Applications of an AMSR-E RFI Detection and Correction Algorithm in 1-DVAR over

WU Ying(吴莹)and WENG Fuzhong(翁富忠)

1 Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration,Nanjing University of Information Science&Technology,Nanjing 210044,China

2 School of Atmospheric Physics,Nanjing University of Information Science&Technology,Nanjing 210044,China

3 NOAA/NESDIS/Center for Satellite Application and Research,College Park,MD 20742,USA

Applications of an AMSR-E RFI Detection and Correction Algorithm in 1-DVAR over Land

WU Ying1,2∗(吴莹)and WENG Fuzhong3(翁富忠)

1 Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration,Nanjing University of Information Science&Technology,Nanjing 210044,China

2 School of Atmospheric Physics,Nanjing University of Information Science&Technology,Nanjing 210044,China

3 NOAA/NESDIS/Center for Satellite Application and Research,College Park,MD 20742,USA

Land retrievals using passive microwave radiometers are sensitive to small fluctuations in land brightness temperatures.As such,the radio-frequency interference(RFI)signals emanating from man-made microwave radiation transmitters can result in large errors in land retrievals.RFI in C-and X-band channels can contaminate remotely sensed measurements,as experienced with the Advanced Microwave Scanning Radiometer (AMSR-E)and the WindSat sensor.In this work,applications of an RFI detection and correction algorithm in retrieving a comprehensive suite of geophysical parameters from AMSR-E measurements using the onedimensional variational retrieval(1-DVAR)method are described.The results indicate that the values of retrieved parameters,such as land skin temperature(LST),over these areas contaminated by RFI are much higher than those from the global data assimilation system(GDAS)products.The results also indicate that the differences between new retrievals and GDAS products are decreased evidently through taking into account the RFI correction algorithm.In addition,the convergence metric(χ2)of 1-DVAR is found to be a new method for identifying regions where land retrievals are affected by RFI.For example,in those regions with much stronger RFI,such as Europe and Japan,χ2of 1-DVAR is so large that convergence cannot be reached and retrieval results may not be reliable or cannot be obtained.Furthermore,χ2also decreases with the RFI-corrected algorithm for those regions with moderate or weak RFI.The results of RFI detected by χ2are almost consistent with those identified by the spectral difference method.

microwave remote sensing,radio-frequency interference(RFI),AMSR-E,1-DVAR

1.Introduction

Early examinations of passive microwave brightness temperature measurements showed evidence of extensive Radio-Frequency Interference(RFI)signals at low microwave frequencies(Li et al.,2004;Njoku et al.,2005;Kidd,2006).A number of passive microwave sensors,such as the Advanced Microwave Scanning Radiometer(AMSR-E)aboard the Earth Observing System(EOS)Aqua platform and the WindSat Radiometer on the U.S.Department of Defense Coriolis satellite,have demonstrated increasing RFI impacts on satellite measurements in C-and X-band channels and on geophysical parameter retrievals(Li et al., 2004,2006;Njoku et al.,2005;Ellingson and Johnson,2006;Kidd,2006;Wu and Weng,2011).Zou et al. (2012)detected RFI signals over land from the FY-3B Microwave Radiation Imager(MWRI), which is similar to AMSR-E but without channels at 6.9 GHz.The successor to AMSR-E,AMSRE-2 aboard the Japanese Shizuku mission(originally GCOM-W1(Global Change Observation Mission 1st-Water))(Kachi et al.,2008)launched on 18 May 2012 (JAXA,2012),is similar to AMSR-E,but is enhanced

for RFI detection with the addition of channels at 7.3 GHz.

Supported by the National Natural Science Foundation of China(41305033,41275043,and 41175035),Priority Academic Program Development(PAPD)of Jiangsu Higher Education Institution,and NOAA/NESDIS/Center for Satellite Applications and Research(STAR)CalVal Program.

∗Corresponding author:wuying-nuist@163.com.

