P2.76
1
Two New Contrail Detection Methods for the Compilation of a Global Climatology of Contrail Occurrence David P. Duda Konstantin Khlopenkov Patrick Minnis Science Systems and Applications, Inc. (SSAI) NASA LaRC P2.76 I. INTRODUCTION One of the recommendations of the Federal Aviation Administration’s (FAA) Aviation-Climate Change Research Initiative (ACCRI) is the development of a global climatology of linear contrail occurrence detectable via satellite remote sensing methods. Such a contrail climatology is necessary for the validation of a new generation of atmospheric models that represent contrail formation explicitly. We present two new automated contrail detection algorithms (CDAs) that can be used to develop this climatology. These CDAs are part of a contrail climatology being developed by NASA LaRC that includes concurrent contrail microphysical and radiative properties such as contrail optical depth, effective particle size and solar and broadband longwave radiative forcing that are derived using the CERES cloud detection and retrieval analyses (Minnis et al., 2008, 2009). II. MODIFIED MANNSTEIN ET AL. CDA The first algorithm is a modified version of the technique described by Mannstein et al. (1999), which detects linear contrails in multi-spectral thermal IR satellite imagery using only two channels (11 & 12 µm) from the Advanced Very High Resolution Radiometer. This method requires only the brightness temperatures (BT) from the IR channels, with no other ancillary data, and can be applied to both day and night scenes. It uses a scene-invariant threshold to detect cloud edges produced by contrails, and 3 binary masks to determine if the detected linear features are truly contrails. However, these masks are not always sufficient to remove all non-contrail edge features. To reduce the number of false positive detections due to lower cloud streets and surface features, we add observations from other infrared radiance channels available on the MODerate-resolution Imaging Spectroradiometer (MODIS) on Terra and Aqua. The new modified method uses additional masks derived from the added thermal infrared channels to screen out linear cloud features that appear as contrails in the original method. III.KHLOPENKOV CDA The second algorithm is based primarily on image analysis and pattern recognition techniques. It uses the BTDs of the MODIS thermal bands producing the highest contrast for the contrails in a particular image. The core of the algorithm consists of several stages of correlation analysis that use predefined linear patterns to match contrails with different widths and orientations. The map of likely contrail vectors obtained from the initial analysis then undergoes a geometry analysis that merges and extends the detected linear fragments by including the pixels missed during the correlation stages. To discriminate contrails from common cirrus streamers, the geometry analysis also suppresses the lines that follow the most dominant local direction. The output contrail mask has different weights for each detected contrail indicating a combined confidence level of the detection. This allows some flexibility for the end user when deciding whether a more strict or more liberal detection is needed. 11-12 μm BT diff. Original CDA Modified CDA Terra MODIS channel 31 (10.8 μm) minus channel 32 (12.0 μm) brightness temperature difference (BTD1) imagery over North Atlantic Ocean (1530 UTC 13 March 2006). Persistent contrails are visible in top part of imagery. Original image (BT8.7 - BT12 μm) aggregated to 2 km/pixel. Apply 5×5 masking to derive the directionality and strength of potential contrail fragments (nodes). Convert original 1 km/pixel image to Laplacian image. For strong nodes, correlate 16×16 block of Laplacian image to closest direction, width and possible offset to assign weight and direction for each node. For given direction and weight of each node, build a map of optimal links between nodes. Different colors show different directions of detected links. The