LS-Fit performs a linear band prediction using least-squares fitting. You can use it to find regions of anomalous spectral response in a dataset. It calculates the coveriance of the input data and uses it to predict the selected band as a linear combination of the predictor bands plus an offset. The difference (residual) between the actual band and the modeled band is calculated and output as an image. Pixels with a large residual (positive and negative) indicate the presence of a feature that was not predicted (for example, an absorption band). The modeled band image is also included in the output. You can calculate the predicted band using existing statistics or from new statistics.

Predict from New Statistics

  1. From the Toolbox, select Spectral > Mapping Methods > Least Squares-Fit New Statistics. The Linear Band Prediction Input File dialog appears.
  2. Select an input file and perform optional spatial and spectral subsetting, then click OK.
  3. Click OK. The LS-Fit Parameters dialog appears.
  4. Select the bands to use as predictors.
  5. Select the band to predict by clicking on the band name in the Select the Model Band list. If you already selected this band as a predictor, it will be de-selected under Select the Predictor Bands.
  6. To subsample the data when calculating the statistics, enter Stats X/Y Resize Factor values less than 1.
  7. Enter an output statistics filename, if desired.
  8. Select output to File or Memory.

The output contains two bands: the modeled band and the residual image. You can compare the actual band and modeled band using dynamic overlays. Pixels in the residual image with large values, both positive and negative, show where the actual and modeled bands differ. ENVI adds the resulting output to the Layer Manager.

Predict from Existing Statistics

Use this option to specify an existing statistics file for band prediction. You can only use bands with covariance values in the statistics file as predictor or model bands.

  1. From the Toolbox, select Spectral > Mapping Methods > Least Squares-Fit Existing Statistics. The Linear Band Prediction Input File dialog appears.
  2. Select an input file and perform optional spatial subsetting, then click OK. The Enter Statistics Filename dialog appears.
  3. Select the statistics file that corresponds with the input data file. The LS-Fit Parameters dialog appears. Only the bands used in the statistics calculations are available for selection as predictor or model bands.
  4. Select the bands to use as predictors.
  5. Select the band to predict by clicking on the desired band name in the Select the Model Band list. If you already selected this band as a predictor, it will be deselected under Select the Predictor Bands.
  6. Select output to File or Memory.

The output contains two bands: the modeled band and the residual image. You can compare the actual band and modeled band using dynamic overlays. Pixels in the residual image with large values, both positive and negative, show where the actual and modeled bands differ. ENVI adds the resulting output to the Layer Manager.