©The Chinese Meteorological Society and Springer-Verlag Berlin Heidelberg 2014

Impacts of RFI on L-band satellite data have also been experienced by the Microwave Imaging Radiometer with Aperture Synthesis(MIRAS)aboard the ESA/SMOS(European Space Agency/Soil Moisture and Ocean Salinity)mission(Camps et al.,2010; Hallikainen et al.,2010;Skou et al.,2010;Anterrieu, 2011;Castro et al.,2012;Mecklenburg et al.,2012; Misra and Ruf,2012;Oliva et al.,2012)and NASA’s AQUARIUS(Le et al.,2007;Misra and Ruf,2008) mission,launched in June 2011.

In order to properly identify and reject RFI contamination,both hardware and software RFI detection and mitigation schemes(Gasiewski et al.,2002;Njoku et al.,2005;Johnson et al.,2006;Li et al.,2006;Ruf et al.,2006;Piepmeier et al.,2008;Wu and Weng, 2011;Lacava et al.,2013)have already been investigated over land.Njoku et al.(2005)examined the spatial and temporal characteristics of the RFI in the 6.9-and 10.7-GHz AMSR-E channels over the global land domain for a 1-yr observation period using spectral indices.Li et al.(2006)developed PCA-based (principal component analysis)land RFI algorithms to detect C-(6.9 GHz)and X-band(10.7 GHz)land RFIs of WindSat measurements through extending the spectral difference technique by using PCA of RFI indices.This method integrated statistics of target emission/scattering characteristics(through RFI indices)and multivariate correlation of radiometer data into a single statistical framework of PCA.Lacava et al.(2013)proposed the multi-temporal robust satellite techniques that can be implemented on C-band AMSR-E data to identify areas systematically affected by different levels of RFI.

In this paper,the RFI characteristics of C-and X-band data of AMSR-E,including the magnitude, extent,and location,are further investigated by using an RFI detection and correction algorithm proposed by Wu and Weng(2011).Emphasis is given to the applications of the algorithm in geophysical parameter retrievals over land using the one-dimensional variational retrieval(1-DVAR)approach.The retrieval convergence metric(or goodness of fit)between brightness temperature(TB)model simulations and satellite measurements is utilized to identify possible RFI-contaminated regions.Section 2 provides a description of AMSR-E missions and the RFI detection and correction algorithm.Section 3 describes the 1-DVAR approach,as well as the metric for identifying convergence.Numerical results are presented in Section 4. Section 5 provides a summary and some conclusions.

2.AMSR-E description

2.1 Mission overview

AMSR-E,launched on 4 May 2002,is a 12-channel,dual-polarization conically scanning passive microwave radiometer with 6 frequencies ranging from 6.9 to 89.0 GHz,which detects faint microwave emissions from the earth’s surface and atmosphere.

Various geophysical parameters can be retrieved from AMSR-E measurements,including water vapor, cloud liquid water,precipitation,sea surface temperature,sea surface wind speed,sea ice concentration, snow water equivalent,and soil moisture.Its global and continuous long-term geophysical record with fine spatial resolution plays an important role in climate change monitoring and provides valuable information for understanding the earth’s climate system,including water and energy circulation.Near real-time products will be used to investigate satellite data assimilation into weather forecasting models and to contribute to improvement of forecasting accuracy.

2.2 AMSR-E RFI detection and correction algorithm

It is well known that AMSR-E measurements at 6.9 and 10.7 GHz over land are seriously contaminated by variable surface radio frequency transmitters.In order to properly identify and reject increasing RFI contamination,Wu and Weng(2011)proposed an RFI identification and correction algorithm for AMSR-E channels.The algorithm is based on mean emissivity spectral characteristics over various land types that are simulated by using a microwave land emissivity model(Weng et al.,2001).An RFI index can be used to detect RFI over land.The larger the RFI index is,

the stronger the RFI contamination is.In this algorithm,the AMSR-E measurements over land with an RFI index greater than 5 K are defined as RFI-contaminated.

