Presentation is loading. Please wait.

Presentation is loading. Please wait.

VIIRS Cloud Mask Validation Exercises

Similar presentations


Presentation on theme: "VIIRS Cloud Mask Validation Exercises"— Presentation transcript:

1 VIIRS Cloud Mask Validation Exercises
A13B-0233 Richard Frey1, Andrew Heidinger2,1, Keith Hutchison3, Denis Botambekov1, and Steven Dutcher1 1Cooperative Institute for Meteorological Satellite Studies, Space Science and Engineering Center, University of Wisconsin-Madison 2Office of Research and Applications, NOAA / NESDIS, Madison, WI, USA 3Center for Space Research, University of Texas, Austin, Austin, TX Introduction The Joint Polar Satellite System (JPSS) Visible/Infrared Imager Radiometer Suite (VIIRS) is a scanning radiometer that will collect visible and infrared imagery and radiometric measurements of the land, atmosphere, cryosphere, and oceans. It extends a series of measurements initiated by the Advanced Very High Resolution Radiometer (AVHRR) and the Moderate Resolution Imaging Spectroradiometer (MODIS). VIIRS data will be used to measure cloud and aerosol properties, ocean color, sea and land surface temperature, ice motion and temperature, fires, and Earth's albedo. Identifying pixels as either cloudy or clear is an essential component of VIIRS data processing. The VIIRS Cloud Mask (VCM) technique incorporates a number of cloud detection tests that determine whether a pixel is obstructed by a cloud or is cloud free. If a cloud is detected, the VCM indicates whether its phase is water, ice, or mixed. Additionally, the VCM specifies whether aerosols, fire, or shadows are detected within the pixel field of view. The VCM performs a series of visible/NIR threshold and ratio tests, as well as brightness temperature threshold and difference tests. The VCM tests have a heritage from the AVHRR and MODIS, where each returns a clear or cloudy result with an associated confidence of clear sky (0-1). Analogous to the MODIS cloud mask (MOD35), the VCM groups its cloud confidence tests into five categories. The minimum confidence from individual tests within a group represents the confidence for that group and the product of all the group confidences is used to determine the overall clear sky confidence value. This poster details some of the methods that will be used to validate the VIIRS cloud mask once data becomes available. Some initial validation of the VCM algorithm has already been done by use of MODIS proxy data and is shown here. Zonal Means of Cloud Cover and Comparisons to other Cloud Masks Four months of VCM output were created using MODIS Aqua L1b and geolocation as proxies for VIIRS inputs. Monthly, zonal means were created for January and April, 2007 (right) and July and October, 2007 (below). These were compared to PATMOS-x, MODIS (MOD35), MODIS CERES, ISCCP-D1, and CALIOP lidar data that were obtained from the GEWEX data set. Daytime data is shown in the plots. Pixels that were labeled cloudy and probably cloudy by the VCM were counted as clouds. Results for January and April (right) show that generally, the VCM finds fewer clouds than the other algorithms from 60S to 30N latitude. In the northern mid-latitude and Arctic regions, dominated by land surfaces, the VCM is comparable to the other data sets. In the Antarctic, the VCM over-clouds relative to the others, including the active sensor CALIOP, which is widely regarded as being the most accurate among these. Keep in mind that the “cleanest” comparison is between VCM and MODIS (MOD35) as both used the same input radiance, geolocation, and ancillary data. E E VIIRS Photo courtesy of Raytheon Space and Airborne Systems Use of CALIOP Cloud Data for Cloud Mask Validation Four months of CALIOP cloud data and MODIS radiance and cloud mask data were collocated and analyzed (January, April, July, October, 2007). This allowed comparisons between MODIS Collection 6 cloud mask (MOD35) and the latest available VIIRS Cloud Mask (VCM) results. MODIS radiance and geolocation data were used as proxies for VIIRS inputs. The VCM was generated using the Low Earth Orbiter Cloud Algorithm Testbed (LEOCAT) software to stage and prepare all radiance, geolocation, and ancillary data sets. Note that the 375-meter imagery resolution variability tests were not performed as there is no way to simulate these bands using proxy MODIS L1b inputs. It is anticipated that these tests will increase the VCM accuracy over oceans, especially at night. The tables at right and below show percent agreement (“hit rate”) between CALIOP and the VCM, and CALIOP and MOD35 for various scene types. It is expected that the VCM should show lesser agreement than MODIS, given that MOD35 has benefited from over ten years of research and development. For example, the VIIRS cloud mask has not been tuned to fully exploit the VIIRS 1.38-micron band since it has less out of band response than MODIS. It appears that the most developed VCM algorithm is non-polar daytime ocean. The VCM is of comparable quality to MOD35 in daytime Arctic conditions. January 2007 Scene Type VCM / MOD35 C6 vs. CALIOP Agreement 60S-60N Day Ocean 89.6% / 91.5% 60S-60N Day Land 83.8% / 85.3% 60S-60N Night Ocean 85.0% / 91.1% 60S-60N Night Land 76.7% / 85.1% 60S-60N Day Desert 81.6% / 84.0% 60S-60N Night Desert 77.2% / 81.0% Antarctic Day 74.5% / 88.4% Plots for July and October (left) show the VCM having fewer cloud detections from about 