modified CDA uses 6 binary masks to check for contrails 1. N=NBTD2 + N6.8 + N12.0 + NBTD > 2.2 2. BTD1 > 0.2 K 3. Gradient(12.0 μm) < 2×STD(12.0 μm) + 1K 4. (GrW – GrS)/GrS > -0.95 5. NBTD3 > 0.0 6. Gradient(BTD3) < 1.5×STD(BTD3) + 0.2K BTD2 = BT8.5 – BT12 μm BTD3 = BT8.5 – BT13.3 μm GrW = Gradient(BT6.8 μm) + 0.03K GrS = Gradient(BT12.0 μm) + 0.03K Merge and extend contrails by joining chains of links into lines, merging lines that are too close together, and extending lines by including weak nodes also. Discard lines with low weight and limited length. Result after merging and extending. Greens are probable contrails, reds are noise. IV. WAYPOINT DATA ACCRI aircraft emissions data were used to provide the location of commercial aircraft around the globe at 1-hour resolution. To provide a standard geographic grid for the contrail and aircraft data, all locations of the detected contrails and flight tracks were re-gridded to a Lambert Azimuthal Equal-Area projection. As a preliminary test of the new CDAs, the contrail masks produced by the modified Mannstein et al. algorithm have been compared to the global waypoint data provided by the FAA. The waypoint data have also been used as a reference to distinguish thick diffused contrails from natural cirrus clouds. V. SUMMARY Two new contrail detection algorithms (CDAs) have been developed to detect linear persistent contrails from 1-km resolution thermal infrared imagery from MODIS. Both retrievals detect fewer but more reliable contrails than the original Mannstein et al. CDA. The Khlopenkov CDA is an independent contrail detection method that can be used as a check of the modified Mannstein algorithm. Additional datasets including flight track data are being used to improve the accuracy of the contrail detection algorithms. The CDAs form the first step in the development of a satellite-based contrail climatology of cont occurrence and optical properties. VI. FUTURE WORK • Use upper tropospheric wind data from GEOS-4 reanalysis to advect ACCRI flight track data. Compare advected flight tracks with contrail mask from CDA to determine conditions where false positive detections are likely. Use wind and CERES cloud product retrievals to reduce number of false positives. • Verify accuracy and compile detection statistics of CDA (including probability of detection, false alarm rate) by visual inspection of test cases. • Process one year of Aqua and Terra MODIS imagery over Northern Hemisphere for complete climatology. VII. REFERENCES Mannstein, H., R. Meyer, P. Wendling, 1999: Operational detection of contrail from NOAA-AVHRR data. — Int. J. Remote Sensing, 20, 1641-1660. Minnis, P., and Co-authors, 2008: Cloud detection in non-polar regions for CERES using TRMM VIRS and Terra and Aqua MODIS data. IEEE Trans. Geosci. Remote Sens., 46, 3857-3884. Minnis, P., and Co-authors, 2009: CERES Edition-2 cloud property retrievals Using TRMM VIRS and Terra and Aqua MODIS data. IEEE Trans. Geosci. Remote Sens., submitted. To discriminate from common cirrus streaks, suppress the links that follow a common direction (shown in purple). Detected lines. BTD1 image from Terra MODIS 1345 UTC 1 Jan 2006 BTD1 image with Modified Mannstein CDA overlay BTD1 image with Modified Mannstein CDA & ACCRI flight track overlay. Width and color of flight tracks indicates their age relative to over- pass time. The contrail detection masks developed from the CDAs form the first st the contrail climatology process. The contrail masks are then integra the NASA LaRC CERES cloud product retrieval system to derive contrail optical properties for a global contrail climatology. OPTICAL DEPTH & LW RADIATIVE FORCING FROM THEORETICAL CALCULATIONS ASSUMING CONTRAIL TEMPERATURE OF 224K BTD1 image Terra MODIS 13 March 2006 1135 UTC Contrail mask Terra MODIS 13 March 2006 1135 UTC Contrail Broadband LW radiative forcing Terra MODIS 13 March 2006 1135 UTC Contrail visible optical depth Terra MODIS 13 March 2006 1135 UTC Acknowledgement: This research is supported by the ACCRI program.