In the RFI correction,two empirically-based equations are derived based on AMSR-E training data under non-RFI contaminations,exploiting the fact that measurements at 18.7 GHz are rarely contaminated,i.e.,a relationship is established between AMSR-E measurements at 10.7 GHz and those at 18.7 or 6.9 GHz.If an RFI is detected at 10.7 GHz instead of 6.9 GHz,the RFI-contaminated AMSR-E measurements at 10.7 GHz are predicted from measurements at 18.7 or 6.9 GHz by using this relationship between two neighborhood frequencies.Also,if an RFI is detected at 6.9 GHz instead of 10.7 GHz, the RFI-contaminated measurements at 6.9 GHz can be predicted from measurements at 10.7 GHz.Again, when RFI contaminations are detected at both 6.9 and 10.7 GHz,RFI-contaminated measurements at these two frequencies can still be predicted from measurements at 18.7 GHz.

Moreover,it is found that AMSR-E measurements have better agreement with simulations in a variety of surface conditions after the RFI-correction algorithm(Wu and Weng,2011).As a result,one could expect to use more RFI-contaminated AMSRE measurements for satellite data retrieval with RFI mitigation.

3.1-DVAR approach

The 1-DVAR algorithm used in this paper is a component of the MIRS(Microwave Integrated Retrieval System)(Boukabara et al.,2011),which uses the Community Radiative Transfer Model(CRTM) (Han et al.,2006;Ding et al.,2011)as the forward and adjoint operators.The 1-DVAR inversion scheme solves the surface and atmospheric parameters simultaneously,e.g.,surface emissivities,temperature,profiles of atmospheric temperature,moisture and rainfall.Besides these primary parameters,other products are derived either by performing simple vertical integration,such as the total precipitable water(TPW),or by performing a more elaborate post-processing,such as the surface rainfall rate(RR)based on the hydrometeor parameters(Boukabara et al.,2011).

The variational approach employed in the inversion scheme seeks to minimize the following cost function,J(X)(Eyre et al.,1993;Boukabara et al.,2011), which measures the fit of the model to the radiances. Assuming Gaussian errors,this cost function can be written as

where Xbis the background state vector;B,which is associated with the back ground state variable Xb,is the error covariance matrix of X;E is the error covariance matrix of observations and/or forward models;Ymis the measurement.Specifically,assuming that we have a forward operator Y that can simulate radiances similar to the measurements without bias,the errors in the satellite observations and priori information are unbiased,uncorrelated,and have Gaussian distributions,and the best estimate of the atmospheric state X minimizes the cost function.The minimization of this cost function is also the basis for variational analysis retrieval(Boukabara et al.,2011). Minimization of the cost function is obtained by using an iterative process that computes the descent direction at state X by solving

The Jacobian matrix K corresponds to the partial derivatives of the radiative transfer with respect to X.

The forward operator is based on the CRTM developed by the Joint Center for Satellite Data Assimilation(JCSDA)in this study.The CRTM produces the simulated radiances as well as the Jacobian matrix K.The parameter χ2is calculated as

where χ2is used as a metric for deciding if convergence has been reached;χ2is also a measure of the goodness of fit of the forward model.Only those channels selected and effectively utilized in the 1-DVAR are used to compute this metric.

4.Results and discussion

4.1 RFI identification and correction

By identifying contaminated AMSR-E TB,affected channels can be ignored to produce retrievals unaffected by RFI.However,the utilization of fewer channels in the retrieval process may increase noise and decrease accuracy since various retrieved parameters are related to the effect of the frequency dependence.As a result,the spectral difference method(Wu and Weng,2011)is used to detect RFI signals over land in this study.In the RFI correction,a relationship between AMSR-E measurements at 10.65 GHz and those at 18.7 or 6.925 GHz is used to predict RFI-contaminated TBs over land(Wu and Weng,2011).

