Adaptive reconstruction of sequential AVHRR imagery of Texas via dynamic compositing using an exponentially weighted polynomial function

A reconstruction system was developed to increase the discrimination capability for imagery which has been modified by residual effects resulting from imperfect sensing of the target and by spatially autocorrelated noise caused by atmospheric attenuation of the signal. As a “dynamic compositing” technique, it recovers missing measurements resulting from cloud cover and sensor noise. The system also utilizes temporal information to enhance the imagery based on an adaptive polynomial filter. The intensity at the next time step can be predicted using an exponentially weighted criterion so that the estimates reflect more recent variations in the process. If sequential images have high temporal resolution, it can be assumed that two images from the same geographical area have identical spatial structures in contiguous time steps. Under this assumption, a missing pixel observation can be estimated directly using observed values of neighboring pixels. The dynamic compositing algorithm presented includes spatial information from observations of neighboring pixels as well as temporal information from the polynomial filter. The Normalized Difference Vegetation Index (NDVI) was computed for Advanced Very High Resolution Radiometer (AVHRR) data collected over Texas from 1991 to 1992, and images were analyzed using the new integrated system for these sequence