The Arctic's perennial sea ice concentrations from a neural network analysis of SMMR-SSM/I data, 1979-2005

Principal Investigators:
Gennady I. Belchansky¹, David C. Douglas²

Investigators:
Ilia V. Alpatsky¹, Vladimir A. Eremeev¹, Ilia N. Mordvintsev, Nikita G. Platonov¹

Abstract

This data set assimilates measurements from several active and passive microwave satellite sensors and includes a 27-year (1979-2005) record of winter (January - March) daily and monthly multiyear sea ice concentrations, derived from the Nimbus-7 Scanning Multichannel Microwave Radiometer and Defense Meteorological Satellite Program Special Sensor Microwave/Imager brightness temperatures using artificial neural networks. These networks were learned on the multiyear sea ice concentrations, estimated using the active and passive microwave observations of Russian Okean-01 #7 and #8 satellites and ERS-1 Synthetic Aperture Radar, and synchronous measurements of the brightness temperature from vertically and horizontally polarized 19GHz channels and vertically polarized 37GHz channel.

Spatial coverage and resolution of concentration maps are the same, as that of SMMR, SSM/I daily gridded products of the NSIDC. Data are provided in single-byte format in the SSM/I polar stereographic projection with pixel size of 25x25 km.

1. Overview

1.1. Data set identification

Winter MY NN sea ice concentrations from Nimbus-7 SMMR, DMSP SSM/I, ERS-1 SAR, and OKEAN-01 #7 and #8.

1.2. Introduction

Arctic sea ice is a significant element of the climate system, because it regulates heat exchange between ocean and atmosphere, salinity and circulation. Therefore, sea ice data is important input for climate modeling. Sea ice habitats have a vital value for the arctic biota, and their studies could help in assessing effects of habitat change, and investigating governing processes in the biosphere and climate. Knowledge of sea ice conditions is important for the navigation in the Arctic and developing natural resources of the coastal area.

Variability and trends of sea ice distribution have been investigated in various studies [Parkinson et al., 1998; Comiso, 2002; Parkinson and Cavalieri, 2002; Serreze et al., 2003; Belchansky et al. 2004, 2005; Stone et al. 2005]. These studies are supported by many sea ice data sets, provided by the National Snow and Ice Data Center (NSIDC, http://nsidc.org), the Applied Physics Laboratory at the University of Washington (APL, http://www.apl.washington.edu), the Arctic and Antarctic Research Institute (AARI, http://www.aari.nw.ru), the National Ice Center (NIC, http://www.natice.noaa.gov), the RADARSAT Geophysical Processor System (RGPS, http://www-radar.jpl.nasa.gov), the Alaska Satellite Facility (ASF, http://www.asf.alaska.edu) and others. Modern sea ice data sets include DMSP SSM/I Daily Polar Gridded Brightness Temperatures, Near Real-Time DMSP SSM/I Daily Polar Gridded Brightness Temperatures, Nimbus-7 SMMR Polar Gridded Radiances and Sea Ice Concentrations, DMSP SSM/I Daily and Monthly Polar Gridded Sea Ice Concentrations, Submarine Upward Looking Sonar Ice Draft Profile Data and Statistics (from the NSIDC), High-Resolution Ice Motion, Deformation, Age, And Thickness (from the RGPS), Buoy measurement data from International Arctic Buoy Program (IABP, from the APL), weekly charts of sea ice form and distribution (from the NIC), various radar sea ice products from the ASF, etc.

We present "The Arctic's perennial sea ice concentrations" data set including the 27-year record (1979-2005) of winter (January - March) daily and monthly perennial (multiyear, MY) sea ice concentrations, and covering the whole Northern polar region. MY sea ice concentrations were derived from the Nimbus-7 SMMR and DMSP SSM/I brightness temperatures [Gloerson et al, 1990; Cavalieri et al. 1990] using artificial neural networks (NN). These NNs were learned on the MY sea ice concentrations, estimated using the active and passive microwave data of Russian OKEAN-01 #7 and #8 satellites and ERS-1 SAR, and synchronous measurements of the brightness temperature (Tb) from SSM/I vertically and horizontally polarized 19GHz channels and vertically polarized 37GHz channel onboard DMSP F8 and F13 satellites.

