A support vector machine model for determining radiative heating profiles

Knowledge of the vertical structure of radiative heating is important for understanding the distribution of energy within the atmosphere, which has a direct impact on many key local and large scale dynamic atmospheric processes. Currently, detailed information on vertical radiative heating profiles can only be derived from instruments measuring vertical cloud property profiles in the atmosphere. Such instruments exist only at a limited number of fixed observational sites (such as the radar and lidar instruments at Atmospheric Radiation Measurement (ARM) program sites) or on polar orbiting satellite platforms which only have twice daily observations. By modeling the relationship between the ARM surface based measurements and geostationary satellite measurements, knowledge of the vertical structure from ARM measurements can be expanded to a much larger spatial scale, allowing retrievals of representative diurnal heating rate profiles from geostationary satellite observations.

A set of states of the radiative heating rate profiles is defined based on distinct patterns of cloud profiles and heating rates in the observational ground data. Validation is done to ensure that the states are both statistically consistent and physical meaningful. We then create a support vector machine model that links observed geostationary satellite data to the set of states allowing 1) the retrieval of a representative heating rate profile for a satellite observation and 2) an understanding of the relationship between the satellite measurements variables and the ground based measurement variables.