21
CDR Program FCDR (SSM/I) Technical Report SSM/I F15 RADCAL Correction for Ocean Data CSU Technical Report Karen Milberger and Wesley Berg March 2012 http://rain.atmos.colostate.edu/FCDR/

SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

  • Upload
    others

  • View
    8

  • Download
    0

Embed Size (px)

Citation preview

Page 1: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report

SSM/I F15 RADCAL Correction for Ocean Data

CSU Technical Report

Karen Milberger and Wesley Berg

March 2012

http://rain.atmos.colostate.edu/FCDR/

Page 2: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 1

TABLE of CONTENTS

1.  INTRODUCTION ...................................................................................................... 4 

2.  FINDING A PREDICTION FOR THE CORRECTION ............................................... 4 

3.  CONSIDERING HOTLOAD IN THE CORRECTION ................................................ 8 

4.  RESULT OF THIS CORRECTION ......................................................................... 10 

5.  ERROR ESTIMATE FOR CORRECTED TB22V .................................................... 13 

6.  TB22V RETRIEVALS OVER LAND ........................................................................ 18 

7.  REFERENCES ....................................................................................................... 19 

Page 3: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 2

LIST of FIGURES

Figure 1: Timeseries of F15 daily mean TB22V over ocean ......................................................... 4 

Figure 2: Predicted vs. Observed TB22V (4-channel regression). ............................................... 5 

Figure 3: Predicated vs. Observed TB22V (6-channel regression). ............................................. 6 

Figure 4: Mean (black) and standard deviation (orange error bars) of TB22V-TB22Vpred(6) for 2005 ocean data. .......................................................................................................................... 7 

Figure 5: Mean (black) and standard deviation (orange error bars) of TB22V - TB22Vpred(6) for 2007 ocean data. .......................................................................................................................... 8 

Figure 6: Time series of delta TB22V and hotload temperature for 2007-2011 ocean data. .......... 9 

Figure 7: Hotload correction factor (black) with quadratic best-fit (red) and number of observations (blue) for 2007-2011 ocean data. .......................................................................... 10 

Figure 8: Timeseries of observed TB22V (blue) and RADCAL-corrected TB22V (purple) with hotload (red) for ocean data 80S-80N, where 4 cool events (2 in 2008, 2 in 2010) are identified in light gray and 2 cold events (in 2009) in dark gray. RADCAL activation in August 2006 is shown with a black line. .............................................................................................................. 11 

Figure 9: Timeseries of F15 mean TB22V over ocean before (top) and after (bottom) the RADCAL correction. Daily mean values are indicated by the black dots with green lines showing the 30-day running mean value. Red dots indicate daily mean values with poor sampling (less than 75% of the typical number of samples). The vertical black line in August 2006 denotes when the RADCAL beacons were turned on. ............................................................................. 12 

Figure 10: Timeseries of F15 mean TB22V over tropical ocean before (top) and after (bottom) the RADCAL correction. .............................................................................................................. 13 

Figure 11: Timeseries of the mean daily RADCAL adjustment r·s (blue) with the estimated error on the adjustment shown as the distance to the light blue lines, together with the Hotload (red) and the resulting corrected TB22V (green) also shown with the estimated error. These images show how both the adjustment and the error increase during cold events. ................................ 17 

Figure 12: Comparison of Mean (TB22V - TB22Vpred) for 2007 Data Over Ocean (blue) and Land (orange). ............................................................................................................................ 18 

Figure 13: Timeseries of F15 mean global TB22V before (top) and after (bottom) the correction for the RADCAL beacon interference, including all data over land and ocean together. ............ 19 

Page 4: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 3

LIST of TABLES

Table 1: Standard deviations and expected values of functions used to estimate error ............. 14 

Table 2: Error estimates for each hotload bin ............................................................................. 15 

ACRONYMS AND ABBREVIATIONS

Acronym or Abbreviation

Meaning

CATBD Climate Algorithm Theoretical Basis Document

CDR Climate Data Record

NCDC National Climatic Data Center

NOAA National Oceanic and Atmospheres Administration

RSS Remote Sensing Systems

TB Brightness Temperature

Page 5: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 4

1. Introduction

The activation of the radar calibration (RADCAL) suite on DMSP spacecraft F15 in August 2006 caused significant contamination of the 22.235 GHz vertical polarization (22V) channel of the SSM/I instrument. Our attempt to remove this contamination follows the approach of Remote Sensing Systems (RSS) described in Hilburn (2009) and Hilburn and Wentz (2008), explains where we deviate from it, and estimates uncertainties of the result. Since there is no theoretical estimate of the amount of contamination, the only way to quantify the impact is by comparing the data recorded before and after RADCAL activation. We first consider the 22V brightness temperature (TB22V) retrievals over ocean.