The work, yielded the creation of the presented data set, has began from developing the methodology of processing synchronous active and passive microwave measurements from Russian OKEAN-01 satellite series. These measurements allowed estimation of several sea ice types and concentrations [Belchansky et al., 1995; Belchansky and Douglas, 2000]. The studies are ongoing, and it is planned to make improvements of the presented data set, by utilizing data from new sensors and new efficient neural network learning algorithms.

The neural network theory and definitions, used in this work, are described in [Belchansky et al., 2004].

At present, several scientific organizations in the world already use this data set, provided by their personal request.

2. Description of source data and their processing

2.1. Sensor/Instrument description

2.1.1. OKEAN-01 satellites, real aperture radar and microwave radiometer RM-08

Parameters of the OKEAN-01 #7 and #8 satellites, and technical characteristics of the sensors, they equipped with, are described in [Belchansky and Douglas, 2000]. Latest information can be found at http://sputnik.infospace.ru/ocean/engl/ocean.htm, and at http://www.nkau.gov.ua.

2.1.2. SMMR, SSM/I

The information about Nimbus-7 and DMSP satellites and SMMR and SSM/I sensors can be found in [Cavalieri et al., 1984; Comiso, 1991] and in the following links:
SMMR: http://nsidc.org/data/docs/daac/smmr_instrument.gd.html
SSM/I: http://nsidc.org/data/docs/daac/ssmi_instrument.gd.html
A reader should also refer to the description of the NSIDC Daily Polar Gridded Brightness temperatures [Gloerson et al, 1990; Maslanik and Stroeve, 1990] (http://nsidc.org/data/docs/daac/nsidc0001_ssmi_tbs.gd.html).

2.1.3. ERS-1 Synthetic Aperture Radar

The information about the ERS-1 SAR and the ERS Geoprocessing System data can be found in [Kwok et al, 1990; Kwok et al, 1995], and in the following links: http://www.asf.alaska.edu (ASF) and http://www-radar.jpl.nasa.gov/rgps/radarsat.html (RGPS).

2.2. Data acquisition and processing

2.2.1. SMMR, SSM/I

SMMR data acquisition, processing and generating daily gridded brightness temperature maps was done by the Goddard Space Flight Center (GSFC) [Gloerson et al. 1990]. DMSP SSM/I data acquisition, bad data filtering, handling geolocation errors, implementation of an antenna pattern correction, and production of daily gridded brightness temperature maps are performed by NSIDC [Maslanik and Stroeve, 1990 updated 2005]. Daily Polar Gridded SMMR and SSM/I brightness temperatures were shipped from the NSIDC on CD-ROM in accomplishment of the Sector of Space Monitoring and Ecoinformation Systems (IEE, Russian Academy of Sciences, Moscow, Russia) request.

2.2.2. ERS-1 SAR

Acquisition and processing of ERS-1 data was performed by the ERS Geoprocessing System (Alaska Satellite Facility, Faibanks) [Kwok and Cunningham, 1993]. Data were included in the ERS-1 Ice Classification Product, provided by the ASF with support of the NASA HQ grant (ASF project "Multisensor satellite monitoring of Arctic sea ice in conditions of global change", approved by the NASA HQ).

2.2.3. OKEAN-01 side-looking real aperture radar and passive microwave radiometer

Acquisition of synchronous active and passive data from OKEAN-01 satellites was performed by the Scientific Research Center "Planeta" (Dolgoprudny, Moscow District, Russia, http://sputnik.infospace.ru), which provided source satellite measurements to the Space Monitoring and Ecoinformation Systems sector. Data processing, including radiometric calibration, geometric correction, and estimating of the MY sea ice fraction was conducted in the Sector of Space Monitoring and Ecoinformation Systems, (IEE Russian Academy of Sciences, Moscow, Russia) [Belchansky et al, 1995; Belchansky and Douglas, 2000].

3. MY sea ice data set characteristics, data organization, and processing

3.1. Spatial and temporal characteristics of MY sea ice data set

Provided MY sea ice concentration maps have the same spatial coverage, pixel size, and projection as the Daily Gridded SMMR, SSM/I brightness temperatures (304x448 pixels, polar stereographic projection, pixel size 25x25 km) [Gloerson et al. 1990; Maslanik and Stroeve, 1990].

