Narrowband to Broadband Conversion with Spatially Autocorrelated Reflectance Measurements

Abstract

A new technique for estimating broadband reflectance from Advanced Very High-Resolution Radiometer (AVHRR) narrowband reflectances in channel 1 and 2 is developed. The data used are simultaneous and coincident narrowband and broadband measurements made by the AVHRR and Earth Radiation Budget Experiment (ERBE) radiometers aboard NOAA-9 during four days in July 1985 in the region north of 60°N. The limitations and inefficiency of classical regressional methods when applied to datasets with high spatial auto-correlation, which is often the case for remotely sensed data, are discussed. A statistical variable, Moran's I, is introduced, which is specifically designed for testing against a null hypothesis of spatial independence. On the basis of Moran's I and a correlogram analysis of the spatial autocorrelation of measured reflectances, the data are sampled to provide a spatially independent dataset. In addition to sampling, the data are also screened with respect to spatial homogeneity. Both scene-dependent and scene-independent regressional models are developed that are based on these spatially independent datasets. The rms errors of the predicted broadband reflectance are found to be 1.0, 1.8, 2.0, and 3.1 for the ocean, land, ice-snow, and cloud data, respectively. The effects of scene discrimination and solar and viewing geometry on the regressions are investigated, and comparisons are made between two-channel and single-channel models. The use of two solar channels is found to give a significant improvement in the predicted broadband reflectance for datasets in which there is no scene discrimination, a small improvement for measurements over land, and no improvement for the other homogeneous scene types. Geometric factors are found to have no significant effect on the regressions.