md anisul islam geoscience australia
DESCRIPTION
Australian Government Geoscience Australia. Evaluation of IMAPP Cloud Cover Mapping Algorithm for Local application. Md Anisul Islam Geoscience Australia. Objective of the study: - PowerPoint PPT PresentationTRANSCRIPT
Geoscience Australia
Md Anisul Islam
Geoscience Australia
Evaluation of IMAPP Cloud Cover Mapping Evaluation of IMAPP Cloud Cover Mapping Algorithm for Local application. Algorithm for Local application.
Australian Government
Geoscience Australia
Geoscience Australia
Objective of the study:
Evaluation of IMAPP Cloud Cover Mapping algorithm (Collection 4 & 5) for local application to utilise the cloud mask for generating higher level Land application products.
Geoscience Australia
Ancillary data
Ecosystem file Land/water mapDEMDaily snow/ice SSTGlobal data assimilation system (GDAS1)
Image Pixel
Labelling the pixels to Surface types
Water Land Desert Snow/Ice
10 Spectral Cloud tests
Group 1 Group 2 Group 3 Group 4 Group 5
Initial Cloud mask obtained from Clear sky conservative approach
Final Cloud mask
Clear Sky Restoral tests
Sunglint regionShallow waterDesert & Land
Cloudy Pixels
Schematic diagram of IMAPP MODIS Cloud Cover Algorithm
Clear Pixels
Geoscience Australia
Group 1 (IR bands):
BT11, BT13.9 & BT6.7
Group 2 (Thermal band differences):
Trispectral TestBT11- BT12, BT11 – BT3.9
Group 3 (Reflective bands):
R0.66 or R0.87 & R0.87/R0.66
Group 4 (NIR thin Cirus):
R1.38
Group 5 (IR thin Cirus): BT3.7 – BT12
Clear sky conservative approach of Initial cloud mask generation
Gi=1,5 = min[Fi] & Initial Cloud mask confidence level = (Product of Gi=1,5)-5
Where, G = Group
Three thresholds used in cloud screening
Confidence Level (F): > 0.95 & ≤ 0.99 - Uncertain clear≥ 0.66 & ≤ 0.95 - Uncertain cloud
Geoscience Australia
Difference between Collection 4 and Collection 5 Cloud Mask Algorithms
Surface type Labelling scheme:
Pixels belonging to Ecosystems: Savanna (Woods), Hot & Mild Grasses andShrubs, Woody Savana are labelled as surface type Land in Collection 4 algorithm, are labelled as Desert for Australian Continent (latitude 11.0 - 40.0 (S)& longitude 110.0 -155.0 (E)) in Collection 5 algorithm.
Major Effects on the above pixels in Collection 5 algorithm are :
Subject to different threshold values (for Desert: -20, -18, -16) of BT11-3.9 spectral test in Collection 5 algorithm as against threshold values of Land (-14, -12, -10) in Collection 4 algorithm
Subject to spectral test of Desert (R0.87) applied in Collection 5 algorithm asagainst spectral tests of Land (R0.66 & R0.87/R0.66 ) applied in Collection 4algorithm
Geoscience Australia
Threshold values:
For BT13.9 test are 224, 226 and 228 (Kelvin) for Collection 4 algorithm and are 222, 224 and 226 (Kelvin) for Collection 5 algorithm
For R0.87/R0.66 test are 0.55, 0.40 & 0.30 for Water in Collection 4 algorithm and are 0.65, 0.55, & 0.45 in Collection 5 algorithm
Additional ancillary data used in Collection 5 Cloud Mask Algorithm:
NOAA optimum Interpolation (OI) Sea Surface Temperature (SST) V2 product at 1 degree resolution – helps to improve the cloud mask over ocean at night.
Global assimilation system (GDAS1) for retrievals of temperature and moisture profile – helps to improve the cloud mask over many areas of
the land.
Geoscience Australia
Full swath images acquired from the Orbit covering maximum Land areas of Australia
Monthly MODIS image acquired between September 2004 to April 2005
Acquisition date Orbit no
2 Sept 2004 25043
20 October 2004 25742
21 November 2004 26208
23 December 2004 26674
9 February 2005 27373
30 April 2005 28538
MODIS DATA ASSESSED IN THE PROJECT:
MODIS TERRA images capturing high temporal and spatial variation of Land areas of Australia
Geoscience Australia
Visual assessment of final Cloud Mask using:
RGB of visual bands Gray scale image bands used in the spectral test
Spectral Analysis of data from sample sites in areas where there are errors as determined by visual assessment
Convertion of image band Digital numbers (DN) to the units of the thresh values:
Convertion of reflective image band DN to reflectance unit & Thermal image band DN to Kelvin
Extraction of sample data from the images
Scatterplots of the bands used in spectral tests having errors versus sample site attributes to determine the amount of errors.
Methodology:
Geoscience Australia
Relective bands 1 4 3 Inverse BT11 – BT3.9
Collection 4 Cloud Mask Collection 5 Cloud Mask
Confident Clear Probably Clear Uncertain Cloudy
20 October 2004
Geoscience Australia
21 November 2004
Relective bands 1 4 3 Relective band 26 (R1.38 Thin Cirus test)
Inverse BT11 – BT3.9 Confident Clear Probably Clear
Uncertain Cloudy
Geoscience Australia
Confident Clear Probably Clear
Uncertain Cloudy
Relective bands 1 4 3
23 December 2004
Relective band 26 (R1.38 Thin Cirus test)
Inverse BT11 – BT3.9
Geoscience Australia
Confident Clear Probably Clear
Uncertain Cloudy
Relective bands 1 4 3
9 February 2005
Relective band 26 (R1.38 Thin Cirus test)
Inverse BT11 – BT3.9
Geoscience Australia
Confident Clear Probably Clear
Uncertain Cloudy
Relective bands 1 4 3
2 September 2004
Relective band 26 (R1.38 Thin Cirus test)
Inverse BT11 – BT3.9
Geoscience Australia
Spectral plots of the cloud free samples taken from Multitemporalimages.
THCCLD - threshold corresponding to HighConfident Cloudy pixels (α)
TV - threshold value for pass or fail (β)
THCCLR - threshold corresponding to High Confident Clear pixels (γ)
Geoscience Australia
Conclusions and Recommendations
Collection 5 cloud mask significantly reduces misclassification of the clear pixels as cloudy pixels relative to Collection 4 Cloud mask. However, collection 5 cloud mask may detect patchy Low cloud as uncertain cloud and uncertain clear pixels (October 04).
Collection 5 cloud mask may be highly sensitive in detecting High thin clouds (November 04, December 04), which may be better evaluated using additional data.
Collection 5 cloud mask may have small errors of misclassifying clear pixels as cloud in the surface type Land (February 05), which may be attributed to BT11-BT3.9 test. Modification of the threshold values of this test may further improve the cloud mask.
Clear pixels of bright desert and salt lake may be misclassified as uncertain clear pixels and cloudy pixels (Sep 04 & April 05), which may be attributed to Reflectance R0.87 test. Modification of threshold values of R0.87 test and or BT11 threshold values used in the clear sky restoral test may be modified to improve the results.