The dataset covers the first three months (January - March) for each year from 1979 through 2005. MY sea ice maps are developed for each day. Monthly averaged MY sea ice maps also exist. Source daily brightness temperature maps have pixels with missing Tb values, and also days exist for which Tb maps are totally missing [Gloerson et al. 1990; Maslanik and Stroeve, 1990; Cavalieri et al., 1999]. When possible, MY sea ice concentrations were calculated from the sensor measurements. When Tb data were missing, MY sea ice estimates were produced using linear interpolation of MY sea ice concentrations from the neighboring pixels or days. A major data gap from 3 December 1987 through 13 January 1988 exists and was not filled.

3.2. Data organization

Data files are stored in the flat binary single-byte integer format. The file naming convention for daily data is yymmdd.bin and avgyymm.bin for monthly data, where yy - last two digits of year, mm - month, dd - day of month.

Examples: 000131.bin (31 January 2000) and avg0101.bin (January 2001).

Monthly collections of daily maps are archived with ZIP archiver. Archive names have form yyyymm.zip, where yyyy - a year, mm - a month.

Data range: 0-104. MY sea ice concentrations have range 0-100, 101 denotes a land (NSIDC land mask was used) and 104 denotes missing data (a circular area around the North pole invisible to sensors due to orbit inclination).

3.3. Data processing sequence

3.3.1. Deriving consistent long-term record of SMMR and SSM/I brightness temperatures

The consistent records of SMMR-SSM/I brightness temperatures from the channels 18/19v, 18/19h, and 37v were produced by correcting for instrument drift and differences in sensor characteristics. Microwave data from DMSP F-8 SSM/I were arbitrarily chosen as a basis. Linear regression method was used to transform passive microwave data from another satellites to the SSM/I F8 basis. Linear regression coefficients were obtained from the synchronous measurements of the Greenland and Antarctic ice sheets [Abdalati et al., 1995; Stroeve et al., 1998; Jezek et al. 1991]. We also applied linear regression coefficients to standardize intersatellite Tb calibrations: F13 SSM/I to F11 SSM/I [Stroeve et al., 1998], and F11 SSM/I to F8 SSM/I [Abdalati et al., 1995].

3.3.2. Extraction of MY sea ice concentrations from OKEAN-01 data set for learning neural networks

Concentrations of MY ice, first-year (FY) ice, and open water (OW) were estimated within a 3 km x 3 km pixel-resolution grid using linear mixture modeling of the measured backscattering cross section (s⁰) and Tb values [Belchansky and Douglas, 2000; Belchansky et al, 2004]. The OKEAN MY maps were averaged daily within SSM/I 25 km x 25 km grid, and MY ice concentration estimates were extracted to be inserted in the neural network learning data set.

3.3.3. Extraction of MY sea ice concentrations from ERS SAR data set for learning neural networks

The ERS images had been individually processed and were disseminated as 100 km x 100 km ice maps with 5-km pixel size in polar stereographic projection. For network learning, the ERS MY ice concentration maps were averaged daily within SSM/I 25-km resolution grid. Then concentration values were extracted from maps for the further creation of NN learning data set.

3.3.4. Creation of NN learning data sets and inversion of MY sea ice concentrations using NN

Data sets for learning neural networks consisted of pairs input - output values, where brightness temperatures from channels 18/19v, 18/19h, and 37v acted as input and corresponding MY sea ice concentrations were output. We used multi-layer perceptrons with 3-20-1 topologies, learned with combined method of the error backpropagation and simulated annealing [Belchansky et al, 2004]. Different NNs were trained for different months of a year.

3.3.5. Weather filter and filling data gaps in MY sea ice concentration maps

The weather filter was applied to MY sea ice concentration maps to eliminate spurious sea ice concentrations in the areas of the open ocean [Gloersen, 1986; Comiso, 1990]. Missing data in the SMMR and SSM/I brightness temperature data set were scattered pixels of empty data, caused generally by the geometrical corrections, and larger areas, caused by sensor failures and gaps between single satellite passes [Cavalieri et al., 1999]. The scattered single empty pixels were filled by the linear interpolation of 8 neighboring non-empty pixel values. The larger empty areas were filled by the linear interpolation of data from the maps of neighboring days.

4. Notes and plans

4.1. Limitations of the data

The data limitations are caused by the sensor resolution, temporal coverage and some lack of the learning data, used to train the neural networks.

Another deficiency of the algorithm is the separate independent learning of different NNs for different months, resulting sometimes in sharp month-to-month transition effects.

MY sea ice maps have a hole around the Northern Pole with latitudes from 84 degrees of SMMR and 87 degrees for the SSM/I. This hole is caused by the absence of the microwave measurements due to the satellite orbit inclination.