The convergence metric(χ2)used in the 1-DVAR approach could also be an excellent filter for detecting microwave data contaminated by RFI over land. Adams et al. (2010)described the geophysical retrieval chi-square probability method to identify regions of the ocean where ocean retrievals are affected by geostationary communication satellites.The magnitude of the χ2statistic depends on the number of measurements used in the retrieval and the number of retrieved parameters;χ2will be low when the forward model provides an excellent match with the AMSR-E TBs.Significant RFI in the AMSR-E TB will result in high χ2because the spectral characteristics of RFI differ from the spectral characteristics of natural sources (Li et al.,2006).

Comparisons of the convergence metric distributions using the 1-DVAR approach between two versions(before and after RFI correction)are shown in Fig.1.We use χ2as a relative measure to show how closely the forward model matches the measured TB. Various reasons,such as inaccurate modeling of the effects of geophysical variations,ice cover,and precipitation,could result in a poor goodness of fit between modeled and measured TB,which is presented by high χ2values.In these cases,the high χ2values are spatially correlated with the geophysical parameters to which AMSR-E is sensitive and will likely cause TB differences at multiple frequencies.Meanwhile,random measurement noise will be random both spatially and temporally,although it is usually frequency independent.However,man-made radiation from active microwave transmitters(or RFI to a radiometer)is distinctly different from natural radiation in terms of intensity,spatial variability,spectral characteristics,and channel correlations.RFI signals typically arise from a wide variety of coherent point target sources,i.e., radiating devices and antennas,which are often directional,isolated,narrowbanded,and coherent.These characteristics of RFI may help to provide the criteria of biases determined by RFI when χ2values are high.

Figures 1b,1e,and 1h show that convergence reaches almost everywhere,except for those regions where strong RFI exists(red dots in the ovals in Figs. 1a,1d,and 1g).This corresponds to those dots(in red)with high values of χ2from 1-DVAR in the ovals in Figs.1b,1e,and 1h.Moreover,it is evident that there is a high consistency between the values of RFI index and χ2.The stronger the RFI contamination, the larger the χ2.When RFI is larger than 10 K, the correlation coefficient between RFI at 6.9 GHz for horizontal/vertical polarization and χ2in the US is 0.912/0.479;the correlation coefficient between RFI at 10.7 GHz for horizontal/vertical polarization and χ2in Europe is 0.917/0.475;and the correlation coefficient between RFI at 10.7 GHz for horizontal/vertical polarization and χ2in Japan is 0.921/0.396.However, convergence reaches over larger regions after the RFI-correction algorithm(Wu and Weng,2011)is applied (oval areas in Figs.1c,1f,and 1i).

4.2 Validations of RFI identification and correction

Land and atmospheric parameters from MIRS products and for various sensors(NOAA-18,NOAA-19,Metop-A,and DMSPF16 SSMI/S)are validated by using NWP(numerical weather prediction)analyses,such as those from the ECMWF and NCEP GDAS (Global Data Assimilation System)(Boukabara et al., 2011).The inconsistency in the results suggests that there is intra-variability between the different references used.

Fig.1.Comparison of the convergence metric distributions between the two versions of 1-DVAR with and without RFI correction based on the AMSR-E data on 3 October 2008 for ascending orbits.Left panels(a,d,g)represent the RFI distribution in the US at 6.9 GHz,Europe at 10.7 GHz,and Japan at 10.7 GHz,respectively;middle panels(b,e, h)represent the convergence metric distributions without RFI detection and correction in the US,Europe,and Japan, respectively;and right panels(c,f,i)represent the convergence metric distributions with RFI detection and correction in the US,Europe,and Japan,respectively.

The NCEP GDAS outputs are taken as a reference in this study.Compared to collocated GDAS analysis,examples of MIRS outputs are presented in Figs.2,3,and 4.Note that in these figures,GDAS grid data are interpolated in time and space to the exact location and time of the AMSR-E measurement (Yang and Weng,2011)before the comparison is performed.

