Polar surface and cloud classification 

Product Description

This Level 2 product is a polar classification map. The polar regions are defined as all earth surfaces lying pole-ward from 60 N or 60 S. The algorithm classifies each pixel of a scene into 8 classes: water cloud, ice cloud, aerosol/dust, water, land, snow/ice, slush ice, and shadow. This product is produced at 30-m spatial resolution and uses a combination of visible, near-infrared and infrared channels. Both daytime and nighttime classifications will be available, with the daytime algorithm applied for solar zenith angles less then 85 degrees, and the nighttime algorithm applied in all cases using only thermal infrared channels. Only the daytime algorithm is available at this time. This is an on-request product.

This data product builds on work over the past decade with imagery taken from LANDSAT TM, AVIRIS, TIMS, and MAS satellite and aircraft sensors. The improved spectral coverage, resolution, and radiometric accuracy enables ASTER to provide remotely sensed polar data of unprecedented accuracy.

Algorithm Description

The methodology implemented in this algorithm is a neural-network classifier. The neural network technique is chosen because of its advantages in accuracy and speed when compared to more conventional techniques such as maximum likelihood. To classify a scene, additional spectral features first are generated from the original channels. Several of these additional features are nonlinear combinations of the original channels, which are constructed to improve separability between classes. Classification is then performed using a trained neural network classifier with as input an optimal subset of all features.

Applications

Since greenhouse forcings are expected to be amplified in the polar regions, the changes in these regions may act as early warning indicators of global climate shifts. Cloud cover is expected to be altered by greenhouse forcings, and cloud changes are expected to have a significant effect on sea ice conditions and regional ice-albedo feedbacks, especially to the polar heat balance which directly affects surface melting. ASTER polar data will be used to monitor changes in surface conditions, notably temperature, albedo, and sea ice breakup.

This ASTER polar data product also provides complementary validation to the global-scale polar data and cloud property retrievals from MODIS. In particular, this data set will be used to cross-validate the MODIS cloud optical thickness and effective particle sizes which directly impact the Earth's radiative budget. In addition, only ASTER has the spatial resolution necessary to fully analyze the three-dimensional effects of clouds.

Constraints

The daytime classification algorithm expects as input all the ASTER channels. Furthermore, the algorithm expects the gain selections for VNIR and SWIR channels to be both "LOW". The algorithm has not been applied to data with other gain selections.

The nighttime algorithm utilizes only the thermal infrared channels as input. Furthermore, the nighttime algorithm is developed by using only thermal infrared channels of daytime images. This is due to the fact that the human experts are less confident in classifying thermal infrared images by visual inspection. Therefore, daytime images are used to create training data sets for the classifier.