (more than 150 instruments across the world) a channel for water vapor retrieval is included as part of the standard instrumental protocol. Calibration and validation requirements for data quality assurance of both aerosol and water vapor channels are demanding at the scale of a large network such as AERONET. In this paper we propose an independent and instrument-specific water vapor calibration, a method based on intercomparisons between sun radiometer and GPS meteorology. For that purpose, the high temporal resolution of GPS observations is more appropriate than that of radiosonde limited to two observations per day. Furthermore, the considered methodology appeared to be a simple means of dynamically adapting calibration constants retrieved by traditional radiometric techniques. The data required to develop and test this new calibration method are presented in section 2. The methodologies related to integrated water vapor (IWV) retrievals from the 940-nm water vapor channel calibration process are outlined in section 3. The results are then presented in section 4 and discussed in section 5.
2. Database description
Table 1 shows the operating sites and observation periods (between 2000 and 2004) for the sun radiometric and GPS databases employed in this study. Meteorological data, acquired alongside GPS recordings, were needed for reducing GPS raw data into IWV output. Sun radiometric IWV data (IWV-SUN) were retrieved from the AERONET/AEROCAN network (Bokoye et al. 2001). The extinction and sky-scanning instrument employed in this network is the CIMEL Electronique multiwavelength and automatic sun radiometer. This instrument permits IWV retrievals from direct solar measurements in the 940-nm water vapor channel (Halthore et al. 1997; Holben et al. 1998). The data corresponding to IWV-SUN were, until recently, processed and archived according to an earlier version of the AERONET IWV retrieval procedure (referred to as V1; Holben et al. 1998). A new retrieval procedure, V2, was introduced in 2004 (Smirnov et al. 2004). Three product levels were defined with AERONET: raw level: 1.0, cloud-screened level: 1.5 (Smirnov et al. 2000), and quality-assured level: 2.0 (Holben et al. 2001). The difference between V1 and V2 retrieval procedures is discussed in sections 4 and 5.
GPS observations from Churchill, Manitoba, Canada (58°43′30″N, 94°07′00″W), and Victoria, British Columbia, Canada (48°23′23.2″N, 123°29′14″W), were derived using dual-frequency GPS receivers from the active network run by the Geodetic Survey Division of Natural Resources Canada. GPS data from Sherbrooke, Quebec, Canada (45°22′N, 71°55′W), was collected from our own dual-frequency GPS receiver system (Bokoye et al. 2003). Note that the Sherbrooke site is referred to as the Centre d’Applications et de Recherches en Télédétection (CARTEL) within the AERONET network. Observational GPS data consist of calibrated satellite code and carrier phase observations from continuous tracking GPS satellites at 30-s intervals. Data are archived daily in Receiver Independent Exchange Format (RINEX), version 2. Precise satellite ephemerides (GPS orbits) and precise satellite clock corrections necessary for GPS-based IWV retrieval were also downloaded from the International GPS Service (IGS) Web site. A meteorological station is associated with each GPS receiver station. Atmospheric pressure and air temperature were recorded at 15-min intervals and archived in a daily file. The processing methods applied to the overall database are briefly outlined in the following section.
3. Methodology
This section describes IWV retrieval procedures using sun radiometry and GPS meteorology. The calibration principle employed for the sun radiometer water vapor absorption channel is first reviewed.
a. Water vapor from sun radiometry
b. Water vapor from GPS meteorology
The principle of IWV determination from the GPS constellation is based on the delay introduced by the atmospheric water vapor in the propagation of the GPS signal. The absolute value of IWV can be computed from raw GPS observation data associated with the meteorological data (the GPS observations include RINEX data and orbital files from the International GPS Service for Geodynamics). In the present study, a geodetic processing software program “GPSpace” (Bokoye et al. 2003) based on the precise point positioning (PPP) technique was used to reduce dual-frequency GPS receiver and IGS orbital data to zenith tropospheric delay (ZTD). The latter can be partitioned into wet delay (ZWD) and hydrostatic delay (ZHD). ZHD is inferred from ground pressure measurements. Knowledge of ZTD and ZHD allows one to retrieve ZWD by a simple difference. ZWD is then calibrated to IWV. The IWV/ZWD calibration constant depends on meteorological conditions. A full description of IWV retrieval from GPS meteorology is given in Bokoye et al. (2003). Note that GPS observations were made every 30 s.
The accuracy of the GPS meteorology method is of the same order as that of radiosonde IWV retrievals, that is, about 0.1 g cm−2 (Bokoye et al. 2003; Mätzler et al. 2002; Niell et al. 2001) on average. Note that in certain cases the application of a correction algorithm to radiosonde raw data is necessary to improve the accuracy associated with this observation technique (Miloshevich et al. 2006). Errors associated with water vapor profiles are observed in the upper atmosphere using independent observation systems like lidar (Soden et al. 2004), but the impact of upper-atmospheric water vapor is limited as compared to the total columnar IWV. In fact, the lower atmosphere represents about 90% of total water vapor.
c. Sun radiometric calibration
In the context of the present study, sun radiometric calibration is assumed to be the process of determining values for the constants Vo(λ), a, and b [Eq. (3)].
d. Sensitivity of a and b to 940-nm filter characteristics
Sun radiometer filter characteristics degrade due to aging caused by the effects of long time solar exposure, temperature cycles, humidity, and lens surface contamination. The extent of the deterioration affects the spectral passband of the filter (Holben et al. 2001).
Note that a and b are retrieved from Eq. (5). Let us consider first the influence of a spectral shift on the evaluation of a and b. Figure 3 shows the derived values of a and b for the nominal 940-nm case and those derived by adding or subtracting a spectral shift of 2, 4, 6, and 8 nm to this nominal central wavelength. There is no direct experimental validation of this sensitivity analysis performed using synthetic transmittance data generated with the MODTRAN radiative transfer code for the above spectral shift values (with a maximum extent of ±8 nm) and the water vapor column content. However, several studies in spatial remote sensing (Vermote and Kaufman 1995; Bréon and Jackson 1999) reported spectral drifts of several nanometers of considered water vapor channel bandpass due to atmospheric contamination on the measurement or a degradation due to solar radiation effects.
The derived (a, b) values for the nominal 940-nm filter spectral response remain in the range of those reported by Halthore et al. (1997) and are comparable to the fixed AERONET V1 values as well as to the AERONET V2 values (see Table 7 where the V2 values of both a and b range between 0.6 and 0.7). When the central wavelength of the filter shifts (Fig. 3), a significant systematic decrease in the constant a is observed between 0-nm spectral shift and the extreme limits of the ±8-nm shift. The corresponding maximum relative variation in a is, respectively, −68% and −18% for the forward and backward shifts (Fig. 3). The constant b decreases moderately between the 0-nm and +4-nm spectral shifts by −7% and increases by about +6% for both the forward and backward 8-nm shifts. The combined variation of (a, b) introduces a strong variation in IWV retrievals. In fact, a typical error in IWV retrievals of more than 100% is obtained between a +4-nm spectral shift case and the nominal 940-nm case. These results indicate that the calibration process should take into account the possibility of spectral shifts as reported by Heath et al. (1998), who investigated environmental effects on filter central wavelength and bandwidth.
In the AERONET V2 procedure, a and b are computed using state-of-the-art spectroscopy (Rothman et al. 2003) and a measured (if available) filter function (which is different for each instrument). The LBLRTM radiative transfer code (Smirnov et al. 2004) is used to compute atmospheric transmittances in the 940-nm region and to subsequently extract a and b using the slope and intercept of the simulated function log(Tw) versus m × IWV (in similar fashion to the fitting procedure carried out in Fig. 2). For a given atmospheric state (a given value of IWV) computations are carried out over a range of airmass values (a range of m × IWV values). Note that each AERONET instrument has its own unique set of a and b values depending on the filter configuration. They are considered fixed until the filter is changed.
The effect of possible changes in filter response amplitude and the subsequent impact on integrated transmittance variation was also considered (without spectral shift). For example a reduction of filter output signal amplitude could lead to a decrease of water vapor channel signal measurement sensitivity, which would increase observation errors.
These errors could be reduced with a normalization of the signal. The influence of the amplitude can be linked also to some variations in filter spectral response shape, which is not studied in this paper.
For this purpose Tw was simulated using MODTRAN and the 940-nm nominal passband filter response (Fig. 1) while adding a certain percentage of random noisy signals (values between 0 and ± a certain percentage of random noise generated using the Matlab software random function). Then Tw was computed for each simulated IWV-GPS measurement. Our objective was to analyze the influence of random changes in filter passband amplitude on a and b retrievals. Table 2 shows the retrieved values of a and b for 0%, 10%, 30%, and 50% random noise in the nominal 940-nm filter passband response. These a and b values were always retrieved from log(Tw) versus m × IWV fits to the ensemble of input IWV-GPS values. The relative difference between the addition of 0% and 50% uncertainty in the nominal amplitude is −29.6% and −7.5%, respectively, for a and b (Table 2).
The above analysis shows that the coefficients a and b can undergo significant variations as a result of changes in the nominal 940-nm filter characteristics. These variations directly influence the IWV retrieval represented by Eq. (5). The constants a and b are typically computed using a nominal filter spectral response; while it would appear that it is necessary to account systematically for real variations in filter properties, the development of such a network-wide procedure by frequent and direct measurement of filter characteristics would be practically impossible on an operational scale (i.e., beyond the relatively infrequent updates provided by the AERONET V2 protocol). A dynamical combination of sun radiometry and GPS meteorology is the proposed alternative approach (see discussion in section 4b). This technique would allow for the determination of a and b without having to develop a protocol for the frequent measurement and updating of filter characteristics. Such a procedure appears to be very important in the framework of the development of an accurate worldwide sun radiometer–derived water vapor climatology. In ground-based sun radiometry the IWV retrieval is generally made in parallel to atmospheric aerosols observations through τa(λ) determination. Unfortunately, in existing sun radiometer networks including aerosols and water vapor channels, the logistical support is generally prioritized for τa(λ) determination, the main objective of such networks like AERONET. Our proposed procedure for A, a, and b determination in the water vapor absorption channel can be considered as a contribution to increase the water vapor sun radiometer network accuracy using a single sun radiometer site-to-sun radiometer site mobile GPS unit on a limited timeframe window.
4. Results
a. IWV-SUN versus IWV-GPS
Time series of IWV values derived from sun radiometry [V1 downloaded data] and GPS meteorology are shown in Fig. 4 for the three AERONET/AEROCAN sites. A general agreement in trend can be observed for the two independent measurement systems with a typical IWV increase during the summer. Table 3 summarizes the results of the comparison between IWV values derived from AERONET 940-nm sun radiometry [IWV-SUN (V1)] data and IWV values retrieved from simultaneous GPS observations (IWV-GPS) for each instrument. The comparison was performed for both AERONET levels 1.5 and 2.0 aerosol optical depth values [τa(940)] employed as the correction term of Eq. (3) [in Table 3, level 1.5 and level 2.0 refer only to the application of different interpolated τa(940) values]. As compared to the reference IWV-GPS data, the IWV root-mean-square error (rmse) for level 1.5 ranges between 0.14 g cm−2 (Churchill in 2002) and 0.48 g cm−2 (Sherbrooke in 2001) with a three-station average of 0.23 ± 0.11 g cm−2. The latter is about 22% of the mean observed IWV-GPS value of 1.00 ± 0.66 g cm−2 for all 2672 observations. The bias error ranges between −0.41g cm−2 (Sherbrooke in 2001) and 0.12 g cm−2 (Sherbrooke in 2003) with a three-station mean bias of −0.09 ± 0.16 g cm−2. No statistically significant improvement is obtained using the level 2.0 results (Table 3) relative to the level 1.5 results. Thus, the influence of non-cloud-screened versus cloud-screened atmospheric aerosol measurements on the 940-nm water absorption channel is weak for the data ensemble employed in our study. This suggests, as well, that the impact of AOD variations on IWV retrievals is minimal. AOD correction, such as the dynamic correction employed in the AERONET V2 protocol, does nonetheless contribute to an uncertainty reduction in water vapor retrieval (particularly in a hazy atmosphere). Except for Sherbrooke in 2001, the errors in Table 3 remain generally in the range of 0.15–0.2 g cm−2. This is similar to most IWV comparisons between various techniques such as GPS meteorology, radiosonde, numerical weather prediction models, or microwave radiometers (Liou et al. 2000; Mätzler et al. 2002; Niell et al. 2001; Pugnaghi et al. 2002). The large error obtained for Sherbrooke in 2001 may have been linked to a substantial calibration degradation, which may, in turn, have been induced by instrument misalignment or contamination.
Table 4 gives an overview of the calibration constants (A, a, and b) derived for each site and instrument, with the associated fitting error. An example of the retrieval approach for deriving these calibration constants using the IWV-GPS as reference is shown in Fig. 5 for the largest rmse of Table 4 [Eq. (6); Y versus slant water vapor]. The observed variations are reasonably well described by the regression fit (solid line in Fig. 5), despite some scatter possibly due to anomalous observations (cloud coverage, sun radiometer misalignment, etc.). Note that the fitting process itself [Eq. (5)] could also induce variations in (a, b) that are not strictly independent.
The AERONET A values are also reported in the same table for comparison. The AERONET V1 calibration constant (A) does not vary significantly (Table 4). This fact is linked to the constancy of a and b values in the V1 procedure. The mean difference between A from the AERONET V1 protocol and the derived GPS-based calibration values is about 5%. The derived (a, b) values of Table 4 are somewhat different from the nominal AERONET V2 values (see next section) and from those reported by Halthore et al. (1997) or Schmid et al. (1996). Filter degradation (i.e., changes in center wavelength and bandwidth characteristics) with time over the dataset period, or sun radiometer calibration errors (in A), may explain such differences in a and b. Note also that even though each CIMEL sun radiometer is equipped with the same type of ion-assisted deposition interference filters (Holben et al. 1991), the comparison with (a, b) values from literature are subject to the difference between filter responses.
The most important observation that can be drawn from the results of Table 4 is the fact that these constants change significantly in time and between instruments. This suggests different changes through time induced by differences in filter properties as suggested in the previous section. Such variations could not be entirely compensated for by adjusting the constant A (which is effectively what is done as part of the AERONET V1 protocol). The new AERONET V2 protocol (Smirnov et al. 2004) takes the filter characteristics in the a and b retrieval into account.
The use of these GPS-based constants for calibrating the sun radiometer calibration is analyzed in the next section and discussed in comparison with AERONET V2 protocol in section 5.
b. Analysis of the improvement obtained in the IWV-SUN retrieval using GPS-based calibration
This section describes the evaluation of the GPS-based IWV calibration compared with the V1 and V2 AERONET-based procedures. The evaluation is performed using the database corresponding to IWV-GPS data resampled to the sun radiometer times to create an ensemble of IWV-GPS points matched to the times of all sun radiometer signals. For a given instrument and period, this joint database was partitioned into two randomly selected subsets that were nearly equal in terms of number of observations and totally independent. The first subset is used to simultaneously compute the constants A, a, and b (GPS-based calibration) and the second is used for the comparison between the three calibration procedures, that is, IWV-GPS versus IWV-SUN (AERONET V1 calibration), IWV-GPS versus IWV-SUN (AERONET V2 calibration), and IWV-GPS versus IWV-SUN (GPS-based calibration). In this comparison, sun radiometric data were screened to retain only the data falling within the limit of uncertainty corresponding to typical radiosonde accuracies of 10% in IWV [as previously shown in Fig. 5, some points depart significantly from a fit to Eq. (5)]. These outliers might correspond to residual cloud cover or misalignment of the instrument in the solar direction. They can increase errors associated with IWV retrievals, and in turn, in the errors of the calibration constants derived from the linear regression-based procedure [Eq. (5)]. We thus chose to consider only values within 10% accuracy of IWV. This limit of uncertainty corresponds to a 1% departure from the model, that is, in the relative difference between predicted and observed logarithm of the sun radiometric signal as shown in Fig. 6a.
The above 1% threshold uncertainty criterion in Y was applied to the 940-nm signal for the first dataset yielding the constants A, a, and b (see the Saturna Island example in Fig. 6b, which is the filtered version of Fig. 5). One can note that in this case the application of the above screening procedure (and the subsequent reduction in the number of points) did not change the derived A, a, and b values. However, the 1% criterion in Y allowed the screening of outlier radiometric records that corresponds to bad observations (cloud coverage, instrument misalignment, sensor contamination, etc.).
The above approach was applied to the two databases having the greatest number of points, that is, Saturna Island (in 2002, instrument 81) and Sherbrooke (in 2003, instrument 82). It must be noted that the generation of a time-synchronous database between AERONET V1 and V2 databases and the GPS-retrieval database significantly reduced the number of GPS retrieval points. The retrieved A, a, and b values used for the sun radiometer calibration are reported in Table 5. The (IWV-SUN versus IWV-GPS) scatterplots for both AERONET V1 and GPS-based calibration are compared in Figs. 7a,b, respectively, for each site. In both cases, the GPS-based calibration significantly reduces the bias observed for IWV-SUN relative to the AERONET V1 and V2 calibration: the absolute value of the bias error in Table 6 indicates a reduction by a factor 3 and 11, respectively, for Saturna Island (in 2002) and Sherbrooke (in 2003). The bias reduction and the small but systematic rmse reduction of 30% and 40%, respectively, support the validation of the IWV calibration approach for the two test databases. One can note that results from AERONET V1 and V2 procedures are nearly the same for the two sites. However, Smirnov et al. (2004) reported more significant differences with a slope of 0.85, a y intercept of −0.05, and a correlation coefficient of 0.99 for a regression analysis between V2 and V1 carried out over a number of different sites. A V1 and V2 comparison using a large ensemble of data from the Sherbrooke site (1995–2004 data downloaded in March 2005, 15 615 observations) revealed a slope of 0.99, a y intercept of 0.00, and a correlation coefficient of 0.99. These results suggest that the difference between V1 and V2 depends on the site and may vary with the sun radiometer calibration history.
The results of this section indicate that improved calibration can be achieved using the GPS calibration methodology. This calibration approach can be easily implemented as part of a sun radiometer network calibration protocol by deploying one or more mobile bifrequency GPS receiver units to all stations during a regular calibration cycle.
5. Discussion
The new AERONET calibration procedure (Smirnov et al. 2004) referred to as V2 in this paper produces a and b values for each filter instrument. The calibration coefficient A is thereafter evaluated for a field instrument by intercomparison with a reference master sun radiometer. This new procedure resulted in a computed IWV reduction of about 13%–15% relative to the older AERONET V1 procedure (Smirnov et al. 2004) for the AERONET site of Greenbelt, Maryland. For the sites considered in this study the difference between V1 and V2 appeared to be statistically nonsignificant.
The IWV comparison between the AERONET V2 procedure and our GPS-based calibration approach for the Saturna Island site yielded an rmse of 0.24 g cm−2 and a bias error of 0.22 g cm−2 while the results for the Sherbrooke site were 0.15 g cm−2 rmse and a bias of −0.09 g cm−2. Table 6 highlights the difference between the AERONET V1 and V2 procedures and the GPS meteorology for various sites. Table 7 shows the a and b coefficients derived using V2. The results yield a and b values that differ from those of Table 5. This may be due to several factors including the fact that A, a, and b are not strictly independent in the iteration retrieval algorithm even if the set of (A, a, b) triplets that we employed yielded the lowest rmse. In other words, different triplet sets could give nearly the same rmse values.
Furthermore, the influence of the period over which regressions were performed on the A, a, and b retrieval procedure was investigated. Retrieval of the above parameters on a daily basis (computation of a and b for a daily observations) for the Saturna Island site (see Fig. 8) highlights the fact that the varying atmospheric conditions (as represented by different days of the year) can significantly affect the values of the derived constants. The coefficient of variation (standard deviation/mean) are, respectively, 1.28% for A, 9.07% for a, and 14.90% for b. The mean values of the daily-based averaged triplets over 3 months of measurements (Figs. 8 a–c, day 20 to day 180 in 2002) indicate a relative stability with a = 0.40 ± 0.07, b = 0.61 ± 0.11, and A = 9.15 ± 0.09. These values are reasonably close to the triplets derived from all 3-month data groupings (Fig. 9): a = 0.35, b = 0.78, and A = 9.36. The above sets of A, a, and b values remain different from the AERONET V2 mean values (see Table 7), respectively, by 4.32%, 43.8%, and 17.98% in terms of the coefficient of variation. The difference between the two approaches could result from the influence of possible filter aging effects in the AERONET V2 procedure. In this case, constants a and b remain constant for a given filter until its change; only the constant A is allowed to vary in the calibration fit. Our proposed GPS-based calibration procedure is advantageous in that it integrates all possible error sources such as detection response, bad filter characterization, or possible stray light (internal reflections).
6. Conclusions
The retrieval of IWV using the original AERONET V1 and AERONET V2 procedures can be validated and improved using GPS meteorology. Comparative results between IWV-SUN (V1) and IWV-GPS for seven instrument datasets from three sites in Canada over 3 yr highlight important temporal and site-to-site variations of the sun radiometer IWV accuracy with an rmse ranging between 0.14 and 0.48 g cm−2 and a bias between both databases (0.09 ± 0.16 g cm−2). These changes in accuracy are likely related to variations in the 940-nm filter response. It was demonstrated that the station-to-station and period-to-period variations could have a significant influence on the retrieved values of the three calibration constants (A, a, and b). The new AERONET V2 IWV retrieval procedure that employs instrument-dependent (a, b) constants is an improvement on the earlier V1 procedure.
The relevance of GPS meteorology for IWV calibration in sun radiometry was demonstrated in this work. The results show that the comparison between the retrieved values of IWV-GPS and IWV-SUN using GPS-based calibration reduced the rms and bias error (relative to more traditional nondynamic techniques) to about 0.1 (10%) and <0.06 g cm−2, respectively.
Given the advantage of simple, automatic, and continuous IWV measurements, GPS meteorology appears to be a powerful technique for assuring the quality and operationalization of water vapor retrievals.
The authors are grateful to Aichatou Ibrahim and Ovidu Pancrati for their help in data processing. We wish to thank Brent Holben and the AERONET staff at the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC) for their commitment to the concept of providing timely public access to AERONET data. We are very grateful in particular to Alexander Smirnov for his contribution in terms of data and information on the AERONET IWV products. We wish to also thank Jim Freemantle and Bruce McArthur (MSC, Environment Canada) for their continuing contribution to AEROCAN, the Canadian sun radiometers network. The financial support of the Canadian Foundation for Climate and Atmospheric Sciences (CFCAS), FQRNT, Québec, and NSERC Canada is gratefully acknowledged. We would also like to acknowledge the three anonymous reviewers, whose contributions have enriched the paper.
REFERENCES
American Geophysical Union, 1995: Water vapor in the climate system. Special Rep., 17 pp. [Available from the American Geophysical Union, 2000 Florida Ave., N.W., Washington, DC 20009.].
Angstrom, A., 1929: On the atmospheric transmission of Sun radiation and on dust in the air. Geogr. Ann., 12 , 130–159.
Bevis, M., , Businger S. , , Chiswell S. , , Herring T. , , and Anthes R. , 1994: GPS meteorology—Mapping zenith wet delays onto precipitable water. J. Appl. Meteor., 33 , 379–386.
Bokoye, A. I., , Royer A. , , O’Neill N. T. , , Fedosejevs G. , , Teillet P. M. , , McArthur L. J. B. , , and Cliche P. , 2001: Characterization of atmospheric aerosols across Canada from a ground-based sunphotometer network: AEROCAN. Atmos.–Ocean, 39 , 429–456.
Bokoye, A. I., , Royer A. , , O’Neill N. T. , , Cliche P. , , McArthur L. J. B. , , Teillet P. M. , , Fedosejevs G. , , and Thériault J-M. , 2003: Multisensor analysis of integrated atmospheric water vapor over Canada and Alaska. J. Geophys. Res., 108 .4480, doi:10.1029/2002JD002721.
Bréon, F-M., , and Jackson D. , 1999: Evidence of atmospheric contamination on the measurement of the spectral response of the GMS-5 water vapor channel. J. Atmos. Oceanic Technol., 16 , 1851–1853.
Bruegge, C. T., , Conel J. E. , , Green R. O. , , Margolis J. S. , , Holm R. G. , , and Toon G. , 1992: Water vapor column abundance retrieval during FIFE. J. Geophys. Res., 97 , 18759–18768.
Coleman, T. F., , and Li Y. , 1994: On the convergence of reflective Newton methods for large-scale nonlinear minimization subject to bounds. Math. Programm., 67 , 189–224.
Coleman, T. F., , and Li Y. , 1996: An interior, trust region approach for nonlinear minimization subject to bounds. SIAM J. Optimiz., 6 , 418–445.
Dubuisson, P., , Dessailly D. , , Vesperini M. , , and Frouin R. , 2004: Water vapor retrieval over ocean using near-infrared radiometry. J. Geophys. Res., 109 .D19106, doi:10.1029/2004JD004516.
Gradinarsky, L. P., , Johansson J. M. , , Bouma H. R. , , Scherneck H-G. , , and Elgered G. , 2002: Climate monitoring using GPS. Phys. Chem. Earth, 27 , 335–340.
Halthore, N. R., , Thomas F. E. , , Holben B. N. , , and Markham B. L. , 1997: Sun photometric measurements of atmospheric water vapor column abundance. J. Geophys. Res., 102 , D4. 4343–4352.
Heath, D. F., , Hilsenrath E. , , and Janz S. , 1998: Characterization of a “hardened” ultrastable UV linear variable filter and recent results on the radiometric stability of narrow band interference filters subjected to temperature/humidity, thermal/vacuum and ionizing radiation environments. Proc. SPIE, 3501 , 401–411.
Held, I. M., , and Soden B. J. , 2000: Water vapor feedback and global warming. Annu. Rev. Energy Environ., 25 , 441–475.
Holben, B. N., , Eck T. F. , , and Frazer R. S. , 1991: Temporal and spatial variability of aerosol optical depth in the Sahel region in relation to vegetation remote sensing. Int. J. Remote. Sens., 12 , 1147–1163.
Holben, B. N., and Coauthors, 1998: AERONE—A federated instrument network and data archive for aerosol characterization. Remote Sens. Environ., 66 , 1–16.
Holben, B. N., and Coauthors, 2001: An emerging ground-based aerosol climatology: Aerosol Optical Depth from AERONET. J. Geophys. Res., 106 , 12067–12097.
Inamdar, A. K., , and Ramanathan V. , 1998: Tropical and global scale interactions among water vapor, atmospheric greenhouse effect, and surface temperature. J. Geophys. Res., 103 , D24. 32177–32194.
Ingold, T., , Schmid B. , , Mätzler C. , , Demoulin P. , , and Kämpfer N. , 2000: Modeled and empirical approaches for retrieving columnar water vapor from solar transmittance measurements in the 0.72, 0.82, and 0.94 μm absorption bands. J. Geophys. Res., 105 , D19. 24327–24344.
Johnsen, K-P., , and Kidder S. Q. , 2002: Water vapor over Europe obtained from remote sensing sensors and compared with a hydrostatic NWP model. Phys. Chem. Earth, 27 , 371–375.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kasten, F., 1966: A new table and approximate formula for relative optical air mass. Arch. Meteor. Geophys. Biochlimatol. Ser. B, 14 , 206–223.
Kneizys, F. X., , Shettle E. P. , , Abrieu L. W. , , Chetwynd J. , , Anderson G. , , Gallery W. , , Selby J. , , and Clough S. , 1988: User’s guide to LOWTRAN-7. AFGL-TR-88-0177, Hanscom Airforce Base, MA, 140 pp.
Liou, Y-A., , and Huang C-Y. , 2000: GPS observation of PW during the passage of a typhoon. Earth Planets Space, 52 , 709–712.
Liou, Y-A., , Huang C-Y. , , and Teng Y-T. , 2000: Precipitable water observed by ground-based GPS receivers and microwave radiometry. Earth Planets Space, 52 , 445–450.
Mätzler, C., , Martin L. , , Guerova G. , , and Ingold T. , 2002: Assessment of integrated-water-vapor data at Bern from GPS, sun photometry, microwave radiometry and radiosonde. Proc. Second Workshop of COST Action 716, Postdam, Germany, GeoForschungsZentrum Potsdam (GFZ).
Miloshevich, L. M., , Vömel H. , , Whiteman D. N. , , Lesht B. M. , , Schmidlin F. J. , , and Russo F. , 2006: Absolute accuracy of water vapor measurements from six operational radiosonde types launched during AWEX-G and implications for AIRS validation. J. Geophys. Res., 111 .D09S10, doi:10.1029/2005JD006083.
Minschwaner, K., , and Dessler A. E. , 2004: Water vapor feedback in the tropical upper troposphere: Model results and observations. J. Climate, 17 , 1272–1282.
Niell, A. E., , Coster A. J. , , Solheim F. S. , , Mendes V. B. , , Toor P. C. , , Langley R. B. , , and Upham C. A. , 2001: Comparison of measurements of atmospheric wet delay by radiosonde, water vapor radiometer, GPS, and VLBI. J. Atmos. Oceanic Technol., 18 , 830–850.
Press, W. H., , Flannery B. P. , , Teukolsky S. A. , , and Vetterling W. T. , 1986: Numerical Recipes: The Art of Scientific Computing. Cambridge University Press, 678 pp.
Pugnaghi, S., , Boccolari M. , , Fazlagic S. , , Pacione R. , , Santangelo R. , , Vedel H. , , and Vespe F. , 2002: Comparison of independent integrated water vapor estimates from GPS and Sun photometer measurements and a meteorological model. Phys. Chem. Earth, 27 , 355–362.
Reagan, J. A., , Pilewskie P. A. , , Scott-Fleming I. C. , , Herman B. M. , , and Ben David A. , 1987: Extrapolation of Earth-based solar irradiance measurements to exo-atmospheric levels for broad bands and selected absorption bands observations. IEEE Geosci. Remote Sens., 25 , 647–653.
Rothman, L. S., and Coauthors, 2003: The HITRAN molecular spectroscopic database: edition of 2000 including updates through 2001. J. Quant. Spectrosc. Radiat. Transfer, 82 , 5–44.
Schmid, B., , Thome K. J. , , Demoulin P. , , Peter R. , , Mätzler C. , , and Sekler J. , 1996: Comparison of modeled and empirical approaches for retrieving columnar water vapor from solar transmittance measurements in 0.94 μm region. J. Geophys. Res., 101 , D5. 9345–9358.
Schmid, B., and Coauthors, 2001: Comparison of columnar water-vapor measurements from solar transmittance methods. Appl. Opt., 40 , 1886–1896.
Smirnov, A., , Holben B. N. , , Eck T. F. , , Dubovick O. , , and Slutsker I. , 2000: Cloud screening and quality control algorithms for the AERONET data base. Remote Sens. Environ., 73 , 337–349.
Smirnov, A., , Holben B. N. , , Lyapustin A. , , Slutsker I. , , and Eck T. F. , 2004: AERONET processing algorithms refinement. Proc. AERONET Workshop, El Arenosillo, Spain, NASA/GSFC Aeronet Project.
Soden, B. J., , Wetherald R. T. , , Stenchikov G. L. , , and Robock A. , 2002: Global cooling after the eruption of Mount Pinatubo: A test of climate feedback by water vapor. Science, 296 , 727–730.
Soden, B. J., , Turner D. D. , , Lesht B. M. , , and Miloshevich L. M. , 2004: An analysis of satellite, radiosonde, and lidar observations of upper tropospheric water vapor from the Atmospheric Radiation Measurement Program. J. Geophys. Res., 109 .D04105, doi:10/1029/2003JD003828.
Vermote, E., , and Kaufman Y. K. , 1995: Absolute calibration of AVHRR visible and near-infrared channels using ocean and cloud views. Int. J. Remote Sens., 13 , 2317–2340.
Volz, F. E., 1974: Economical multispectral sun radiometer for measurement of aerosol extinction from 0.44 μm to 1.6 μm and precipitable water. Appl. Opt., 13 , 1732–1733.
Nominal spectral response of the 940-nm CIMEL sun radiometer (dotted line) superimposed on the water vapor transmittance simulated with MODTRAN in midlatitude summer conditions (continuous line). The 940-nm narrow-band filter has a spectral bandwidth [full width at half maximum (FWHM)] of 10 nm.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Variation of the 940-nm atmospheric transmittance calculated using the MODTRAN radiative transfer code as a function of observed slant water vapor (air mass times IWV). The IWV values used to carry out this simulation were retrieved from GPS meteorology measurements at the Sherbrooke site (45°22′N, 71°55′W). The spectral response employed to generate these points was the nominal (passband) curve of Fig. 1.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Influence of the 940-nm filter spectral shift on the a and b parameters. These values were derived using the MODTRAN (version 3.7) radiative transfer code to generate transmissions for an ensemble of IWV and m values.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Time series of IWV records from both GPS meteorology and (AERONET V1) sun radiometry for three Canadian AEROCAN sites that undergo different climatic regimes: (a) Churchill (58°45′N, 94°5′W), (b) Saturna Island (48°46′N, 123°07′W), and (c) Sherbrooke (45°22′N, 71°55′W) (see Table 1 for the measurement period).
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Example of the (A, a, b) retrieval procedure [Eq. (5)] in the case of 2002 Saturna Island observations. The points represent the aerosol and Rayleigh corrected transmission for the sun radiometer 940-nm channel [the Y term from Eq. (6)] as a function of the slant water vapor content.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
(a) Uncertainty in the IWV retrieval as a function of the relative difference between the predicted and observed IWV values in the 940-nm band for the Saturna Island site (48°46′N, 123°07′W). The same plot profile is observed for the others considered sites. A 10% IWV uncertainty (accuracy of water vapor determination from radiosonde) corresponds to a 1% relative deviation in the CIMEL sun radiometer logarithm signal prediction. (b) Variation of the Y term [from Eq. (6)] with the slant water vapor content shown in Fig. 5, and corrected for outlier radiometric points (points greater than 10% IWV or 1% deviation threshold have been excluded).
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Comparison between IWV-GPS and IWV-SUN (AERONET V1 calibration; rhombus symbols), IWV-GPS and IWV-SUN (AERONET V2 calibration; plus symbol), and IWV-GPS and IWV-SUN (GPS calibration; gray star symbols) for the (a) Saturna Island and (b) Sherbrooke sites. The statistical results of the regressions are reported in Table 7.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Day-to-day variation of calibration constants (a) a, (b) b, and (c) and the associated daily number of observations. (d) For the Saturna Island (48°46′N, 123°07′W) site during the year 2002.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
Plot of Y as a function of m × IWV corresponding to the single set of a, b, and A values extracted from the overall observations of the sun radiometer 940-nm channel for the Saturna Island (48°46′N, 123°07′W) site during the year 2002.
Citation: Journal of Atmospheric and Oceanic Technology 24, 6; 10.1175/JTECH2011.1
AEROCAN/AERONET sun radiometer sites, their corresponding GPS receiver locations, and the period of simultaneous GPS and sun radiometer observations.
Influence of variations in the filter spectral response (variable random noise) on the a and b coefficients. These values were generated using MODTRAN transmission outputs convolved with the nominal filter response of Fig. 1.
Comparison between GPS- and the sun radiometer–derived IWV (AERONET, V1) procedure (the level 1.5 and 2.0 labels refer only to the method of extrapolation of the AOD to 940 nm). The statistical parameters for the comparison are No. obs: number of observations, rmse: root-mean-square error between y and x, bias: bias error [mean (y − x)], and r: correlation coefficient. The slope and yo values are defined from the linear regression: y = slope × x + Yo, where y and x stand, respectively, for dependent (IWV-SUN) and independent (IWV-GPS) variables.
CIMEL sun radiometer calibration parameters derived from the modified Langley technique compared with AERONET V1 results. For the AERONET V1 protocol, Note that A is variable while the values of a and b are fixed at 0.61 and 0.59, respectively [see Eq. (7)]. For the GPS meteorology technique, A, a, and b are variable [see Eq. (5)]. Rmse corresponds to the root-mean-square error in the Y term resulting from the curve fitting procedure.
Sun radiometric calibration constants A, a, and b retrieved using GPS measurements from half of the database considered. The second half of the database is used to validate the approach (see Table 7).
Comparison between the AERONET calibration approach and the GPS-based calibration approach (see Table 3 for the statistical parameter meanings).
Values of a and b derived from the AERONET V2 procedure for several CIMEL sun radiometers and their associated sites.
