Geostatistical prediction of vegetation amount using ground and remotely sensed data

Abstract

Geostatistical methods, which assume spatial dependence, have an untapped potential to map vegetation amount using ancillary data from remote sensing images. Two geostatistical methods, cokriging and conditional simulation, were constructed with aspatial regression in terms of their accuracy and uncertainty description.

For a synthetic data set constructed from imaging spectrometer data, spatial regression was most accurate when ground and spectral variables were very closely related (r between the data exceeding .89). Cokriging was more accurate in all other situations. Conditional simulation, though not as accurate, was superior to the other two methods in reproducing the univariate and spatial characteristics of vegetation amount. The sample size was 300 and the sampling fraction was .3%. For a real data set from western Montana, USA, over 300 ground measurements of conifer canopy cover made in each of two years by the US Forest Service and collocated NDVI values from Landsat TM were used to predict canopy cover in a 97 square km² subarea where the sampling fraction was .03%. The nonlinear aspatial regression model between canopy cover and NDVI had statistically identical parameters in both years, but prediction intervals were very wide and accuracy was low at test points. Cokriged maps had much higher accuracy, but were affected by the small sampling fraction and clumped of ground measurements. Conditionally simulated realizations using collocated cokriging displayed the desirable aspects of cokriging at the same time as presenting plausible global and spatial distributions of canopy cover and were therefore preferable to the cokriged maps.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata