Product Validation Report

Transcription

1 European Space Agency GOME Evolution project Product Validation Report GOME Evolution Climate Product vs. NCAR GNSS GOME Evolution Climate Product vs. ARSA Version: Final version Date:

2 Issue: Revision: 0 Page 2 Document Change Record Document, Version DOC, draft DOC, draft DOC, draft DOC, Final version v1.0, v1.1, v2.0, Date Changes Originator Original version A. Danielczok, M. Schröder Feedback from DLR and MPI-C implemented Updated GOME Evolution climate product data version: v2.01 A. Danielczok, M. Schröder A. Danielczok, M. Schröder Final version A. Danielczok, M. Schröder

3 Issue: Revision: 0 Page 3 Table of Contents 1 Introduction Purpose Definitions, acronyms and abbreviations References Structure of the document Methods and reference data GOME Evolution Climate product of H 2 O GNSS ARSA Methodology GOME EVL climate product vs. GNSS General investigation of GOME-EVL climate product vs. GNSS Seasonal investigation Global distribution of bias and RMS Stability GOME EVL climate product vs. ARSA General investigation of GOME EVL climate product vs ARSA Seasonal investigation Global distribution of bias and RMS Stability Conclusions Acknowledgement... 20

4 Issue: Revision: 0 Page 4 1 Introduction 1.1 Purpose This document presents results of the comparison of the monthly Total Column Water Vapor (TCWV) climate product from the Global Ozone Monitoring Experiment Evolution (GOME-EVL) project and the Global Navigation Satellite System GNSS as well as radiosonde records of the Analysed Radio- Soundings Archive (ARSA, version 2.7) as reference. The data product is based on the measurements from GOME, SCIAMACHY and GOME-2 on-board ERS-2, ENVISAT and Metop-A. The investigation considers bias, RMS and stability as well as zonal and seasonal dependencies using data from June 1995 December 2014 in case of GOME Evolution and GNSS and June 1995 March 2015 in case of GOME Evolution and ARSA. 1.2 Definitions, acronyms and abbreviations ARSA Envisat ERS EUMETSAT GNSS GOME GOME-EVL Metop RMS SCIAMACHY TCWV Analysed Radio Soundings Archive Environmental Satellite European Remote Sensing European Organisation for the Exploitation of Meteorological Satellites Global Navigation Satellite System Global Ozone Monitoring Experiment GOME-Evolution Meteorological Operational Satellite Root Mean Square Error Scanning Imaging Absorption Spectrometer for Atmospheric Cartography Total Column Water Vapour 1.3 References Beirle et al., The MPIC/DLR Climate water vapor product:a consistent time-series of H2O columns from GOME/SCIAMACHY/GOME-2, to be submitted to ESSD, Grossi, M., Valks, P., Loyola, D., Aberle, B., Slijkhuis, S., Wagner, T., Beirle, S. and Lang, R.: Total column water vapour measurements from GOME-2 MetOp-A and MetOp-B, Atmos. Meas. Tech., 8(3), , doi: /amt , Scott, N., and the ARA/ABC(t)/LMD team: QUASAR - Quality Assessment of Satellite and Radiosonde data. CM SAF Visiting Scientist Report, CDOP-2 AVS Study 13_03, 18 December 2015, available at Wang, J., L. Zhang, A. Dai, T. Van Hove, and J. Van Baelen (2007), A near-global, 2-hourly data set of atmospheric precipitable water from ground-based GPS measurements, J. Geophys. Res., 112, D11107, doi: /2006jd Structure of the document An overview of the GOME Evolution Climate product, the GNSS and the ARSA data records and the methodology is given in section 2. Section 3 and 4 then present the comparison of the GOME Evolution data to both reference datasets. In section 5 conclusions are given.

5 Issue: Revision: 0 Page 5 2 Methods and reference data 2.1 GOME Evolution Climate product of H 2 O The MPI-C/DLR GOME Evolution Climate product of TCWV (in the following: GOME EVL climate product) is derived from the satellite instruments GOME on-board ERS-2, SCIAMACHY on Envisat and GOME-2 on Metop-A. Particular focus of the climate product is the consistency amongst the different sensors to avoid jumps from one instrument to another. This is reached by applying robust and simple retrieval settings (similar to the settings of the operational GOME-2 retrieval described in Grossi et al., 2015) consistently. Most of all, the systematic effects caused by the large differences in pixel size are avoided by merging SCIAMACHY and GOME-2 observations to GOME spatial resolution. In addition, the GOME-2 swath is reduced to that of GOME to have consistent viewing geometries. Remaining differences between the different sensors are quantified during overlap periods and are corrected for in the homogenized time series. For details of the climate product retrieval see Beirle et al., 2017.The GOME EVL climate product v1 was released on 20 December It provides time series of monthly mean TCWV on a 1 x1 grid, covering the time period Note that GOME/SCIAMACHY/GOME-2 measurements are always taken at 10:30/10:00/9:30 local time, respectively 2.2 GNSS The atmospheric delay of the signal sent by a satellite and received by a ground station is depending on pressure, temperature and total amount of water vapor, where the latter can be calculated by knowledge of pressure and temperature. Those were gained from synoptic observations as well as from the reanalysis of NCEP/NCAR. The GNSS data base version includes data from 1995 to In total, 997 stations are specified. More information and calculations of the GNSS data base can be found in Wang et al. (2007). A map of the global distribution of collocated stations is shown in section ARSA Here we focus on a brief description of the Analysed RadioSoundings Archive (ARSA, version 2.7, The ARSA database was produced at and provided by ARA/ABC(t)/LMD, Paris, France. ARSA is mainly based on radiosonde observations that have successfully completed extensive qualitative and quantitative tests: the required minimum information content per profile is to have valid measurements from surface up to 30 hpa for temperature profiles and from surface to 300 hpa for water vapor profiles. Moreover, in order to give a continuous description of the atmospheric state from the surface to the top of the atmosphere (~0.002 hpa), these radiosonde observations have been extended above their highest measured point with ERA-Interim data (temperature, water vapor and ozone up to 0.1 hpa) and then with SciSat ACE FTS level2 data (from 0.1 hpa to the top the atmosphere at hpa). In the upper troposphere ERA- Interim is included based on an empirical function and bias correction which ensures radiative consistency with IASI. During various validation efforts it could be shown that this consistency is also observed relative to HIRS and AMSU-B (Scott et al., 2014). Each ARSA observation contains 43 level-profiles of temperature, water vapor mass mixing ratio and ozone mass mixing ratio. In some cases, the radiosounding profiles needed to be completed with ozone profiles or were extrapolated. More information and calculations of the ARSA data base can be found in the ARSA README (available at A map of the global distribution and their respective number of available measurements is displayed in figure 2-1.

6 Issue: Revision: 0 Page 6 Figure 2-1: Number of profiles at ARSA stations. 2.4 Methodology We compare monthly mean TCWV data of GOME EVL climate product with TCWV data of NCAR GNSS and ARSA. The GNSS data record is based on perceptible water in mm what can be directly converted into Total Column Water Vapor in kg/m². The ARSA data record is based on water vapor mass mixing ratio (µ) profiles. Specific humidity is calculated by: sh = µ µ + 1 TCWV can then be calculated by integrating the profile from surface to top of the atmosphere: (1) TCWV = 1 g!_!"#!_!"#$%&' sh dp (2) where g = 9.81 m/s² is the gravitational acceleration. Since the data of GOME EVL climate product was given as monthly means, GNSS as well as ARSA data were averaged on a monthly basis, using the arithmetic mean of all valid values of a month for each station:! TCWV!"#$!!"_!"#$ = TCWV! n with n being the number of valid observations. GNSS and ARSA monthly means thus represent averages of the full diurnal cycle. To take the difference in temporal sampling into account, a separate evaluation is done for ARSA and GNSS monthly mean TCWV based on measurements done at 10:00 ± 1 hour local time (in the following: 10:00 mean TCWV). The collocation criterion is 100 km as often found in literature. The nearest neighbor is used if several collocations were found at the same GNSS/ARSA station or GOME Evolution. Only stations providing at least 20 collocations were taken into account. GOME Evolution data with warning flag = 1, meaning that the mean GOME1 scan angle systematically deviates from 0 (Beirle et al., 2017), was declined for all collocations. This applies mostly to regions in and between northern India and Kazakhstan, where GOME calibration measurements were performed. In addition to the consideration of (3)

7 Issue: Revision: 0 Page 7 all collocations 8 GNSS stations in at islands or costal stations in high altitudes are declined (cf. sec. 3). The stability analysis is based on linear least-square regressions (see Biases as well as RMS are provided and discussed as well.

8 Issue: Revision: 0 Page 8 3 GOME EVL climate product vs. GNSS The number of collocations of GNSS and GOME-EVL climate product is displayed in figure 3-1. The northern hemisphere includes most of the stations, in particular North America and Europe. Some mountain stations at coasts and islands above 1000 m are excluded (e.g. Hawaii: MAUI, 3045 m; MKEA, 3729 m; KOKB, 1152 m; KOKV, 1152 m; in total 8 stations). The total number of collocations is Figure 3-1: Global distribution of the number of collocations for each GNSS station. Taking the 10:00 mean TCWV reduces the number of collocations only slightly because GNSS data is provided every 2 hours for most stations. The total number is The global distribution of the number of collocations is similar to fig General investigation of GOME-EVL climate product vs. GNSS The GOME EVL climate product reveals reasonable agreement with GNSS data (fig. 3-2). The bias is kg/m² (rel. bias: -5.24%) while the RMS is 4.35 kg/m² (rel. RMS: 22.82%). As already mentioned above, some costal stations in high altitudes are excluded, which reveal a large positive bias. The median of the relative difference of GOME-EVL climate product and GNSS (fig. 3-2, right) is large for low GNSS TCWV, what is expected as the relative difference becomes larger with small reference values. After GNSS TCWV exceeds approximately 10 kg/m² the relative difference stays constant at a negative value. Figure 3-2: Left: Scatter plot of TCWV of all GNSS Stations and GOME EVL climate product; Right: Median of relative difference of GOME-EVL climate product and GNSS TCWV including 5%, 25%, 75% and 95% percentile vs. GNSS TCWV.

9 Issue: Revision: 0 Page 9 Taking the 10:00 mean TCWV as reference shifts the bias closer to 0 kg/m². Iit is kg/m² (rel. bias: -3.93%). Relative and absolute RMS are close to RMS of total mean TCWV (abs. RMS: 4.34 kg/², rel. RMS: 23.13%). 3.2 Seasonal investigation The seasonal investigation is carried out for the comparison of GOME EVL climate product and GNSS. Associated results are displayed in figure 3-3. The absolute largest bias and RMS occur in summer with a bias of kg/m² (rel. bias: -5.69%) and RMS of 5.55 kg/m² (rel. RMS: 20.35%). The smallest bias and RMS are found in winter. Thus, RMS and bias exhibit a seasonal cycle with maximum/minimum values in summer/winter. Because of larger variability associated with larger TCWV values in combination with the annual cycle of TCWV, this is expected. Figure 3-3: Scatter plots of TCWV of GOME EVL climate product and GNSS (station altitude <500 m) in case of spring (upper left), summer (upper right), fall (lower left) and winter (lower right).

10 Issue: Revision: 0 Page Global distribution of bias and RMS The global distributions of bias and RMS are displayed in figures 3-4 and 3-5 in absolute and relative values, respectively. The values of relative bias appear mostly random. The absolute bias reveals a zonal dependence. It tends to larger negative and positive values in the tropics compared to higher latitudes. As the TCWV is larger in the tropics, the increase of absolute bias is reasonable. The bias tends to positive values over the tropical ocean, whereas it is negative at some coastal stations in low latitudes. Figure 3-4: Global distribution of the bias between GOME EVL climate product using GNSS as truth, in absolute (left) and relative (right) values. The absolute RMS also exhibits a zonal dependence with larger RMS in the tropics, caused by the effect of larger variability of TCWV. Relative RMS appears mostly random. Neglecting some large relative RMS values at coastal stations in low latitudes, relative RMS appears to be larger in high latitudes, where TCWV is lower. Especially at coastal stations in the low latitudes the values of absolute bias, relative and absolute RMS are large. Figure 3-5: Global distribution of RMS of GOME EVL climate product and GNSS, in absolute (left) and relative (right) values.

11 Issue: Revision: 0 Page Stability The time series of monthly mean bias over 19 years ( ), averaged over all stations is displayed in figure 3-6, left. A negative stability (-1.77 ± 0.37 %/decade) is observed. The p-value is < 0.01, leading to the interpretation that the stability is significantly different from 0 %/decade. Figure 3-6: Left: Relative differences of GNSS and GOME EVL climate product as a function of time; Right: Time series of number of collocations per month: per latitude (upper right) and in total numbers (lower right). Note that for reliable conclusions on the stability the number of collocations (figure 3-7, right) has to be taken into account. In case of the GNSS comparison, the number of collocations increases continuously between 1995 (approx. 50 collocations/month) and 2011 (approx. 500 collocations/month). Between 2011 and 2014 the number of collocations decreases to approximately 300 collocations per month. Because of the low number of collocations at the beginning of the time series the accuracy of the given stability and p-value is questionable. Rejecting stations that were not available over the whole time series lowers the number of collocations to In total 45 stations are left with a constant number of ~40 collocations per month. The resulting stability is displayed in figure 3-7. It is negative (-1.69 ± 0.39 %/decade) and significantly different from 0%/decade (p < 0.01). Compared to the stability analysis done with all stations included, the variation is larger but the time series is more continuous. It seems that this number might be too low for drawing conclusions. Figure 3-7: Left: Relative differences between GOME EVL climate product and data from GNSS stations that were available during the whole time series as a function of time; Right: Time series of number of collocations per month: per latitude (upper right) and in total numbers (lower right).

12 Issue: Revision: 0 Page 12 Taking the 10:00 mean TCWV as reference with stations included, that were available during the whole time series, and investigating stability (fig. 3-8) and its significance gives a stability significantly different from 0 %/decade (stability: ± 0.40 %/decade; p = 0.03). The stability shifts slightly closer to 0 %/decade when taking the 10:00 mean TCWV compared to the total daily mean. However, the variability is large and the number of collocations is too low to draw conclusions. Figure 3-8: Relative differences between data from GNSS (10:00 mean TCWV) and GOME EVL climate product as a function of time, stations available during whole time series. As the median might be more representative compared to the arithmetic mean, another stability analysis based on the monthly median values was done (fig. 3-9). In both cases, the whole daily TCWV mean and the 10:00 mean TCWV, the stability is slightly negative and significantly different from 0%/decade. For the 10:00 mean TCWV (fig. 3-9, right) the stability again is closer to 0%/decade. Figure 3-9: Relative differences between median of GOME EVL climate product and GNSS as a function of time; left: whole daily mean of stations available during the whole time series; right: only stations of the 10:00 mean TCWV included that were available during the whole time series.

13 Issue: Revision: 0 Page 13 4 GOME EVL climate product vs. ARSA The number of collocations between ARSA and GOME EVL climate product is displayed in figure 4-1 (left). As found in the comparison of GOME EVL climate product and GNSS, the northern hemisphere includes most of the stations, here the number of collocations is larger, with a total number of collocations of Figure 4-1: Global distribution of the number of collocations for each ARSA station. Left: including all data for ARSA monthly mean TCWV; Right: ARSA 10:00 mean TCWV. Most of the ARSA data is provided for 0:00 and 12:00 UTC. Calculating the 10:00 mean TCWV for all ARSA stations reduces the total number of collocations to Some areas (e.g. India) are not included anymore (fig. 4-1, right). 4.1 General investigation of GOME EVL climate product vs ARSA The comparison of GOME EVL climate product and ARSA exhibit a bias of kg/m² (rel. bias: -8.32%) and RMS of 5.59 kg/m² (rel. RMS: 24.38%). Compared to the GNSS comparison, bias and RMS are larger, but the bias tends to the same (negative) sign. Figure 4-2: Left: Scatter plot of TCWV from GOME EVL climate product vs. GNSS; Right: median of relative difference of GOME EVL climate product and ARSA including 5%, 25%, 75% and 95% percentile vs. GNSS TCWV.

14 Issue: Revision: 0 Page 14 The median of the relative difference of GOME EVL climate product and ARSA (fig. 4-2, right) appears to be similar to the GOME-GNSS-comparison. Here the relative difference reaches a constant value at ARSA TCWV of approximately 5 kg/m². In contrast to the GNSS comparison, ARSA TCWV includes larger values, up to ~75 kg/m², whereas GNSS reaches a maximum at approximately 60 kg/m². With TCWV above 60 kg/m² the median decreases distinctly. The number of collocations has to be taken into account. It is low at these high TCWV values (> 60 kg/m²: ~950 collocations, 0.6% of total number; > 65 kg/m²: ~300 collocations, 0.2% of total number) and thus results at TCWV values larger than ~60 kg/m 2 need to be considered with care. Taking the ARSA 10:00 mean TCWV as reference shifts the bias to a small positive value (0.19 kg/m²; rel. bias: 0.82 %). Because of the lower number of available collocations, the absolute RMS is larger compared to the examination using all ARSA data (6.06 kg/m²; rel. RMS: %). It has to be mentioned that some areas with large negative bias (e.g. Southeast Asia) are not included in the 10:00 mean TCWV calculation. 4.2 Seasonal investigation The seasonal investigation of GOME EVL climate product and ARSA is displayed in figure 4-3. As already found in the comparison to GNSS, RMS is largest in summer and smallest in winter caused by the same effect of larger variability. The bias is largest in spring and in summer. It also reaches a minimum in winter. Figure 4-3: Scatter plots of TCWV from GOME EVL climate product and ARSA in case of spring (upper left), summer (upper right), fall (lower left) and winter (lower right).

