Spatial Backfitting of Roller Measurement Values from a Florida Test Bed

Abstract

Modern earthwork compaction rollers collect location and compaction information as they traverse a compaction site. These data are indirectly observed through non-linear measurement operators, inherently multivariate with complex correlation structures, and collected in huge quantities. The nature of such data was investigated at a large, atypically compacted test bed in Florida, USA. Exploratory analysis of this data through detrending and empirical semivariogram estimation is performed. A second analysis using a sequential, spatial backfitting algorithm is used to investigate the importance of driving direction of the roller.