description
P2.76. Two New Contrail Detection Methods for the Compilation of a Global Climatology of Contrail Occurrence David P. DudaKonstantin Khlopenkov Patrick Minnis Science Systems and Applications, Inc. (SSAI) NASA LaRC. INTRODUCTION - PowerPoint PPT Presentation
Transcript of P2.76
Two New Contrail Detection Methods for the Compilation of a Global Climatology of Contrail Occurrence
David P. Duda Konstantin Khlopenkov Patrick Minnis Science Systems and Applications, Inc. (SSAI) NASA LaRC
P2.76
I. INTRODUCTIONOne of the recommendations of the Federal Aviation Administration’s (FAA)Aviation-Climate Change Research Initiative (ACCRI) is the development of a global climatology of linear contrail occurrence detectable via satellite remote sensing methods. Such a contrail climatology is necessary for the validation of a new generation of atmospheric models that represent contrail formation explicitly.
We present two new automated contrail detection algorithms (CDAs) that can be used to develop this climatology. These CDAs are part of a contrail climatology being developed by NASA LaRC that includes concurrent contrail microphysical and radiative properties such as contrail optical depth, effective particle size and solar and broadband longwave radiative forcing that are derived using the CERES cloud detection and retrieval analyses (Minnis et al., 2008, 2009).
II. MODIFIED MANNSTEIN ET AL. CDAThe first algorithm is a modified version of the technique described by Mannstein et al. (1999), which detects linear contrails in multi-spectral thermal IR satellite imagery using only two channels (11 & 12 µm) from the Advanced Very High Resolution Radiometer. This method requires only the brightness temperatures (BT) from the IR channels, with no other ancillary data, and can be applied to both day and night scenes. It uses a scene-invariant threshold to detect cloud edges produced by contrails, and 3 binary masks to determine if the detected linear features are truly contrails. However, these masks are not always sufficient to remove all non-contrail edge features. To reduce the number of false positive detections due to lower cloud streets and surface features, we add observations from other infrared radiance channels available on the MODerate-resolution Imaging Spectroradiometer (MODIS) on Terra and Aqua. The new modified method uses additional masks derived from the added thermal infrared channels to screen out linear cloud features that appear as contrails in the original method.
III. KHLOPENKOV CDAThe second algorithm is based primarily on image analysis and pattern recognition techniques. It uses the BTDs of the MODIS thermal bands producing the highest contrast for the contrails in a particular image. The core of the algorithm consists of several stages of correlation analysis that use predefined linear patterns to match contrails with different widths and orientations. The map of likely contrail vectors obtained from the initial analysis then undergoes a geometry analysis that merges and extends the detected linear fragments by including the pixels missed during the correlation stages. To discriminate contrails from common cirrus streamers, the geometry analysis also suppresses the lines that follow the most dominant local direction. The output contrail mask has different weights for each detected contrail indicating a combined confidence level of the detection. This allows some flexibility for the end user when deciding whether a more strict or more liberal detection is needed.
11-12 μm BT diff. Original CDA Modified CDATerra MODIS channel 31 (10.8 μm) minus channel 32 (12.0 μm) brightnesstemperature difference (BTD1) imagery over North Atlantic Ocean (1530 UTC13 March 2006). Persistent contrails are visible in top part of imagery.
Original image (BT8.7 - BT12 μm)aggregated to 2 km/pixel.
Apply 5×5 masking to derive thedirectionality and strength of potential
contrail fragments (nodes).
Convert original 1 km/pixel image to Laplacian image.For strong nodes, correlate 16×16 block of Laplacianimage to closest direction, width and possible offset to
assign weight and direction for each node.
For given direction and weight of eachnode, build a map of optimal links
between nodes. Different colors showdifferent directions of detected links.
The modified CDA uses 6 binary masks to check for contrails1. N=NBTD2 + N6.8 + N12.0 + NBTD > 2.22. BTD1 > 0.2 K3. Gradient(12.0 μm) < 2×STD(12.0 μm) + 1K4. (GrW – GrS)/GrS > -0.955. NBTD3 > 0.06. Gradient(BTD3) < 1.5×STD(BTD3) + 0.2K
BTD2 = BT8.5 – BT12 μmBTD3 = BT8.5 – BT13.3 μmGrW = Gradient(BT6.8 μm) + 0.03KGrS = Gradient(BT12.0 μm) + 0.03K
Merge and extend contrails by joining chains of links into lines, merging lines that are too close together, and extending lines by including weak nodes also.
