Engineering topics

Craig, R. G.

Publications and source records attributed to Craig, R. G..

6 records

NASA NTRS1984

The spatial structure of terrain - A process signal in satellite digital images

Pattern recognition procedures applied to Landsat imagery carry an implicit assumption that the digital data are independently distributed. That assumption is incorrect over virtually any terrain. Deviations from independence occur because slopes follow a systematic pattern of variation arising from the slope-forming processes. That pattern can be identified using the stochastic process methodology of Box and Jenkins.Angles of adjacent slopes are autocorrelated and the bidirectional reflectance function transfers these systematic slope changes to the sensor. Imagery becomes autocorrelated through this transfer. Autocorrelation in the imagery can be removed through direct calculation from a digital elevation model or by use of stochastic process methodology. The latter has the advantage that the residuals are white noise; and it is applicable in any area, even where a D.E.M. is unavailable. The stochastic process signal can be used to study terrain processes.

Craig, R. G.↗

NASA NTRS1983

Effects of autocorrelation upon LANDSAT classification accuracy

Richmond, Virginia and Denver, Colorado were study sites in an effort to determine the effect of autocorrelation on the accuracy of a parallelopiped classifier of LANDSAT digital data. The autocorrelation was assumed to decay to insignificant levels when sampled at distances of at least ten pixels. Spectral themes developed using blocks of adjacent pixels, and using groups of pixels spaced at least 10 pixels apart were used. Effects of geometric distortions were minimized by using only pixels from the interiors of land cover sections. Accuracy was evaluated for three classes; agriculture, residential and "all other"; both type 1 and type 2 errors were evaluated by means of overall classification accuracy. All classes give comparable results. Accuracy is approximately the same in both techniques; however, the variance in accuracy is significantly higher using the themes developed from autocorrelated data. The vectors of mean spectral response were nearly identical regardless of sampling method used. The estimated variances were much larger when using autocorrelated pixels.

Craig, R. G.↗

NASA NTRS1982

Evaluation of terrain complexity by autocorrelation

The topographic complexity of various sections of the Ozark, Appalachian, and Interior Low Plateaus, as well as of the New England, Piedmont, Blue Ridge, Ouachita, and Valley and Ridge Provinces of the Eastern United States were characterized. The variability of autocorrelation within a small area (7 1/2-ft quadrangle) to the variability at widely separated and diverse areas within the same physiographic region was compared to measure the degree of uniformity of the processes which can be expected to be encountered within a given physiographic province. The variability of autocorrelation across the eight geomorphic regions was compared and contrasted. The total study area was partitioned into subareas homogeneous in terrain complexity. The relation between the complexity measured, the geomorphic process mix implied, and the way in which geobotanical information is modified into a more or less recognizable entity is demonstrated. Sampling strategy is described.

Craig, R. G.↗

NASA NTRS1981

Precision in the evaluation of Landsat autocorrelation - The terrain effect

The autocorrelation present in Landsat data can be described by a two-parameter model. Presence of this autocorrelation seriously inflates estimates of the variance of a set of pixels. The degree of inflation is always serious but varies markedly according to the values of the parameters phi and Theta. Thus corrections of the effects of the model require precise estimates of these parameters. Two hypotheses are proposed to explain the variation of phi with location. Several lines of evidence are presented which support the idea that phi is induced directly by the autocorrelation structure of the terrain being sensed. It is suggested that use of this relation will allow economical estimates of phi for any scene of interest.

Craig, R. G.↗

NASA NTRS1980

Sources of variation in Landsat autocorrelation

Analysis of sixty-four scan lines representing diverse conditions across satellites, channels, scanners, locations and cloud cover confirms that Landsat data are autocorrelated and consistently follow an Arima (1,0,1) pattern. The AR parameter varies significantly with location and the MA coefficient with cloud cover. Maximum likelihood classification functions are considerably in error unless this autocorrelation is compensated for in sampling.

Craig, R. G.↗

NASA NTRS1979

Autocorrelation in Landsat data

Many computer algorithms for the analysis of Landsat data have a statistical basis which requires that the observations comprise independent samples. Four distinct methods are employed to show that this assumption commonly is not fulfilled for these data. Each leads to a similar conclusion; the data must be sampled no closer than every 10th pixel in order to yield independent estimators. The implications of this are illustrated with a simple example.

Craig, R. G.↗