GISdevelopment.net ---> AARS ---> ACRS 1992 ---> Poster Session R

Estimation of cloud type, its amount and precipitation area using NOAA – AVHRR data

M. Bayasgalan, M. Erdenetuya
National Remote Sensing Centre, Mongolia Ulaanbaatar – 11


Abstract
NOAA-AVHRR digital gave a possibility to estimate cloud types, amounts and precipitation area over continuous large space within short time and with less expenditure, based on digital image data processing methods of thresholding, multivariate statistical analyzing and supervised classification. The results of this work are used for environment monitoring, especially in field of weather forecasting, expansion of the ground observation data range in terms of time and space.

Introduction
One of the main factor, affected on natural environment state is the climate, that is determined by dynamically changeable weather. Therefore the system intended for natural environment monitoring should include weather diagnosis analysis and forecasting components.

The weather characteristics, especially the cloud coverage and precipitation area are very changeable and the ground observation discrete data from few meteorological stations are not enough for their full estimation.

The remotely sensed data from meteorological satellite NOAA can provide very important and useful information in meteorology, mainly in field of weather forecasting. The satellite data is much significant for these purposes, due to its continuity, repetition and wide coverage area of observation.

Since, 1988 we are receiving and processing multi channel AVHRR data from NOAA satellite and developing some methods to estimate cloud coverage and to detect precipitation area using the radiance temperature, spectral albedo and some cloud statistical parameters. In this paper we have briefly introduced the methods and some results of our work.

Data and Methodology
We have used NOAA-AVHRR and ground meteorological observation data over territory of Mongolia. We have developed methods for estimation of cloud amounts, cloud types and delineation of precipitation area.

1 Cloud amounts
The recognition of cloudness is most important task in its estimation. It can realized using following methods:
  1. Threshold method : It is used the radiance temperature of channel 4 (T4) and spectral albedo of channel 2(A2).
  2. Discriminant analysis method : In this method for cloud recognition, the ground observation data has been selected as source data and based n discriminant function there is extracted cloudness over those territory where ground measurement data is absent or rare.
  3. Supervised classification : In this method the training area was selected interactively or such area where has ground observation cloud data.
After recognition of cloud we calculated its amounts using following simple expressing as ratio of cloudy and non-cloudy pixel numbers:

N = P / Pn * 10 (1)

Where,
P is a number of cloud pixel (one pixel has 1.1 sq. km resolution)
Pn is a square of an area, where calculated cloud amount (40 sq. km)
    2 Cloud types
    Different kind of clouds are expressed with their own characteristics in each channel of AVHRR data. According to many properties of cloudness have been interactively determined cloud type and picked out their corresponding radiance temperature of channel 3 and 4 (T3, T4), spectral albedo of channel 2(A2) and their deviations (Sigma).

    3 Precipitation area M
    In this work we have used the cloud top radiance temperature, derived from AVHRR data for 3 and 4 channels and also cloud types and amounts as predictors.

    Result and Discussions

    1 Cloud Amounts

    1. Threshold method : Based on this method have been selected the threshold values of -30°C, in winter season and -5°C in Summer season for radiance temperature of channel 4 (T4). Threshold values for spectral albedo of channel 2 (A2) are 30 for winter and 20 for summer.
    2. Discriminant analysis method : In this method we have selected T3 and T4 as bets predictors and used following discriminant function :

      D cl = 0.054 * T3 – 0.174 * T4 – 1.46 (2)

      Where, T3 and T4 are values of radiance temperature for channel 3 and 4

    3. Supervised classification method : This method was more suitable when it is impossible to use threshold and discriminant methods for estimation cloud especially for winter time, when differences between cloud top temperature and cold surface temperature became less, even when cloud became that landcover. The final result shows that 83.8% of case-study for cloud amount had differences of 1 - 2 value between calculated and observed parameters and 8% of its has 5 value of differences. The main errors are observed in case of micro-scale cumulus (Cu) cloud.
    2 Cloud type
    Every cloud type has different kind of properties, for example, the thick cumulus nimbus (Cb) cloud has a high reflectance value in each channel, clear-cut edge, small square and low (less than -50°C) radiance temperature (T rad). The T rad value of cumulus (Cu) cloud ranges from -10° to -35°C and has high spatial variation. But stratus (St) cloud has T rad value of (-5°C) – (-25°C), grey-white color and low spatial variation. In Fig. 1 – 4 are shown examples of source image data, cloud amounts and types, estimated by above mentioned methods over territory of Mongolia, on 1 August 1992. Using statistical values of cloud top temperature, albedo their spatial and temporal variations and cloud amounts have been classified some types of cloud. The strato-cumulus (Sc), strong cumulo-nimbus (Cb) and stratus (St) types of clouds are classified successfully by these methods (83.7% its accuracy), but classification results of cirrus (Ci), cumulus (Cu) and very low layer clouds are not good.








    3 Precipitation area :
    We have determined such main precipitated cloud structures as wave in front, cloud stream of cold front and the cloud vortex, in which delineated the precipitation area, occupying central part of cloud stream, and head part of cloud vortex. Most of long time continuous precipitation are observed in case of cloud vortex (55%) and cloud stream (34%). Fig. 2 is shown the delineated precipitation are on that date.

    Also in this work it was analysed the relationships between rainfall and cloud type, rainfall and cloud amount. (Table 1 and 2). The precipitation probability is increased with increase of cloud amount.

    Table 1 The relationship between precipitation and cloud amounts Cloud amounts
    Cloud amounts Number of examples
    With precipitation Without precip.
    9 – 10
    5 – 8
    0 – 4    		 151
    9
    -    		 42
    17
    101

    The results show, that in summer season the delineation method of precipitation area is more suitable than in winter season, especially for convective rainfall. Its accuracy for summer is 83 – 91%, but in winter time the precipitation area of small scale cloud or lower layers’ cloud is delineated unsuccessfully.

    Table 2 Number of precipitation reiteration in different types of cloud
    Cloud type Number of example
    Cb
    Sc
    Ns
    Ac, Sc
    St    		 70
    62
    29
    6
    2

    The results of the study are presented in the form of :
    1. Classified image on the screen with plotting of cloud amounts, cloud types, cloud top temperatures, precipitation probabilities and with delineation of cloud precipitation area.
    2. Hardcopy of maps with above mentioned parameters and it is used for operative work.
    3. Photo slides, which can be used for other scientific purposes.
    Conclusion
    1. Multichannel NOAA-AVHRR data can be practically and effectively utilized in weather forecasting on large territory, especially in case of study for cloud coverage and precipitation area.
    2. Threshold method is more suitable for summer season, when the difference between land and cloud temperature is quite more.
    3. Discriminant analysis method is more convenient in that cases when it is difficult to use threshold method of cloud estimation and when there are clouds distributed over large space.
    Reference
    1. Afifi, L, Eizsh.S.Statistical analyses. Moscow, 1982.
    2. Enclein. K, Pleston.E. Statistical methods for computers. Moscow 1986.
    3. Bayasgalan. M, Erdenetuyra.M. Estimation of cloud type and amount by NOAA-AVHRR data. journal of Mongolian Hydrometeological research Institute, Ulaanbaatar – Mongolia Vol 15.