2. Finding a Prediction for the Correction

The timeseries of daily mean TB22V over ocean is shown in Figure 1. Not only is there a large jump (> 10 K) at the time of RADCAL activation, but as noted by Hilburn and Wentz (2008), there are significant time-dependent changes, most notably in 2009. Following Hilburn and Wentz (2008), we developed a linear regression model to predict TB22V based on the other low resolution channels (TB19V, TB19H, TB37V, TB37H). We used data prior to the RADCAL activation from 2005 and screened the data based on the following the conditions:

Ocean only Quality flag indicates no issues | Latitude | < 60 No rain in field of view (based on the flags developed by Stogryn et al., 1994)

Figure 1: Timeseries of F15 daily mean TB22V over ocean

Page 6: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 5

The condition used to identify rain over ocean is from Stogryn et al. (1994):

((TB37V – TB37H) < 50) or (TB19V > TB37V) or (TB19H > 185) or (TB37H > 210)

Based on this data subset we get the following regression equation:

TB22Vpred(4) = 0.4096 TB19V + 1.0319 TB19H + 1.1913 TB37V – 0.9270 TB37H – 104.3902

Figure 2 shows a scatter density plot of the observed TB22V versus the predicted values from the 4-channel regression. As indicated by the figure, there is a significant curvature, or a residual scene temperature dependence between the observed and predicted values.

Figure 2: Predicted vs. Observed TB22V (4-channel regression).

As a result, any correction using this prediction will lead to significant biases between warm scenes (land) and cold scenes (ocean). To eliminate the scene temperature dependence in the predicted TB22V estimates, we tried including the 85 GHz channels, which are more sensitive to changes in water vapor than the 19 and 37 GHz channels. This resulted in the following six channel regression model:

TB22Vpred(6) = 0.1444 TB19V + 1.1013 TB19H + 0.6625 TB37V – 1.0975 TB37H + 0.0062 TB85V + 0.3245 TB85H + 3.5265

(1)

Figure 3 shows the scatter density plot of the observed TB22V versus the predicted values from the 6-channel regression, which includes the 85 GHz channels. Although

Page 7: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 6

not perfect, the scene temperature dependence has been largely eliminated and the residual biases are within a few Kelvin of the observed values.

Figure 3: Predicated vs. Observed TB22V (6-channel regression).

A plot of the mean and standard deviation of TB22V – TB22Vpred(6) as a function of scan position is shown in Figure 4, which appears as expected with mean values near zero across the scan. As indicated in Figure 3, however, there is a fair amount of scatter. The standard deviations, indicated by the orange error bars in Figure 4, are around 2K, due to the fact that there is significant information in the 22 GHz channel than can’t be captured on a single pixel basis.

Page 8: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 7

Figure 4: Mean (black) and standard deviation (orange error bars) of TB22V-TB22Vpred(6) for 2005 ocean data.

However, as indicated in Figures 3 and 4, the 6-channel regression retrieval does a good job predicting the mean values across the scan and over a wide range of scene temperatures. Given the improvement over the 4-channel results, the subsequent analysis was done using the 6-channel regression shown above. It should be noted that this differs from the 4-channel prediction used by Hilburn and Wentz (2008) and Hilburn (2009).

Applying the 6-channel prediction to the data from 2007 following the activation of the RADCAL beacons, we computed the mean and standard deviation of TB22V – TB22Vpred(6) for each scan position and get a similar result (Figure 5) to that shown in Hilburn (2009), although with slightly smaller standard deviations presumably due to using the 6-channel prediction.

Page 9: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 8

Figure 5: Mean (black) and standard deviation (orange error bars) of TB22V - TB22Vpred(6) for 2007 ocean data.

These differences can subsequently be applied to the observed TB22V to correct for the RADCAL interference, at least during 2007 when the effect was relatively consistent as shown in Figure 1. Unfortunately, as determined by Hilburn and Wentz (2008) and seen in Figure 1, the RADCAL impact changes over time and thus cannot be treated as a static correction. Following the approach developed by Hilburn and Wentz (2008), the differences shown in Figure 5 are subsequently used to describe/quantify the variation in the RADCAL bias across the scan. This variation across the scan is subsequently referred to as the function r and it is treated as a static (independent of time) correction, but is a function of the scan position (or cell). Thus:

r(cell) = mean difference for specified cell (i.e. scan position) as shown in Figure 5.