GDAS is the system used by the Global Forecast System(GFS)model to place observations into a gridded model space for the purpose of starting or initializing weather forecasts with observed data(Zheng et al.,2009;Yan and Weng,2011;Yang and Weng,2011). GDAS adds the following types of observations to a gridded 3-D model space:surface observations,balloon data,wind profiler data,aircraft reports,buoy observations,radar observations,and satellite observations.Currently,GDAS produces global analyses of temperature,water vapor profiles,and land parameters,such as land skin temperature(LST),soil moisture,snow depth,etc.,four times a day(0000,0600, 1200,and 1800 UTC)with a spatial resolution of approximately 0.3°after assimilating the conventional and satellite data.

An example of the LST difference between products from GDAS and MIRS over snow-free and snowcovered land surfaces on 3 October 2008 is presented in Fig.2.The middle panels in Fig.2 show the LST difference(LSTMIRS-LSTGDAS)between outputs from MIRS before RFI correction and GDAS,while the right panels in Fig.2 represent the LST difference between outputs from MIRS after RFI correction and GDAS.The GDAS LST products are given in the left panels.In addition,note that the accuracy of the skin temperature estimate from GDAS over the highlatitude regions is poor due to surface snow and ice cover.

Overall,major features of the surface temperature from MIRS are consistent with GDAS products. The differences are found to be more pronounced in the RFI-contaminated regions.For example,based on Fig. 2,the retrieved LST over these areas contaminated by RFI is much higher than GDAS LST products,which are represented by ovals in the middle panels.Meanwhile,the differences of LST between the retrieval and GDAS are evidently decreased through taking the RFI correction algorithm into account(shown in the ovals in the right panels).For the LST comparisons,similar to water vapor or TPW presented later,new retrievals are obtained by using RFI-corrected AMSR-E brightness temperatures as renewed inputs to the retrieval system.Furthermore,in regions with much stronger RFI,such as England,Italy,and Japan,the convergence metric(χ2)of 1-DVAR is too large and the convergence cannot be reached.Therefore,retrieval results cannot be obtained since the retrievals are unre-

liable.The LST statistics of studied areas before RFI correction show a mean bias of 2.42 K and a standard deviation of 6.29 K,while the statistics after RFI correction show a mean bias of 1.73 K and a standard deviation of 5.59 K.Additionally,note that the penetration of microwaves is as much as a few centimeters inside the soil,and this penetration depth is dependent on the frequency and on the type of soil,which creates both systematic biases and scattered differences.

Fig.3.Comparison of water vapor at 850 hPa derived from GDAS products and MIRS using AMSR-E data on 3 October 2008 for ascending orbits.Left panels(a,d,g)represent the water vapor at 850 hPa derived from GDAS products in the US,Europe,and Japan,respectively;middle panels(b,e,h)represent the water vapor difference at 850 hPa between products derived from MIRS without RFI correction and GDAS in the US,Europe,and Japan,respectively; and right panels(c,f,i)represent the water vapor difference at 850 hPa between products derived from MIRS with RFI correction and GDAS in the US,Europe,and Japan,respectively.

Another way to validate the AMSR-E RFI detection and correction algorithm is to compare the atmospheric moisture at 850 hPa derived from the MIRS algorithm and that provided by the GDAS analysis. Figure 3 shows the differences in the humidity field at 850 hPa as retrieved by MIRS using AMSR-E data on 3 October 2008,and as provided by the GDAS analysis.From these figures,we can see that the majority of moisture plumes and other large-scale features of GDAS are well captured by the MIRS retrievals. However,for those regions with weak or moderate RFI intensity,such as the US(Fig.1a),the values of MIRS retrievals(Fig.3b)are much higher than those from GDAS(Fig.3a).Meanwhile,in those regions

with the existence of extremely strong RFI,such as England,Italy,and Japan,the convergence metric (χ2)from 1-DVAR is so large that the retrievals are rejected in MIRS(Figs.3e and 3h),which are similar to the case of retrieved LST(Figs.2e and 2h).However, those retrievals with abnormally high values due to RFI in Figs.3b,3e,and 3h are mitigated in Figs.3c, 3f,and 3i by applying the RFI correction algorithm. The rejected retrievals due to strong RFI are also reobtained.The statistics of atmospheric moisture at 850 hPa for the studied areas before RFI correction show a mean bias of 0.36 g kg-1and a standard deviation of 2.56 g kg-1,while the statistics after RFI correction show a mean bias of 0.12 g kg-1and a standard deviation of 2.15 g kg-1.

