International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 Intern 
  
  
it will provide the real-time data that was required. NOAA-12 knowledge-free automatic segmentation of the imagery, this | input 
local area coverage (LAC) data was acquired at our local showed a high contrast and allowed for better evaluation of | classif 
ground receiving station November 20, 2003 in monsoon brightness temperature and cloud heights. The segmentation | object 
season, as it will provide better understanding into rain baring technique of eCognition creates a hierarchical network of land, | proces 
clouds. sea and cloud objects in different scales, which represents the | be ver 
image information. The classified clouds show attributes that | object 
AVHRR data channels 1.2 and 4 were used to visually correlate to their shape information, spectral statistics and | impro 
differential clouds based on the height and shape (Fig 4a). Five relations to nearest objects. Only level 1 classification. was | 
classes of features representing represent the land, sea, low performed using the nearest neighbour sampling. Due the 5 Refe 
stratus cloud, mid altos cloud, and high cirrus clouds (Fig.4b) limited spatial resolution. of AVHRR data level 2 and 3 | 
using eCognition object-oriented image classification software classification could not be performed as the involve sub object | Anagni 
to differential image object. Class objects were extracted using classification. | satellite 
the infrared thermal channel 3,4,5 of the AVHRR data by : | studies 
| Arkin, 
coveras 
| GATE 
Arkin, 
distribu 
Ba, M. 
| activity 
| lakes of 
| channel 
| 
| Bendix, 
| Techniq 
distribu: 
| Int. J. R 
| 
| Bendix, 
| Norther 
perspect 
| Gruber, 
| convecti 
| Definien 
55 [Low Stratus<2000 [ Mia Altos 2000-6000 [mem |High Cirrus 6000-1800m | useclass; 
2, No. 6, 
Fig 4 Cloud Type Identification and Classification | .  httpz//wy 
| 
; : : : : - 3 2 | Karlsson 
Bi-spectral techniques based on the relationship between cold enhancements and filters are then applied to improve the visual | Resolut 
and brightness temperature of clouds were used to evaluate interpretation of the clouds. Rainfall is estimated based on the | S So ni 
ae D . i T Da | Swedish 
precipitation probability. The NIR and IR channels 3,4,and 5 of assumption that every cloud pixel has a constant unit rain-rate | 601 76 N 
the data were processed for temperature and brightness. In an of 3mm*', which is appropriate for tropical precipitation over 
infrared (IR) image cold clouds are high clouds, so the colors 2.5? x 2.5? areas around the equator. Total cold cloud cover Todd. M 
typically highlight the colder regions Mid height clouds with Tp and the portion of the catchment covered by cloud determines | ] 1999 E 
below 235k were identified as cumulonimbus cloud with a rainfall intensity. | basi . 
high probability to precipitate. Lower probabilities were | GST 
| technique 
associated to warm but bright stratus cloud and thin cirrus 
cloud that were cold but dull. The study area shows warm but 3 Conclusion ToddsM 
bright non-precipitating stratus cloud (Fig 4c). 1995 
Cloud classification was performed using feature object uoo: Sal 
oriented techniques on NOAA AVHRR data. Although the ver has 
temperatu 
Cold clouds with temperature below 235k threshold value are 
taken as indication of rain. The intensity and dimensions of rain 
bearing pixels were measured after a mask has been applied 
clouds have been highlighted (Fig 4d). Various 
coarse resolution of 1.1km of AVHRR data did not allow for 
high-level classification, object orient classification however 
and proved effective for the cloud type identification due to the 
722