3. Considering Hotload in the Correction

As shown in the timeseries of TB22V in Figure 1, there are spikes in 2009 and 2010 suggesting a change in the RADCAL impact. Hilburn (2009) showed that these increases in the observed TB22V values are correlated with decreases in the temperature of the hotload. The dependence of TB22V – TB22Vpred(6) (or delta TB22V) on the hotload temperature is clearly evident in the time series of these two values, which is shown in Figure 6 (note that the right-hand axis for the hotload is reversed with colder values corresponding to increases in delta TB22V). Note that in the subsequent

Page 10: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 9

discussion a moderate decrease in the hotload temperature is referred to as a cool event and a large decrease is referred to as a cold event.

Figure 6: Time series of delta TB22V and hotload temperature for 2007-2011 ocean data.

To investigate the relation between delta TB22V and hotload temperature, we use the multiplicative model described in Hilburn (2009). Instead of the static correction TB22Vcorrected = TB22V – r(cell) as discussed above, the multiplicative model is modified to account for the relationship to the hotload temperature as follows:

TB22Vcorrected = TB22V – r(cell) · s(hotload) (2)

where s is a function of the hotload temperature. Since the hotload temperature changes with the temperature of the spacecraft, and thus changes with time, this attempts to capture the time-dependent changes in the RADCAL impact shown in Figure 1. Figure 7 shows the resulting adjustment factor s, which is calculated as (TB22V – TB22Vpred(6)) / r for each pixel over the 5-year period from 2007 through 2011 and then binned as a function of the hotload temperature using 1-Kelvin bins. The mean and standard deviation for the binned values of s are shown along with the number of observations in each hotload temperature bin (shown in blue) in Figure 7, which is similar to Figure 7 in Hilburn (2009).

Page 11: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 10

Figure 7: Hotload correction factor (black) with quadratic best-fit (red) and number of observations (blue) for 2007-2011 ocean data.

For the function s, Hilburn and Wentz (2008) used a quadratic best-fit to the factor (TB22V-TB22Vpred) / r for hotload ≤ 298 K and the value of 1 for hotload > 298 K. The function shown in red in figure 7 is the result of fitting a quadratic through the mean binned values (note that the resulting fit is slightly different if the individual observations are used rather than the mean binned values). Since the resulting quadratic fit doesn’t fit the mean binned values very well, particularly at the warm end, we decided to use the mean binned values of s in the subsequent RADCAL correction. Unlike Hilburn and Wentz (2008), we use the binned mean values for all hotload temperature including those > 298 K. For the multiplicative adjustment factor s, therefore, we use the mean binned values shown in Figure 7, or:

s(hotload) = binned mean adjustment factor shown in Figure 7.

4. Result of this Correction

Figure 8 shows the resulting time series of TB22V corrected for the RADCAL impact (purple) using the multiplicative correction described above. The original TB22V is plotted in blue with the hotload temperature in red.

Page 12: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 11

Figure 8: Timeseries of observed TB22V (blue) and RADCAL-corrected TB22V (purple) with hotload (red) for ocean data 80S-80N, where 4 cool events (2 in 2008, 2 in 2010) are

identified in light gray and 2 cold events (in 2009) in dark gray. RADCAL activation in August 2006 is shown with a black line.

Figures 9 and 10 show the impact of the RADCAL beacons on TB22V after August 13, 2006. Figure 9 shows the result for all data over the ocean and Figure 10 limits the coverage to tropical oceans between 40S and 40N latitude. As the top panel in each of these figures shows, the impact of the RADCAL beacons was largest in early and late 2009, which corresponds to the periods of low hotload temperature (cold events) as shown in Figure 6.

Page 13: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 12

Figure 9: Timeseries of F15 mean TB22V over ocean before (top) and after (bottom) the RADCAL correction. Daily mean values are indicated by the black dots with green lines showing the 30-day running mean value. Red dots indicate daily mean values with poor

sampling (less than 75% of the typical number of samples). The vertical black line in August 2006 denotes when the RADCAL beacons were turned on.

The correction applied to remove the impact of the RADCAL beacons is a substantial improvement from the uncorrected data. However, there does appear to be a residual signal, which is particularly noticeable in early 2009 in the tropical oceanic mean values. As shown in Figure 10, there is a residual increase in TB22V in the range of 2-4 K. It should be noted that this increase, as well as all of the mean TB22V values shown in Figures 9 and 10, are averaged across the scan. As shown in Figure 5, the impact of the RADCAL beacons is greatest around pixel position 25 with a significant dropoff towards the right edge of the scan. Regardless, as shown in Figure 10, there is a residual time-varying RADCAL signal after the correction has been applied. To understand more about this correction, we estimate the error on the corrected values in the next section.