15 Issue: Revision: 0 Page Global distribution of bias and RMS The global distributions of absolute and relative bias and RMS are displayed in figures 4-4 and 4-5. A zonal dependence of absolute and relative bias appears as already found in the comparison with GNSS. The absolute bias tends to larger (negative) values in the tropics. Here the Total Column Water Vapor is larger. Remarkable are the large negative values of absolute bias at the Indian subcontinent and Southeast Asia. Other high negative absolute biases appear mostly in the tropics at coastal stations. In the GNSS comparison this effect is less obvious because of the low number of available stations in the aforementioned regions (cf. fig. 3-4). But the few available GNSS collocations support the results of the comparison with ARSA. The same applies to the absolute RMS. In general the absolute and relative bias tends to negative values at most stations, except for stations in the tropical ocean, where the bias tends to positive values. Figure 4-4: Global distribution of bias of GOME EVL climate product and ARSA comparison, in absolute (left) and relative (right) values. The absolute and relative RMS also reveal a zonal dependence, caused by the effect of larger variability and high absolute TCWV values in the tropics. The Indian Subcontinent and Southeast Asia as well as some other equatorial coastal stations stick out with large absolute RMS values. In addition to the larger relative RMS in higher latitudes, the relative RMS appears to be larger at some coastal stations. Figure 4-5: Global distribution of RMS of GOME EVL climate product and ARSA, in absolute (left) and relative (right) values.

16 Issue: Revision: 0 Page Stability The time series of the monthly mean bias, averaged over all stations is displayed in figure 4-6. A slight positive stability (0.89 ± 0.23 %/decade) is observed. The stability is, probating the stability found in the GNSS comparison, significantly different from 0%/decade (p < 0.01). The number of collocations (fig. 4-6, right) is nearly constant over time. In contrast to the GNSS comparison, the number of collocations is larger, also at the beginning of the time series in Figure 4-6: Left: Relative differences of GOME EVL climate product and ARSA as a function of time; Right: Time series of number of collocations per month: per latitude (upper right) and in total numbers (lower right). The stability analysis with ARSA 10:00 mean TCWV and data from all stations (fig. 4-7, left) reveals a negative stability (-1.79 ± 0.43 %/decade), not significantly different from 0 %/decade (p > 0.05), what is contradictory to the previous analysis. It has to be mentioned that the p-value is close to The variability is larger and the number of collocations per month is Figure 4-7: Left: Relative differences between GOME EVL climate product and ARSA (10:00 mean TCWV) as a function of time; left: all stations included; right: only stations included that were available during the whole time series. The stability analysis was also done after rejecting stations that were not available during the whole time series (fig. 4-7, right). The number of collocations reduces to approximately 65 per month. Here, the stability is slightly larger. It is significantly different from 0 %/decade (p = 0.03).

17 Issue: Revision: 0 Page 17 Figure 4-8: Relative differences between median of GOME EVL climate product and ARSA as a function of time; left: all stations included; right: only stations of the 10:00 mean TCWV included that were available during the whole time series. Another stability analysis was done on the basis of median instead of arithmetic mean (fig. 4-8). Here the stability is not significantly different from 0%/decade using all stations (fig. 4-8, left). The stability based on the median of the 10:00 am mean and stations, that were available during the whole time series (fig 4-8, right) again reveals a negative stability (-1.17 ± 0.46 %/decade) significantly different from 0%/decade (p-value = 0.01). Here the low number of collocations (about 65 per month) and the large variability have to be taken into account.

