Skip to content

Surface Water Mapping

Maps surface water from a true-colour orthomosaic, following section 2.4 of Chan, Dabros & McDermid (2026):

  1. Draws points_per_polygon random points inside each training polygon and reads the orthomosaic's red, green and blue values (bands 1–3) at each point.
  2. Trains an rpart decision tree on those values, and estimates its accuracy with 5-fold cross-validation on the training points.
  3. Classifies every cell of the orthomosaic. Cells the tree labels water become 1, all others 0.
  4. If a CHM is given, sets cells with canopy taller than max_canopy_height to 0. The CHM is resampled to the orthomosaic's grid first.
  5. Sets water patches (8-connected) smaller than min_patch_area to 0.

The training polygons need a class field. Polygons labelled water (in any case) are surface water; use any other labels for the rest, for example ground and shadow as in the paper. rpart needs at least 20 training points before it splits, so the tool warns if the tree ends up with no splits.

Parameter Type Required Description
in_ortho raster Yes True-colour orthomosaic; bands 1–3 are red, green and blue.
in_training vector (polygon) Yes Training polygons with a class field, as file.gpkg\|layer for a GeoPackage.
out_folder directory (output) Yes Folder for the outputs.
in_chm raster No Canopy height model, for example from Generate CHM or DSM from LiDAR Data.
max_canopy_height float No Cells with a CHM above this height (m) can't be water. The default is 0.2.
min_patch_area float No Water patches smaller than this area (m²) are removed; 0 keeps them all. The default is 0.04 (20 × 20 cm).
points_per_polygon int No Random training points per polygon. The default is 2.
seed int No Seed for the training points and cross-validation folds. The default is 1.
Output Contents
surface_water.tif 1 for surface water, 0 for everything else.
cv_accuracy.csv, cv_confusion_matrix.csv 5-fold cross-validation accuracy and confusion matrix.
training_points.gpkg The random training points and their classes.
decision_tree.txt The fitted tree's splits. 
from beratools.tools.rpy_surface_water import rpy_surface_water

rpy_surface_water(in_ortho="ortho.tif", in_training="training.gpkg|training",
                  in_chm="chm.tif", out_folder="surface_water")

Command line

bt surface_water   # show the parameters
bt surface_water --in_ortho ortho.tif --in_training "training.gpkg|training" --in_chm chm.tif --out_folder <out_dir>