Thin ice thickness retrieval in the Kara Sea using MODIS and HIRLAM data

1. Thin ice thickness retrieval in the Kara Sea using MODIS and HIRLAM data Marko Mäkynen Finnish Meteorological Institute Ice Research Group

2. Introduction • FMI has retrieved thin ice thickness charts for the Kara Sea and eastern Barents Sea using MODIS and HIRLAM data. • Uncertainty figures for the MODIS thickness charts were also estimated. • In SMOSIce project the SMOS thickness charts are compared to MODIS charts: • Comparison of two remote sensing products with independent ice thickness estimates. • MODIS charts are not ‘validation’ data. They don’t show the ‘true’ ice thickness. • Agreements/disagreements between the two charts helps to fine tune the SMOS ice thickness algorithms.

3. Thin ice thickness from MODIS Physical basis:Thin ice thickness from ice surface temperature can be estimated on the basis of surface heat balance equation.Major assumptions here are that the heat flux through the ice and snow is equal to the atmospheric flux and temperature profiles are linear in ice and snow. Method presented e.g. in: Yu & Rothrock (1996). Thin ice thickness from satellite thermal imagery. Geophys. Res. 101(C10), 25753-25766. Requirement:The approach works only under cold cloud-free weather conditions (air temperature < -10°C). Using only nighttime data:Uncertainties related to the effects of the solar shortwave radiation and surface albedo are excluded. Reliable method for MODIS cloud masking needed! HIRLAM as weather forcing data. Parametrizations needed:e.g. snow vs. ice thickness, snow and ice thermal conductivity. 3

4. Thin ice thickness from MODIS Utilizing parameterization used by Yu and Rothrock (1996) and Russian Sever data (‘runway’ data) we determined snow vs. ice thickness parameterization for the Kara Sea. hs=0 cm, for hi<5 cm Same as Yu&Rothrock hs=0.05·hs, for 5 cm hi 20 cm Same as Yu&Rothrock hs=0.09·hs, for hi>20 cm The amount of Sever data for ice thickness less than 40 cm is very small. 4

5. Barents and Kara Seas study area Red dots are weather stations. Coverage 1500 by 1350 km. 5

6. Data sets Cloud masked MODIS ice surface temperature images: 48 images for 1 Jan – 30 Apr 2010 81 images for 1 Nov 2010 – 30 Apr 2011 HIRLAM (v. 7.3) weather forcing data computed in 20 km grid. Air temperature, wind speed, relative humidity, downward longwave radiative flux. Boundary conditions from ECMWF. The sea ice concentration input for HIRLAM was based on the ECMWF operational sea surface temperature analysis. Weather station data for HIRLAM validation. Landmask derived from the NASA’s MODIS 250 m land-water mask product (MOD44W). 6

7. MODIS data processing MODIS nighttime data rectified to polar stereographic projection with 1 km pixel size using MODIS SwathTool. MODIS nominal resolution 1 km. Sensor scan angle limited to 40 degrees at which the resolution is 2 km. MODIS cloud masking: The quality of the cloudmask in the MODIS/Terra Sea Ice Extent 5-Min L2 Swath 1km - MOD29 product is sometimes not good enough. Thus, we cloudmasked the data using 1) three cloudtests (11-3.9 μm BTD, 3.9-12 μm BTD and 6.7 μm BT) with thresholds empirically determined for the Kara Sea, and 2) manual methods using RGB-images. 7

8. MODIS data processing Cloudmasking performed with 10x10 km pixel blocks. We want to identify large cloud-free areas for the ice thickness retrieval. Block based cloudmask is less ‘grainy’ than pixel based mask in MOD29 product. After automatic cloudtests morphological operations are performed; e.g. to fill isolated small holes. Thermal RGB-images: 1) brightness temperature channels 20 (red), 31 (green) and 32 (blue), and 2) channel difference 32-31 (red), difference 31-22 (green) and channel 31 (blue) (used for Meteosat SEVIRI data and called as NightMicrophysical). All data processing done with Matlab. Image Processing Toolbox needed for the cloud masking. MODIS ice thickness image shows thickness in the 0 to 1 m range and various masks (no data, land, clouds, thickness retrieval unsuccessful or over 1 m, air temperature too high). 8

