SNAP Data Processors - Idepix MODIS Algorithm Specification

Algorithm Specification

Neural network classification

The Idepix classification algorithm for MODIS is based on a neural network approach. The following MODIS bands are used as input for the neural net:

EV_250_Aggr1km_RefSB.1
EV_500_Aggr1km_RefSB.3
EV_500_Aggr1km_RefSB.4
EV_500_Aggr1km_RefSB.5
EV_500_Aggr1km_RefSB.7
EV_1KM_Emissive.23
EV_1KM_Emissive.25
EV_1KM_RefSB.26
EV_250_Aggr1km_RefSB.1
EV_250_Aggr1km_RefSB.1
EV_250_Aggr1km_RefSB.1

As output, the neural net finally provides per pixel one of the properties 'clear', 'cloud sure', 'cloud ambiguous' or 'snow/ice'.

Brightness and whiteness

In a second step, two additional quantities, the 'whiteness' and the 'brightness', are introduced to derive a 'bright' flag as well as additional cloud indicators:

'cloudSureFromWhiteness'
'cloudSureFromBrightness'
'cloudAmbiguousFromWhiteness'
'cloudAmbiguousFromBrightness'

A bright spectrum means that the intensity of the spectral curve (related to the albedo) should present relatively high values. Therefore, cloud brightness is calculated for each pixel as the integral of spectrum, and differs from the average of the spectral channels since it takes into account the distribution of the energy along the spectrum.

A white spectrum means that the spectral signature must be flat along the spectrum. The first derivative of the spectral curve should present low values, but noise and calibration errors may reduce the accuracy in the estimation of the spectrum flatness when computing the spectral derivative in channels with similar wavelengths.

This retrieval mainly follows the approach described in more detail in [1].

Cloud flagging modes

The final cloud flagging can be performed in two 'strengths', depending on the combination of the three indicators 'cloudFromNeuralNet (cNN)', 'cloudFromWhiteness (cWhi)' and 'cloudFromBrightness (cBri)':

'CLEAR_SKY_CONSERVATIVE': 'cloud' if 'cNN' OR ('cWhi' AND 'cBri')
'CLOUD_SKY_CONSERVATIVE': 'cloud' if 'cNN' OR 'cWhi' OR 'cBri'

Consequently, the 'CLEAR_SKY_CONSERVATIVE' usually results in overall 'less clouds', and the 'CLOUD_SKY_CONSERVATIVE' results in 'more clouds'.

Additional properties

The following additional pixel properties are provided from the classification:

'cloud_buffer': The pixel is located in a 'safety buffer' around an area classified as cloud. The width of the buffer is given in pixels.
'land': the pixel is located over land
'coastline': the pixel is a coastline pixel

The 'land' and 'coastline' pixels are identified from an SRTM land/water mask [2].

Final classification flags

From the algorithm steps outlined above, the following final classification flags are provided for MODIS:

'INVALID' (input invalid or no classification possible)
'CLOUD_AMBIGUOUS'
'CLOUD_SURE'
'CLOUD' (which means either CLOUD_AMBIGUOUS or CLOUD_SURE)
'CLOUD_BUFFER'
'SNOW_ICE'
'COASTLINE'
'LAND'
'BRIGHT'

Known issues

The pixel classification algorithm for MODIS has been developed and optimized mainly for ocean applications, e.g. the OceanColour project, [3]. Therefore, the classification results may be poor under certain conditions over land.

The following pixel properties are currently NOT provided for MODIS:

'CLOUD_SHADOW'
'MIXED_PIXEL'
'GLINT_RISK'

References

[1] Ackerman, S. A., Strabala, K. I., Menzel, W. P., Frey, R. A.,Moeller, C. C., Gumley, L. E., Baum, B., Wetzel-Seeman, S., and Zhang, H.: Discriminating clear sky from clouds with MODIS. Algorithm Theoritical Basis Document (MOD35). Version 6.1, October 2010

[2] Farr, T. G., et al., The Shuttle Radar Topography Mission, Rev. Geophys., 45, RG2004, doi:10.1029/2005RG000183. (2007)

[3] OceanColour Project Web Site: www.esa-oceancolour-cci.org