18 Issue: Revision: 0 Page 18 5 Conclusions This report provides a comparison of the GOME Evolution Climate product monthly mean TCWV data with GNSS and ARSA monthly averaged TCWV as references. The comparison with GNSS exhibits a negative stability in all considered cases (cf. tab. 5-1), significantly different from 0%/decade. According to the number of collocations at the beginning of the time series the accuracy of the stability using all stations is questionable. Considering stations that were not available during the whole time series and recomputing stability also implies a negative stability significantly different from 0 %/decade, in the latter case the number of collocations is too low to draw conclusions. The bias appears to be kg/m² while the RMS is 4.35 kg/m². The seasonal investigation exposes that the absolute bias and RMS are largest/smallest in summer/winter. The global distribution of bias and RMS appears to have a slight zonal dependence caused by zonal variability of TCWV. Coastal stations in low latitudes stick out with larger biases and RMS. Results related to the use of ARSA data as reference support the most features found in the comparison with GNSS, except for the negative stability. Using the total mean reveals a slightly positive stability significantly different from 0%/decade. Bias and RMS are larger compared to GNSS (Bias: kg/m²; RMS: 5.60 kg/m²). Table 5-1: Overview of stability and p-value for all investigated cases Reference Data Condition Stability [%/decade] p-value Number of collocations per month GNSS Total mean, all stations ± GNSS Total mean, stations full time avail ± GNSS 10:00 mean, all stations ± GNSS 10:00 mean, stations full time available ± GNSS Median total, all stations ± GNSS Median total, stations full time available ± GNSS Median 10:00, all stations ± GNSS Median 10:00, stations full time available ± ARSA Total mean, all stations 0.89 ± ARSA Total mean, stations full time avail 0.46 ± ARSA 10:00 mean, all stations ± ARSA 10:00 mean, stations full time available ± ARSA Median total, all stations 0.41 ± ARSA Median total, stations full time available 0.06 ± ARSA Median 10:00, all stations ± ARSA Median 10:00, stations full time available ± Conspicuous are high values of absolute bias and RMS at the Indian subcontinent and Southeast Asia as well as other coastal stations in low latitudes. The number of GNSS station in this region is too low

19 Issue: Revision: 0 Page 19 for concrete conclusions, but the few available stations support the results of the ARSA comparison with large biases and RMS. The comparison to both reference data sets reveals a negative bias. The overall results are summarized in table 5-2. Table 5-2: Overview of bias and RMS for all investigated based on total mean calculation / 10 am ± 1 hour mean calculation Reference Data Season Bias [kg/m²] Relative Bias [%] RMS [kg/m²] Relative RMS [%] Number of Collocations GNSS All / / / / / GNSS Spring / / / / / GNSS Summer / / / / / GNSS Fall / / / / / GNSS Winter / / / / / ARSA All / / / / / ARSA Spring / / / / / ARSA Summer / / / / / ARSA Fall / / / / / ARSA Winter / / / / / Calculating the GNSS monthly mean on the basis of measurements taken at 10:00 ± 1 hour local time reveals that absolute and relative bias shift slightly closer to 0 kg/m² and 0 %, respectively. The stability found for the 10:00 mean calculation is negative and significantly different from 0%/decade (cf. tab. 5-1). Results related to ARSA 10:00 mean TCWV as reference reveal a bias closer to 0 kg/m 2 compared to the previous collocations with all measurement times included. Here, the overall bias is positive (0.19 kg/m²), related to a large positive bias in summer. Absolute and relative RMS are larger compared to the total mean because of a lower total number of available collocations. It has to be taken into account that stations with large negative biases (India, Southeast Asia) are not included in this comparison and that the global distribution of available collocations is not as reasonable as for the ARSA total mean TCWV. Therefore bias and RMS have to be taken with care in the case of ARSA 10:00 mean TCWV. The stability found using data from ARSA with all stations included reveals stability not significantly different from 0%/decade. Rejecting stations that were not available during the whole time series contradicts the GNSS analysis with a negative stability significantly different from 0%/decade. The comparison based on monthly mean TCWV containing only measurements taken at 10 am ± 1 hour reveals a bias slightly closer to 0 kg/m² for both reference data sets. The results for the ARSA comparison have to be taken with care because of the global distribution of number of collocations. Overall, the differences found are moderate and are mainly related to the climate retrieval, i.e. simplified AMFs, reduced spatial resolution, and cloud masking based on O 2 absorption. Relative to GNSS sound conclusions on the stability can hardly be drawn. Applying the strict set of collocation criteria to ARSA leads to a stability of ± 0.42 %/decade, which is significantly different from 0%/decade. Note that in this case the noise level is large and that ARSA data has not been homogenized in time.

20 Issue: Revision: 0 Page 20 6 Acknowledgement The ARA/ABC(t)/LMD group and as well as NCAR/UCAR/EOL are acknowledged for producing and making available the ARSA and GNSS data, respectively.