60S to 30N for October, but extending to 60N for July. Arctic regions again show the VCM comparable to the other imager algorithms, but northern mid-latitudes show the VCM having less (July) or near the least (October) zonal cloud amounts. Large differences between CALIOP and the others (20S to 0 latitude) are due to the lidar’s enhanced sensitivity to very thin cirrus relative to the passive instruments. Note that the VCM imager band (I-band) variability tests were not performed because there is no way to simulate these with MODIS radiance data. This could lead to a slight decrease in overall VCM cloud amounts over water surfaces. The VCM over-clouds in Antarctic regions where sea ice and high elevation, snow-covered lands dominate. April 2007 Scene Type VCM / MOD35 C6 vs. CALIOP Agreement 60S-60N Day Ocean 87.8% / 91.4% 60S-60N Day Land 83.1% / 86.7% 60S-60N Night Ocean 84.0% / 90.6% 60S-60N Night Land 81.4% / 87.2% 60S-60N Day Desert 83.5% / 86.5% 60S-60N Night Desert 82.0% / 84.2% Polar Day 82.9% / 84.9% July 2007 Scene Type VCM / MOD35 C6 vs. CALIOP Agreement 60S-60N Day Ocean 85.4% / 91.2% 60S-60N Day Land 85.4% / 88.3% 60S-60N Night Ocean 83.4% / 90.4% 60S-60N Night Land 85.5% / 89.7% 60S-60N Day Desert 86.1% / 88.0% 60S-60N Night Desert 86.5% / 86.3% Arctic Day 84.3% / 85.4% October 2007 Scene Type VCM / MOD35 C6 vs. CALIOP Agreement 60S-60N Day Ocean 90.0% / 92.3% 60S-60N Day Land 85.2% / 89.0% 60S-60N Night Ocean 84.9% / 91.2% 60S-60N Night Land 83.9% / 89.9% 60S-60N Day Desert 86.4% / 90.2% 60S-60N Night Desert 87.0% / 88.3% Polar Day 76.4% / 83.4% Diagnostics With use of collocated CALIOP, MODIS, and VCM cloud mask results, the zonal mean differences seen above may be investigated. The three plots at right show data for pixels where CALIOP and MODIS detected clouds, but the VCM did not, for July 2007 from 0-60N (see above). The oceanic clouds undetected by the VCM are tropical and subtropical low clouds (cloud edges?) and high clouds throughout the 0-60N range. The middle plot shows reflectances for these clouds and MODIS cloud mask thresholds. The plot at far right shows cloud heights for land cases. Diagnostics like these will quickly show areas that need improvement and suggest cloud test threshold changes. Global Maps of Cloud Cover and Comparisons to the MODIS Cloud Mask (MOD35) Shown below are maps of monthly mean cloud fraction (60S-60N latitude), where MODIS L1b and geolocation are used as proxies for VIIRS data. The top two maps show mean cloud fraction for January 2007 (left) and July 2007 (right). The bottom two show VCM minus MODIS (MOD35) cloud fraction for the same months. Differences between VCM and MOD35 are mostly < Exceptions are some arid and high elevation regions, as well as dust-prone areas near the African continent. The MODIS cloud mask has difficulty distinguishing between clouds and dust in the dusty areas. The differences in arid regions will be largely mitigated by a change in a NDVI threshold that defines “desert” in the VCM algorithm. Note that large areas of ocean show differences of up to about (cyan color) that are in opposite hemispheres between the two months (SH in January, NH in July). This is most likely due to MOD35 detecting more clouds than the VCM in sun-glint regions. VCM Cloud Fraction Warm Water Clouds Thin Cirrus Opaque Colder Smoke RGB (0.41-µm; 1.6-µm; 11.0-µm µm) Manually-Generated Cloud Mask VCM Cloud Fraction Main Points 1) VCM validation plans are in place; some methods are shown on this poster and include: spatial and temporal comparisons of monthly mean cloud amounts from other sensors and algorithms, manual analysis of VIIRS imagery and VCM results, pixel-by-pixel collocation with CALIOP, Aqua MODIS radiance data, and Aqua MODIS (MOD35) cloud mask Use of MODIS proxy input data shows: In daytime, the VCM generally detects fewer clouds than other algorithms e.g., PATMOS-x, MODIS (MOD35 and CERES), and ISCCP. Threshold adjustments can be made to find more cloud edges and thin cirrus (1.38 µm cloud test not yet tuned to the VIIRS instrument). The VCM over-clouds in some arid regions. This can be mitigated by tuning the top-of-canopy NDVI threshold that indicates use of the “desert” processing path. The VCM shows comparable quality to MODIS (MOD35) in daytime Arctic and Northern Hemisphere midlatitude winter conditions. The VCM is less developed for nighttime conditions, relative to daytime. Manual Analysis Results from the VIIRS Cloud Mask will also be compared against analyses of cloud fields made by experts using multispectral VIIRS satellite imagery. The figures at left show an example using MODIS data. A false color image is shown at top left, the manually-generated cloud mask is at top right, the VCM results are at bottom left. Note that cloud phase may also be evaluated by use of false color imagery, shown at bottom right. Manual analysis will play a large role during the VCM Intensive Calibration and Validation (ICV) period, scheduled to begin mid-February, Longer term validation and monitoring will necessarily rely more on methods like the others shown on the poster. VCM Cloud Phase Confidently Probably Cloudy Confidently Probably Clear VCM Cloud Confidence VCM Minus MODIS Cloud Fraction January 2007 July 2007 VCM Minus MODIS Cloud Fraction VCM Minus MODIS Cloud Fraction January 2007 July 2007


Download ppt "VIIRS Cloud Mask Validation Exercises"

Similar presentations


Ads by Google