improvements to the caliop surface detection algorithm ......s = 29.5, h= 0.55 ice 1 0.031 2s h...
TRANSCRIPT
Improvements to the CALIOP Surface Detection AlgorithmMark Vaughan, Kam-Pui Lee, Anne Garnier, Brian Getzewich
Surface Returns in Multi-Shot Averages
At 532 nm, backscatter from theEarth’s surface is spread overthree contiguous range bins, withthe peak signal occurring in eitherthe first (uppermost) or second ofthese bins. Because the 1064 nmdata is averaged to 60-m on-board, the 1064 nm can extendover 4 bins
As shown in the line plots below,we take advantage of the con-sistent geometry of the surfacereturns by using a derivative testto locate the onset of the surfacesignal. Derivatives are computedusing a two-point backward differ-ences applied to the attenuatedbackscatter coefficients.
Surface Returns At Single-Shot Resolution
Signal at 4.0°S Derivatives at 4.0°S Signal at 1.0°S Derivatives at 1.0°S Signal at 12.6°N Derivatives at 12.6°N Signal at 28.7°N Derivatives at 28.7°N
i i 1
i i i 1
ddZ Z Z
a) Use a digital elevation map (DEM) to define a vertical surface search region
b) Compute the scaled RMS background signal (i.e., convert the background signal into a pseudo-attenuated backscatter coefficient)
c) Compute derivatives of the attenuated backscatter profiles in the surface search region
d) Determine minimum and maximum derivatives and peak signal value in the search region
e) Apply tests to both the signal and the derivatives to see if the surface return can be reliably detected
TEST 1 : is the minimum derivative altitude higher than the maximum derivative altitude?
TEST 2 : is the peak signal greater than TP,l times the scaled RMS background, where TP,l is aconfigurable runtime constant (i.e., does the maximum signal rise significantly above thebackground noise)?
TEST 3 : is the distance between the minimum and maximum derivatives less than Nl range bins?
TEST 4 : are the minimum derivative altitudes at 532 nm and 1064 nm within 2 or fewer ranges bins ofeach other?
V4 Surface Detection Procedure
Rugged Terrain
Ocean Surfaces
Test 2: Thresholding The figureto the right illustrates the relationbetween the value of TP,l (dashedline) and the overlying integratedattenuated backscatter (γ′) abovethe surface. γ′ for an opaque layerdepends on lidar ratio; e.g., γ′ forice clouds and water clouds differsby a factor of 2. The attenuationof the surface peak depends onoptical depth, not γ′, so the sur-face can be detected under a widerange of γ′ values.
Performance Metrics: V3 vs. V4The blue line in the figure to the leftshows the frequency of V4 sur-facedetections relative to V3 as a functionof overlying γ′. For rea-sonably clearskies (γ′ < 0.02 sr -1), the V3 and V4algorithms perform equally well, asindicated by the detection ratio of 1.However, in turbid scenes with highoverlying γ′, the V4 algorithm performssub-stantially. Likewise, in totally opa-que situations, V4 reports fewer falsepositives (ratio < 1 for γ′ > 0.085)
V3 Failure V3 Success
V4 Failure 0.3481 0.0047
V4 Success 0.1503 0.4969
i i 1
i i i 1
ddZ Z Z
i i 1
i i i 1
ddZ Z Z
TP,l times scaled RMS background
28.0°S120.0°W
30.5°S120.7°W
33.0°S121.3°W
35.5°S122.0°W
38.0°S122.8°W
40.5°S123.6°W
43.0°S124.4°W
45.5°S125.3°W
48.0°S126.2°W
Performance Example : 2007-08-27T08-59-28ZN
V3 SurfaceDetection
V4 SurfaceDetection
Depending on the terrain, multi-shot averaging can smear the discrete single shot surface return over multiple range bins that may not always be contiguous, and thus a different detection scheme is require
Procedure
a) Use single-shot detectionresults to define a verticalsurface search region
b) Compute the scaled RMSbackground signal
c) If over ocean, locate surfacereturn using the derivativetechnique
d) Otherwise, define surface topas the first (i.e., highest) pointin the search region for whichthe attenuated backscattercoefficient exceeds TP,l timesthe scaled RMS background.The surface extent is definedby the last (lowest) point thatexceeds this threshold.
Downstream Data Product BenefitsSingle Layer Opaque ROI August 2013
t < 2
t < 1
IIR Error Mitigation: the distribution ofeffective emissivities of V3 “opaque” cirruswith backscatter centroid altitudes above 7km and centroid temperatures colder than-35°C shows that ~6% of the layers have
effective emissivities smaller than 0.63 (avisible optical depth of ~2). For ~1% ofthe layers, the effective emissivity issmaller than 0.4, equivalent to an opticaldepth less than ~1. 7.5% of the layershave a value of γ′ smaller than 0.022 sr-1,but only 33% of those layers exhibiteffective emissivities smaller than 0.63.
Lidar Ratio Retrievals: seethe poster by Young et al.,“CALIOP’s V4 Algorithmsfor Retrieving OpticalProperties of Opaque IceClouds”.
Accurate surface detectionis essential for generatingaccurate estimates of lidarratio in opaque layers.
Opaque Water CloudsS = 18.9, h = 0.42
2H O
10.063
2 S
h
Opaque Ice CloudsS = 29.5, h = 0.55
ice
10.031
2 S
h
“clear air”
clear 0.011
DEM Errors(from GTOPO30)
TP,l = 24
Comparing the V4 and V3 Algorithms
Poster presented at the CALIPSO-CloudSat Science Team Meeting held 1-3 March 2016 in Newport News VA