4.2. Plans

It is planned to improve MY sea ice concentration data set by including RADARSAT MY sea ice concentrations from the RGPS [Kwok and Cunningham, 2000], brightness temperatures from the new Advanced Microwave Scanning Radiometer (AMSR) onboard Aqua satellite, estimating MY sea ice concentrations for the whole year, and utilizing new efficient algorithms of the neural network learning.

5. Data Access (Software)

MY sea ice data can be easily loaded into any image processing software, allowing importing of the raw binary data. Some examples of programmatic data access are given below.

The sample C function, reading single daily MY sea ice map into a computer memory:

#include <stdio.h>
#include <stdlib.h>

unsigned char *load_image(char *image_name){
 unsigned char *image;
 FILE *fp;
 int nrows=448, ncols=304;

  if((image=(unsigned char *)malloc(nrows*ncols))==NULL) 
    return NULL;
  if((fp=fopen(image_name,"rb"))==NULL) {
    free(image); 
    return NULL;
  }
  fread(image,1,ncols*nrows,fp);
  fclose(fp);
  return image;
}

This function allocates the memory, loads MY sea ice map from a file into it and returns the pointer to it. In case of failures the function returns NULL.

The call load_image("020104.bin") will load into memory MY sea concentration map for the date January 4th, 2002, provided the file 020104.bin, containing this map, is in the current directory.

Same operations in R (http://www.r-project.org):

 cmy<-matrix(data=readBin("020104.bin","integer",n=304*448,size=1,signed=FALSE),ncol=304,byrow=TRUE)

After executing this instruction, the matrix variable cmy will be created in the interpreter memory. The matrix dimensions are 304 columns by 448 rows; each element contains the value of MY sea concentration. Please, note that R indexes matrices by a row first, therefore, the matrix should be transposed and vertically flipped in order to get properly oriented image with the function image.

Same operations in the Interactive Data Language (IDL) by the Research Systems, Inc (http://www.rsinc.com):

data_dims=[304,448]
filename="020104.bin"
cmy=read_binary(filename,data_type=1,data_dims=data_dims)

IDL has the function read_binary since the version 5.3.

6. Acknowledgments

This work was carried out with the support of the International Arctic Research Center, University of Alaska Fairbanks. We would like to acknowledge the Alaska Satellite Facility (Fairbanks) for providing ERS-1 sea-ice concentrations, the National Snow and Ice Data Center (University of Colorado) for providing the SMMR, SSM/I Daily Polar Gridded Tb, and the Scientific Research Center "Planeta" for providing OKEAN-01 satellite data.

7. References

Abdalati, W., K. Steffen, C. Otto, and K. Jezek (1995), Comparison of brightness temperatures from SSM/I instruments on the DMSP F8 and F11 satellites for Antarctica and the Greenland ice sheet, Int. J. Remote Sens., 16, 1223-1229.

Belchansky, G.I., Douglas, D.C., Mordvintsev, I.N. and Ovchinnikov, G.K. (1995). Assessing trends in Arctic sea ice distribution using Kosmos-Okean satellite series. Polar Record 31(177): 129-134.

Belchansky, G. I., and D. C. Douglas (2000). Classification methods for monitoring Arctic sea-ice using Okean passive / active two-channel microwave data, Remote Sens. Environ., 73, 307- 322.

Belchansky, G. I. and D. C. Douglas. (2001). Seasonal comparisons of sea ice concentration estimates derived from SSM/I, OKEAN and RADARSAT data. Remote Sens. Env. 81:1-15.

Belchansky G.I., Douglas D.C., Alpatsky I.V., Platonov N.G. (2004). Spatial and Temporal Multiyear Sea Ice Distributions in the Arctic: A Neural Network Analysis of SSM/I Data, 1988-2001. J. Geophys. Res., 109 (C12), doi:10.1029/2004JC002388.

Belchansky G.I., Douglas D. C., Eremeev V.A., Platonov N.G. (2005). Variations in the Arctic's perennial sea ice cover: A neural network analysis of SMMR-SSM/I data, 1979-2004. Geophys. Res. Lett., 32, L09605, doi:10.1029/2005GL022395.

Cavalieri, D.J., P. Gloersen, and W.J. Campbell. (1984). Determination of sea ice parameters with the NIMBUS-7 SMMR. Journal of Geophysical Research 89(D4):5355-5369.