According to the studies of Boukabara et al. (2011),it is noticed that the standard deviation over land using NOAA-18 data is consistent between the comparisons made with ECMWF data and those made with GDAS data,ranging between 54%at 300 hPa and 30%at the surface for the land case.However,the bias is not consistent between the ECMWF and GDAS as references.It is also different from the assessment results obtained when comparing to radiosondes.The uncertainty computed by using the radiosondes as a reference is similar at the surface with that obtained using ECMWF or GDAS(Boukabara et al.,2011).

Fig.4.Comparison of TPW derived from GDAS products and MIRS using AMSR-E data on 3 October 2008 for ascending orbits.Left panels(a,d,g)represent the TPW provided by the GDAS analysis in the US,Europe,and Japan, respectively;middle panels(b,e,h)represent the TPW difference between products derived from MIRS without RFI correction and GDAS in the US,Europe,and Japan,respectively;and right panels(c,f,i)represent the TPW difference between products derived from MIRS with RFI correction and GDAS in the US,Europe,and Japan,respectively.

The global TPW has proved to be a useful prod-

uct for many applications,including short-term precipitation forecasting and studies of the hydrological cycle.The retrieved moisture profile is vertically integrated to generate the TPW from MIRS,which by definition ensures that there is consistency between the profile and the TPW.The TPW from MIRS was found to be valid over all surface types,except when there is precipitation since it is radiometrically difficult to distinguish the water vapor signature from the liquid water signature.

Figure 4 presents a set of TPW maps corresponding to the MIRS retrieval and the GDAS-based products.It is shown that MIRS compares favorably to GDAS in terms of the distribution of the features as well as the statistical performance.The TPW statistics of the studied areas before RFI correction show a mean bias of-2.75 mm and a standard deviation of 9.88 mm,while the statistics after RFI correction show a mean bias of-3.37 mm and a standard deviation of 8.56 mm.

The two estimates(GDAS and MIRS)of TPW differ in snow and ice covered areas,which could be attributed to the poor accuracy of the skin temperature estimate from GDAS over these types of surfaces (Boukabara et al.,2011).Boukabara and Weng(2008) assessed the TPW global coverage from MIRS using a number of different reference datasets,and the assessments were stratified by surface background types as well as by sensors.

5.Summary and conclusions

An RFI detection and correction algorithm is applied in retrieving a comprehensive suite of geophysical parameters from AMSR-E measurements using the MIRS 1-DVAR method.The suite of parameters includes a set of derived and post-processed products also derived from MIRS.In the retrieval process,the RFI correction algorithm based on the natural channel correlations between AMSR-E measurements at neighborhood frequencies is used to predict and correct RFI-contaminated brightness temperatures.The results show that the difference caused by RFI between new retrievals and GDAS products,which are taken as reference,is evidently decreased.In particular,for those regions with much stronger RFI,the convergence metric(χ2)of 1-DVAR is significantly decreased after RFI correction such that the convergence can be reached and the retrieval results can be obtained.Furthermore,χ2from 1-DVAR could be utilized not only as a metric of goodness of fit between modeled and measured brightness temperature,but also as a detector to identify geographical regions of possible RFI.The χ2will be high when the forward model provides a poor match with the AMSR-E TB. The AMSR-E TB measurements are significantly contaminated by RFI since the spectral characteristics of RFI differ from the spectral characteristics of natural sources.The results of RFI detected by χ2are almost consistent with those identified by the spectral difference method.

Adams,I.S,M.H.Bettenhausen,P.W.Gaiser,et al.,2010: Identiflcation of ocean-reflected radiofrequency interference using WindSat retrieval chisquare probability.IEEE Geosci. Remote Sens. Lett.,7,406-410.