Discard lines with low weight and limited length.Result after merging and extending. Greens are probable contrails, reds are noise.
IV. WAYPOINT DATAACCRI aircraft emissions data were used to provide the location of commercialaircraft around the globe at 1-hour resolution. To provide a standard geographicgrid for the contrail and aircraft data, all locations of the detected contrails andflight tracks were re-gridded to a Lambert Azimuthal Equal-Area projection.
As a preliminary test of the new CDAs, the contrail masks produced by themodified Mannstein et al. algorithm have been compared to the global waypointdata provided by the FAA. The waypoint data have also been used as a referenceto distinguish thick diffused contrails from natural cirrus clouds.
V. SUMMARYTwo new contrail detection algorithms (CDAs) have been developed to detectlinear persistent contrails from 1-km resolution thermal infrared imagery fromMODIS. Both retrievals detect fewer but more reliable contrails than theoriginal Mannstein et al. CDA. The Khlopenkov CDA is an independentcontrail detection method that can be used as a check of the modified Mannsteinalgorithm. Additional datasets including flight track data are being used toimprove the accuracy of the contrail detection algorithms. The CDAs form thefirst step in the development of a satellite-based contrail climatology of contrailoccurrence and optical properties.
VI. FUTURE WORK• Use upper tropospheric wind data from GEOS-4 reanalysis to advect
ACCRI flight track data. Compare advected flight tracks with contrail mask from CDA to determine conditions where false positive detections are likely. Use wind and CERES cloud product retrievals to reduce number of false positives.
• Verify accuracy and compile detection statistics of CDA (including probability of detection, false alarm rate) by visual inspection of test cases.
• Process one year of Aqua and Terra MODIS imagery overNorthern Hemisphere for complete climatology.
VII. REFERENCESMannstein, H., R. Meyer, P. Wendling, 1999: Operational detection of contrailsfrom NOAA-AVHRR data. — Int. J. Remote Sensing, 20, 1641-1660.
Minnis, P., and Co-authors, 2008: Cloud detection in non-polar regions forCERES using TRMM VIRS and Terra and Aqua MODIS data. IEEE Trans.Geosci. Remote Sens., 46, 3857-3884.
Minnis, P., and Co-authors, 2009: CERES Edition-2 cloud property retrievalsUsing TRMM VIRS and Terra and Aqua MODIS data. IEEE Trans. Geosci.Remote Sens., submitted.
To discriminate from common cirrusstreaks, suppress the links that follow
a common direction (shown in purple).
Detected lines.
BTD1 imagefrom Terra MODIS
1345 UTC1 Jan 2006
BTD1 image withModified MannsteinCDA overlay
BTD1 image withModified MannsteinCDA & ACCRI flight track overlay. Widthand color of flighttracks indicates theirage relative to over-pass time.
The contrail detection masks developed from the CDAs form the first step inthe contrail climatology process. The contrail masks are then integrated intothe NASA LaRC CERES cloud product retrieval system to derive contrailoptical properties for a global contrail climatology.
OPTICAL DEPTH & LW RADIATIVE FORCING FROM THEORETICAL CALCULATIONS ASSUMING CONTRAIL TEMPERATURE OF 224K
BTD1 image
Terra MODIS13 March 20061135 UTC
Contrail mask
Terra MODIS13 March 20061135 UTC
Contrail BroadbandLW radiative forcing
Terra MODIS13 March 20061135 UTC
Contrail visibleoptical depth
Terra MODIS13 March 20061135 UTC
Acknowledgement: This research is supported by the FAAACCRI program.