Page 14: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 13

Figure 10: Timeseries of F15 mean TB22V over tropical ocean before (top) and after (bottom) the RADCAL correction.

5. Error Estimate for Corrected TB22V

For the correction TB22Vcorrected = TB22V – r(cell) · s(hotload), the error in TB22Vcorrected is estimated as the standard deviation of r(cell) · s(hotload) by using the formula for the variance of a product of independent variables r and s:

2 2 2 2 2 2 2·r s r s s r r s (3)

where σ2 is the variance and µ is the expected value. Since r is a function of cell and s is a function of hotload, we need to consider whether µr and σr are also functions of cell and whether µs and σs are also functions of hotload.

µr µr is the function of cell shown in Figure 5 which varies from about 5.3 at cell 64 to a maximum of 13.23 at cell 25, with a mean over all cells of around 10. The maximum value of 13.23 can be used to get an estimate of the largest error across the scan.

σr Since the standard deviations seen in Figures 4 and 5 are almost constant across different cells, we can use the constant value of 2.0 for σr.

µs µs is the function of hotload shown in Figure 7.

Page 15: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 14

σs Since the standard deviations shown as error bars in Figure 7 vary from about 0.2 to almost 0.4, i.e. some are almost double others, we need to account for this variation, so σs is a function of hotload. We use the same hotload bins as define the function s itself and for each bin, its standard deviation is σs.

Table 1: Standard deviations and expected values of functions used to estimate error

Using the four functions listed above, we calculate the value of 2·r s using Equation 3 for

each 1-K hotload bin over the range of observed hotload temperatures, which vary between 257 K and 311 K. The resulting values are given in the table below.

Hotload (K) µs σs 2·r s (K2) σr·s (K)

257.5 2.66 0.39 55.50 7.45

258.5 2.51 0.38 50.83 7.13

259.5 2.36 0.34 43.33 6.58

260.5 2.26 0.30 37.12 6.09

261.5 2.28 0.30 37.28 6.11

262.5 2.23 0.32 38.11 6.17

263.5 2.17 0.29 34.27 5.85

264.5 2.13 0.27 31.15 5.58

265.5 2.14 0.27 31.35 5.60

266.5 2.11 0.26 30.31 5.51

267.5 2.10 0.26 29.50 5.43

268.5 2.04 0.25 27.75 5.27

269.5 1.99 0.26 27.65 5.26

270.5 1.92 0.26 26.48 5.15

271.5 1.86 0.26 25.61 5.06

272.5 1.80 0.25 23.89 4.89

273.5 1.77 0.25 23.41 4.84

274.5 1.72 0.23 21.76 4.66

275.5 1.66 0.22 20.05 4.48

276.5 1.61 0.22 18.75 4.33

277.5 1.55 0.21 17.46 4.18

278.5 1.50 0.21 17.18 4.15

279.5 1.46 0.21 16.56 4.07

280.5 1.42 0.22 16.70 4.09

281.5 1.37 0.22 16.52 4.06

282.5 1.35 0.23 16.92 4.11

Page 16: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 15

283.5 1.32 0.23 16.75 4.09

284.5 1.27 0.23 15.82 3.98

285.5 1.22 0.22 14.95 3.87

286.5 1.18 0.22 14.55 3.81

287.5 1.15 0.22 14.03 3.75

288.5 1.12 0.22 13.49 3.67

289.5 1.08 0.22 13.02 3.61

290.5 1.08 0.21 12.93 3.60

291.5 1.07 0.21 12.73 3.57

292.5 1.05 0.21 12.53 3.54

293.5 1.03 0.21 12.34 3.51

294.5 1.01 0.21 12.00 3.46

295.5 1.00 0.21 11.89 3.45

296.5 1.00 0.21 11.73 3.42

297.5 1.00 0.21 11.72 3.42

298.5 1.02 0.22 12.60 3.55

299.5 1.01 0.21 11.69 3.42

300.5 0.97 0.20 11.17 3.34

301.5 0.98 0.21 11.52 3.39

302.5 0.99 0.21 11.83 3.44

303.5 1.00 0.21 12.09 3.48

304.5 0.97 0.21 11.86 3.44

305.5 0.91 0.21 10.88 3.30

306.5 0.88 0.20 10.50 3.24

307.5 0.89 0.21 10.79 3.28

308.5 0.88 0.18 9.00 3.00

309.5 0.90 0.19 9.95 3.15

310.5 0.92 0.19 10.14 3.18

Table 2: Error estimates for each hotload bin

A timeseries of the mean daily value of the correction r(cell) · s(hotload), together with the estimated error on this correction, σr·s, is shown in Figure 11. As shown in this figure, the mean error is typically around 3 K, although this increases during the cool and cold events due to the increased magnitude of the correction. With mean error estimates of ~3K and values increasing to well over 5 K during the 2009 cold events, the RADCAL corrected data is clearly not suitable for climate applications. Hilburn and Wentz (2008) came to the same conclusion stating that the RADCAL-corrected data may be suitable for many weather applications, but is not of suitable accuracy/stability and therefore should not be used in climate analyses. Correspondingly, the quality flag

