Training Rasters and Covariates

Reference > Training Rasters and Covariates

These Datasheets hold the inputs to the 1-Train Classifier stage.

Training Rasters

Each row pairs a multi-band image with a user-classified raster of the same area. Add one row per image; rows can share a Timestep value or use different ones.

Field Description
Timestep Integer identifying the image (labelled with the Raster group name set in Terminology). Outputs are reported by this value.
Training Raster A multi-band raster (for example, a GeoTIFF of satellite imagery). Each band becomes a predictor. All training and predicting rasters must have the same bands in the same order.
User Classified Raster A single-band raster with the same extent, resolution and coordinate reference system as the training raster. By default it is treated as binary: the lowest value becomes absence (0) and the highest becomes presence (1). To use a multi-class raster, set Target Class Options. Cells with no data are not sampled.

Rasters without a coordinate reference system are given a generic planar one. Training and user-classified rasters that don’t match in extent or resolution are checked and aligned before sampling, and an image that can’t be read is skipped with a warning.

Covariates

Optional rasters added to every training raster as extra predictors, such as elevation or land cover.

Field Description
Covariate Raster A single-band raster covering the training area.
Data Type Continuous for numeric values, or Categorical for class codes (treated as factors by the model).

The same covariates should also be entered in the Predicting Rasters and Covariates Datasheets, so the model sees the same predictors at prediction time.