Anterrieu,E.,2011:On the detection and quantification of RFI in L1a signals provided by SMOS.IEEE Trans.Geosci.Remote Sens.,49,3986-3992.

Boukabara,S.A.,and F.Weng,2008:Microwave emissivity over ocean in all-weather conditions:Validation using WindSat and airborne GPS-dropsondes. IEEE Trans.Geosci.Remote Sens.,46,376-384.

—-,K.Garrett,W.Chen,et al.,2011:MiRS:An allweather 1DVAR satellite data assimilation and retrieval system.IEEE Trans.Geosci.Remote Sens., 49,3249-3272.

Camps,A.A.,J.Gourrion,J.M.Tarongi,et al.,2010: RIF Ranalysis in SMOS imagery.2010 IEEE International Geoscience and Remote Sensing Symposium(IGARSS).Honolulu,HI,IEEE,2007-2010, doi:10.1109/IGARSS.2010.5654268.

Castro,R.,A.Guti´errez,and J.Barbosa,2012:A first set of techniques to detect radio frequency interferences and mitigate their impact on SMOS data. IEEE Trans.Geosci.Remote Sens.,50,1440-1447.

Ding,S.,P.Yang,F.Weng,et al.,2011:Validation of the community radiative transfer model.J.Quant. Spectrosc.&Radiative Transfer.,112,1050-1064.

Ellingson,S.W.,and J.T.Johnson,2006:A polarimetric survey of radio-frequency interference in C-and X-bands in the continental United States using WindSat radiometry.IEEE Trans.Geosci.Remote Sens.,44,540-548.

Eyre,J.R.,G.A.Kelly,A.P.NcNally,et al.,1993: Assimilation of TOVS radiance information through one-dimensional variational analysis.Quart.J.Roy. Meteor.Soc.,119,1427-1463.

Gasiewski,A.J.,M.Klein,A.Yevgrafov,et al.,2002: Interference mitigation in passive microwave radiometry.IEEE International Geoscience and Remote Sensing Symposium.Toronto,Canada,IEEE, 1682-1684.

Hallikainen,M.,J.Kainulainen,J.Seppanen,et al.,2010: Studies of radio frequency interference at L-band using an airborne 2-D interferometric radiometer.2010 IEEE International Geoscience and Remote Sensing Symposium(IGARSS).Honolulu,HI,IEEE,2490-2491,doi:10.1109/IGARSS.2010.5651866.

Han,Y.,van P.Delst,Q.Liu,et al.,2006:Community Radiative Transfer Model(CRTM):Version 1,NOAA Technical Report NESDIS 122.NOAA, Washington,DC,33 pp.

Japan Aerospace Exploration Agency(JAXA)Press, 2012: Launch Resultofthe GlobalChanging Observation Mission 1st-Water“SHIZUKU”(GCOM-W1)and the Korean Multi-Purpose Satellite 3(KOMPSAT-3)by H-IIA Launch Vehicle No. 21. Available online at:http://www.jaxa. jp/press/2012/05/20120518-h2af21-e.html,2012.

Johnson,J.T.,A.J.Gasiewski,B.Guner,et al.,2006: Airborne radio-frequency interference studies at C-band using a digital receiver.IEEE Trans.Geosci. Remote.Sens.,44,1974-1985.

Kachi,M.,K.Imaoka,H.Fujii,et al.,2008:Status of GCOM-W1/AMSR2 development and science activities Sensors,Systems,and Next-Generation Satellites XII.Proceedings of the SPIE,7106,71060P-71060P-8,Meynart,R.,S.P.Neeck,H.Shimoda,et al.,doi:10.1117/12.801228.

Kidd,C.,2006:Radio frequency interference at passive microwaves observation frequencies.Int.J.Remote Sens.,27,3853-3865.

Lacava,T.,I.Coviello,M.Faruolo,et al.,2013: A multitemporalinvestigation ofAMSR-E C-band radio-frequency interference. IEEE Trans. Geosci. Remote Sens., 51, 2007-2015, doi: 10.1109/TGRS.2012.2228487.