Page 17: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 16

on the low resolution channels in the F15 SSM/I FCDR is set to a value of 13 after August 13, 2006, indicating that the RADCAL correction has been applied to the TB22V field and that the resulting data is not suitable for climate applications.

Page 18: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 17

Figure 11: Timeseries of the mean daily RADCAL adjustment r·s (blue) with the estimated error on the adjustment shown as the distance to the light blue lines, together with the Hotload (red) and the resulting corrected TB22V (green) also shown with the estimated error. These images

show how both the adjustment and the error increase during cold events.

Page 19: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 18

6. TB22V Retrievals Over Land

As noted previously the TB22V RADCAL correction was developed using data over tropical and midlatitude oceans that had been screened for rain. This was done to avoid highly variable scenes over land and ice, and/or containing rainfall to better isolate the RADCAL impact. We now consider whether the correction developed based on the screened ocean data can be applied over land scenes. To determine if this is the case we independently analyzed the data over land. As before, we limited the latitude coverage. Using the pre-RADCAL 2005 data that meets the conditions:

Land coverage only Quality flag indicates no issues | Latitude | < 60

the following linear regression model was derived to predict TB22V from the other channels:

TB22Vpred(land) = 0.6504 TB19V + 0.0785 TB19H + 0.2955 TB37V – 0.1400 TB37H - 0.0126 TB85V + 0.0743 TB85H + 16.9551

This regression was then used to determine the RADCAL impact over land using data from 2007 as before. Figure 12 shows that for the 2007 data, the mean differences at each scan position are very similar for the land-derived correction as to those derived from the screened ocean data.

Figure 12: Comparison of Mean (TB22V - TB22Vpred) for 2007 Data Over Ocean (blue) and Land (orange).

Page 20: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 19

Given that the differences between the land-derived and ocean-derived corrections shown in Figure 12 are much less than 1 K for all scan positions, and the errors on the corrected ocean retrievals found in the previous section were greater than 3 K, the RADCAL correction described previously was subsequently applied for all scenes. A time series of the original and RADCAL-corrected TB22V including land, ocean, and high latitudes is shown in Figure 13. While there do appear to be artifacts in the resulting time series, the mean values are well within the stated errors.

The F15 SSM/I FCDR after August 13, 2006 has the RADCAL correction applied to the TB22V values over all scenes and the associated quality flag is set to indicate that the resulting data is not suitable for climate applications.

Figure 13: Timeseries of F15 mean global TB22V before (top) and after (bottom) the correction for the RADCAL beacon interference, including all data over land and ocean

together.

7. References

Colton, M. C. and Poe, G. A. (1999), ‘Intersensor calibration of DMSP SSM/I’s: F-8 to F-14, 1987-1997’, IEEE Transactions on Geoscience and Remote Sensing 37(1), 418–439.

Page 21: SSM/I F15 RADCAL Correction for Ocean Datarain.atmos.colostate.edu/FCDR/Archive_Docs/fcdrtechreports/CSU_FCDR_radcal_correction...RADCAL interference, at least during 2007 when the

CDR Program FCDR (SSM/I) Technical Report  

Page 20

Hilburn, K. (2009), Including temperature effects in the F15 RADCAL correction, Technical Report 051209, Remote Sensing Systems, http://www.remss.com.

Hilburn, K. A. and Wentz, F. J. (2008), ‘Mitigating the impact of RADCAL beacon contamination on F15 SSM/I ocean retrievals’, Geophysical Research Letters 35.

Poe, G. A., Uliana, E., Gardiner, B. and vonRentzell, T. (2006), Mitigation of DMSP F-15 RADCAL Interference with SSM/I, Technical report, NRL, Monterey, CA. Code 7541.

Stogryn, A. P., Butler, C. T. and Bartolac, T. J. (1994), ‘Ocean surface wind retrievals from special sensor microwave imager data with neural networks’, Journal of Geophysical Research 99(C1), 981–984.