9. Thermal RGB image, channels 20, 31, 32. 9

10. Cloud masked thermal RGB image 10

11. MODIS ice surface temperature 11

12. HIRLAM air temperature 12

13. MODIS and HIRLAM based ice thickness 13

14. An example of difference between MOD29 and our IST image. 14

15. An example of difference between MOD29 and our IST image. 15

16. An example of difference between MOD29 and our IST image. 16

17. Comments on HIRLAM accuracy HIRLAM accuracy studied by comparing HIRLAM and coastal weather station datain winter 2010-11 (HIRLAM minus WS data). For air the temperature Ta : mean bias -0.9 K, rms-difference 3.8 K and correlation 0.93. Mean bias increases with decreasing Ta: for the Ta ranges of -10 – -5 ºC and -25 – -20 ºC it is -0.1 ºC and -2.3 ºC. Too low HIRLAM Ta leads to too small retrieved ice thickness. For the wind speed: mean bias -1.2 m/s, rms-difference 3.3 m/s, and correlation 0.67. HIRLAM typically underestimates higher wind speeds (>10 m/s). HIRLAM sea ice mask does not include thin ice and leads, only open water vs. thick ice. A modeling study (Lüpkes et al. 2008) has demonstrated that for sea ice concentrations > 90% small changes in the sea ice fraction have a strong effect on the near-surface Ta. 17

18. Accuracy of the MODIS ice thickness The accuracy of the MODIS thickness charts was studied by: 1) Using estimated or guessed standard deviations and correlations of the input variables to the thickness retrieval the hi uncertainty is estimated with the Monte Carlo method. Results of the HIRLAM accuracy study used here. 2) MODIS thickness charts from consecutive days are compared to each other. Large differences are mainly due to the cloud masking errors and HIRLAM data inaccuracies. Estimates repeatability of the hiretrieval when the true hi change is insignificant. Currently we don’t have any in-situ thickness data for the Kara Sea. 18

19. MODIS hi accuracy: Monte Carlo method hi uncertainty as a function of Ta. 50% uncertainty reached:Ta < -30 °Chi =60 cm -20 °C <Ta < -15 °Chi =25 cm HIRLAM Ta <-20 ºC in 86% of the cases. 19

20. MODIS hi accuracy: comparison of consecutive charts t from 15 to 33 h 10 km block averages for hi to diminish the effect of ice movement. Overall RMSE 7.7 cm. For hi range from 20 to 60 cm in 10 cm bins, RMSE is from 18% to 34%. Cloud masking errors likely the main source of hi differences. 20

21. MODIS hi accuracy: Ta - TS vs. hi relationship Empirical curves from the HIRLAM and MODIS data. When Ta-Ts is close 0 °C then only 1 °C change can cause 10 cm change in hi. Maximum hi depends also on these curves. 21

22. Accuracy of the MODIS ice thickness As the hi uncertainty depends considerably on Ta, wind speed and other variables it is difficult to determine the typical maximum for reliable hi . Combining the results of different accuracy analyses the typical maximum reliable hiis roughly 40-50 cm. Determined using data for both level and deformed ice. MODIS cannot discriminate between them. 22

23. MODIS ice thickness charts 81 charts for 2010-2011 23

24. Conclusions MODIS ice surface temperature and HIRLAM forcing data can be used to estimate ice thickness up to 40-50 cm in the Kara Sea. Problems in MODIS based hi: (1) HIRLAM accuracy: in very cold conditions, even a small change in the ice concentration has a large effect on Ts and Ta (2) snow thickness accuracy (3) sometimes rare availability of cloud-free MODIS data (4) cloud mask errors (fog and thin high clouds) Manual methods used in the MODIS cloud masking. These are not suitable for operative use. Fully automatic cloud masking of nighttime images possible? How VIIRS 0.7 µm Day/Night band with capalility of nighttime imaging under moonlight will improve automatic cloud masking? 24