Le Vine,D.M.,G.S.E.Lagerloef,F.R.Colomb,et al.,2007:Aquarius:An instrument to monitor sea surface salinity from space.IEEE Trans.Geosci. Remote Sens.,45,2040-2050.

Li,L.,E.G.Njoku,E.Im,et al.,2004:A preliminary survey of radio-frequency interference over the U.S. in Aqua AMSR-E data.IEEE Trans.Geosci.Remote Sens.,42,380-390.

—-,P.W.Gaiser,M.H.Bettenhausen,et al.,2006: WindSat radio-frequency interference signature and its identification over land and ocean.IEEE Trans. Geosci.Remote Sens.,44,530-539.

Mecklenburg,S.,M.Drusch,Y.H.Kerr,et al.,2012: ESA’s soil moisture and ocean salinity mission: Mission performance and operations.IEEE Trans. Geosci.Remote Sens.,50,1354-1366.

Misra,S.,and C.S.Ruf,2008:Detection of radiofrequency interference with the aquarius radiometer. IEEE Trans.Geosci.Remote Sens.,46,3123-3128.—-,and—-,2012:Analysis of radio frequency interference detection algorithms in the angular domain for SMOS.IEEE Trans.Geosci.Remote Sens.,50, 1448-1457.

Njoku,E.G.,P.Ashcroft,T.K.Chan,et al.,2005: Global survey and statistics of radio-frequency interference in AMSR-E land observations. IEEE Trans.Geosci.Remote Sens.,43,938-947.

Oliva,R.,E.Daganzo-Eusebio,Y.H.Kerr,et al.,2012: SMOS radio frequency interference scenario:Status and actions taken to improve the RFI environment in the 1400-1427-MHz passive band.IEEE Trans. Geosci.Remote Sens.,50,1427-1439.

Piepmeier,J.R.,P.N.Mohammed,and J.J.Knuble, 2008:A double detector for RFI mitigation in microwave radiometers.IEEE Trans.Geosci.Remote Sens.,46,458-465.

Ruf,C.,S.M.Gross,and S.Misra,2006:RFI detection and mitigation for microwave radiometry with an agile digital detector.IEEE Trans.Geosci.Remote Sens.,44,694-706.

Skou,N.,S.Misra,J.E.Balling,et al.,2010:L-band RFI as experienced during airborne campaigns in preparation for SMOS.IEEE Trans.Geosci.Remote Sens.,48,1398-1407.

Weng,F.,B.Yan,and N.C.Grody,2001:A microwave land emissivity model. J.Geophys. Res.,106, 20115-20123.

Wu Ying and Weng Fuzhong,2011:Detection and correction of AMSR-E radio-frequency interference.Acta Meteor.Sinica,25,669-681.

Yan,B.,and F.Weng,2011:Effects of microwave desert surface emissivity on AMSU-A data assimilation. IEEE Trans Geosci.Remote Sens.,49,1263-1276.

Yang,H.,and F.Weng,2011:Error sources in remote sensing of microwave land surface emissivity.IEEE Trans.Geosci.Remote Sens.,49,3437-3442.

Zheng,W.,J.Meng,H.Wei,et al.,2009:Improvement of satellite data utilization in NCEP operational NWP modeling and data assimilation systems.Proc.2nd Workshop Remote Sens.Model.Surf.Properties, Toulouse,France,June 9-11,14 pp.

Zou,X.,J.Zhao,F.Weng,et al.,2012:Detection of radio-frequency interference signal over land from FY-3B Microwave Radiation Imager(MWRI).IEEE Trans.Geosci.Remote Sens.,40,4994-5003.

:Wu Ying and Weng Fuzhong,2014:Applications of an AMSR-E RFI detection and correction algorithm in 1-DVAR over land.J.Meteor.Res.,28(4),645-655,

10.1007/s13351-014-3075-x.

(Received November 12,2013;in final form May 5,2014)