Using optical flow temporal interpolation of satellite imagery to assist multi-sensor global cloud product composites

Large domain cloud-product composites, important for climate research and
operations in civil and military aviation, maritime, and renewable energy
management, require the synthesis of products retrieved from different
satellites scanning at different times around the globe.  Common
cloud-product composites simply use a pixel within the nearest scan time
(with similar sensor zenith angles) to select as the valid cloud state for
any given image.  Such an assumption leaves clouds out of valid
time position, especially when large temporal differences exist between
the composite valid time and imagery scan time over fast-moving clouds
(i.e. cirrus).  Within this manuscript, we introduce a method for
accounting for cloud motions between different scanned imagers before
composite synthesis, which we term as “Temporal Correction.”  The
method uses an advanced dense (every image pixel) optical flow retrieval
technique coupled with a simple, occlusion reasoning warping method
commonly used for temporal resolution enhancement.  The optical
flow retrieval was tuned using a large comparison between temporally
interpolated full-disk geostationary satellite imagery to corresponding
fine-temporal resolution 1-min scans.  It is demonstrated through
six separate case studies that the temporal correction methodology
improves the correspondence between retrieved cloud-top heights from
various geostationary and low-earth orbiting imagers by ~4.6-13.5% in
reduction of root mean squared error after such products are parallax
corrected and remapped to a common rectilinear grid.  Similar
improvements can be extended to products relevant to 3D cloud
reconstruction, such as cloud-base height.