Cavalieri D., P. Gloerson, and J. Zwally. (1990, updated 2005). DMSP SSM/I daily polar gridded sea ice concentrations. Edited by J. Maslanik and J. Stroeve. Boulder, CO: National Snow and Ice Data Center. Digital media.

Cavalieri, D. J., C. L. Parkinson, P. Gloersen, J. C. Comiso, and H. J Zwally (1999). Deriving long-term time series of sea ice cover from satellite passive-microwave multisensor data sets. J. Geophys. Res., 104, 15,803-15,814.

Comiso, J.C. (1991). Satellite remote sensing of the polar oceans. J. Geophys. Res. 2:295-434.

Comiso J. (1990, updated 2005). Bootstrap sea ice concentrations for Nimbus-7 SMMR and DMSP SSM/I, June to September 2001. Boulder, CO, USA: National Snow and Ice Data Center. Digital Media.

Comiso, J. C. (2002), A rapidly declining perennial sea ice cover in the Arctic, Geophys. Res. Lett., 29, 1956, doi:10.1029/2002GL015650.

Gloersen P., Cavalieri D.J. (1986). Reduction of weather effects in the calculation of sea ice concentrations from the microwave radiances. J. Geophys. Res. 91 (C3), pp. 3913-3919.

Gloerson, P., D. Cavalieri, W.J. Campbell, and J. Zwally (1990). Nimbus-7 SMMR polar radiances and Arctic and Antarctic sea ice concentrations. Boulder, CO: National Snow and Ice Data Center. CD-ROM.

Jezek K., Merry C., Cavalieri D., Grace S., Bender J., Wilson D., and Lamkin D., (1991) Comparison between SMMR and SSM/I passive microwave data collected over the Antarctic ice sheet. Byrd Polar Research Center Technical Report. No. 91-03, The Ohio State University, Columbus, Ohio. 1991. 62 pp.

Kwok, R., J. C. Curlander, R. McConnell and S. S. Pang (1990). An ice-motion tracking system for the Alaska SAR Facility Geophysical Processor System, IEEE J. Ocean Eng., v15, p44- 54.

Kwok, R., and G. F. Cunningham (1993), Alaska SAR Facility Geophysical Processor System Data User's Handbook, Version 2. National Aeronautics and Space Administration, Jet Propulsion Laboratory. JPL D-9526.

Kwok, R., D.A. Rothrock, H.L. Stern, and G.F. Cunningham (1995). Determination of the Age Distribution of Sea Ice from Lagrangian Observations of Ice Motion, IEEE Trans. Geosci. Rem. Sens., vol. 33, #2, pp. 392-400.

Kwok R., Cunningham G.F. (2000) RADARSAT Geophysical processor system. Data user's handbook. Jet Propulsion Laboratory, Pasadena, California.

Maslanik, J., and J. Stroeve. (1990, updated 2005). DMSP SSM/I daily polar gridded brightness temperatures. Boulder, CO: National Snow and Ice Data Center. CD-ROM.

Parkinson C.L., Cavalieri D.J., Gloersen P., Zwally H.J., and Comiso J.C. (1998), Arctic sea ice extents, areas, and trends, 1978-1996, , J. Geophys. Res., 104, 20,837-20,856.

Parkinson, C. L., and D. J. Cavalieri (2002). A 21 year record of Arctic sea-ice extents and their regional, seasonal and monthly variability and trends, Ann. Glaciol., 34, 441-446.

Serreze, M. C., J. A. Maslanik, T. A. Scambos, F. Fetterer, J. Stroeve, K. Knowles, C. Fowler, S. Drobot, R. G. Barry, and T. M. Haran (2003). A record minimum arctic sea ice extent and area in 2002. Geophys. Res. Lett., 30, 1110, doi:10.1029/2002GL016406.

Stone, R. S., D. C. Douglas, G. I. Belchansky, S. D. Drobot, and J. Harris (2005). Cause and effect of variations in western Arctic snow and sea ice cover. 8.3, Proc. Am. Meteorol. Soc. 8 th Conf. on Polar Oceanogr. and Meteorol. San Diego, CA , 9-13 January.

Stroeve, J., J. Maslanik, and L. Xiaoming (1998). An intercomparison of DMSP F11- and F13- derived sea ice products, Remote Sens. Environ., 64, 132-152.