12
An Automated, Dynamic Threshold Cloud Detection Algorithm Jian Liu and Jianmin Xu National Satellite Meteorological Center China Meteorological Administration

An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

  • Upload
    others

  • View
    7

  • Download
    0

Embed Size (px)

Citation preview

Page 1: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

An Automated, Dynamic Threshold Cloud Detection

Algorithm Jian Liu and Jianmin Xu

National Satellite Meteorological CenterChina Meteorological Administration

Page 2: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

An Automated, Dynamic Threshold Cloud Detection Algorithm

1. Data Set2. Cloud detection method3. Discussion

Page 3: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Data Set

FY-2C data

0.55-0.90Visible VIS

3.5-4.0Near infraredIr4

6.5-7.3Water vaporIR3

11.5-12.5Infrared splitIR2

10.3-11.3Far infrared IR1

WAVELENGTH(μm)

CHANNEL NAME

CHANNEL ID

Page 4: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Cloud Detection Method

1. Getting dynamic cloud detection threshold for 32*32 pixels area by histogram analysisDifferent surface typeDEM modify

2. Cloud detection threshold validationCurve fit to check threshold

3. Cloud detectionDynamic thresholdMulti-channel data

Page 5: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Dynamic Threshold

450 500 550 600 650 700 7500

20

40

60

80

100

120

140

Num

ber o

f Pix

els

(331-BT)*5

C

400 450 500 550 600 650 700 750 800-10

0

10

20

30

40

50

60

70

80

Num

ber o

f Pix

els

(3310-bt)*5

Peak

Dynamic Threshold

original histogram analysis smoothed histogram

area average processing

Page 6: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Land

Mask

Page 7: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Using DEM data to modify dynamic threshold

Page 8: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

threshold validation_update dynamic threshold

0 2 4 6 8 10 12 14 16 18 20 22

265

270

275

280

285

BT

Hour

Land Gauss fit

0 2 4 6 8 10 12 14 16 18 20 22250

255

260

265

270

275

280

285

290

BT

Hour

Desert Gauss fit

* Dynamic Cloud detection threshold

Page 9: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Cloud Detection

1. Dynamic threshold method 2. Visible channel data 3. Deviation analysis4. Relationship analysis _Brightness difference

infrared & water vapor channelinfrared split channel

4. Multi-day composed brightness temperature

Page 10: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Case study

CLD image

Page 11: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Discussion

1. The algorithm performs well for most area.2. In high latitude regions, the cloud detecting

methods failed sometimes due to strong surface temperature inversions.

3. Some surface conditions may make this approach inappropriate, most notably over snow and ice condition.

4. Some cloud types such as thin cirrus, low stratus at night, and small cumulus are difficult to detect because of insufficient contrast with the surface radiance.

Page 12: An Automated, Dynamic Threshold Cloud Detection …...Data Set FY-2C data VIS Visible 0.55-0.90Ir4 Near infrared 3.5-4.0 IR3 Water vapor 6.5-7.3 IR2 Infrared split 11.5-12.5 IR1 Far

Thanks