Unit 09 - Model tuning

Let’s improve our NDVI model created in Unit 08 - Modeler. The model operates in a current computation region, it would be better to define region based on user input, eg. by city area. Then NDVI would be computed only within user defined area.

NDVI values range from +1.0 to -1.0. Areas of barren rock, sand, or snow usually show very low NDVI values (for example, 0.1 or less). Sparse vegetation such as shrubs and grasslands or senescing crops may result in moderate NDVI values (approximately 0.2 to 0.5). High NDVI values (approximately 0.6 to 0.9) correspond to dense vegetation such as that found in temperate and tropical forests or crops at their peak growth stage. Let’s classify NDVI into 3 major classes:

  • Class 1: from -1.0 to 0.2
  • Class 2: from 0.2 to 0.6
  • Class 3: from 0.6 to 1.0

The desired output will be vector map with NDVI classes. Let’s also eliminate too small areas.

From GRASS perspective computation will be performed by several steps/modules:

  1. Erase cloud mask from input city region (v.overlay: not operator)
  2. Set computation region based on modified input city region (g.region)
  3. Set mask (r.mask)
  4. Compute NDVI values (i.vi)
  5. Reclassify NDVI values into classes (r.recode)
  6. Set nice color table for raster map classes (r.colors)
  7. Convert raster classes into vector areas (r.to.vect)
  8. Remove small areas (join them with adjacent areas by v.clean)

New modules can be added to the existing model in standard way by grass-module-add Add command (GRASS module) to the model. New commands are added to the end of a computation workflow which is not desired in this case. Good news: we can reorder commands (items) in Items tab.


Fig. 66 Reorder items - commands in the model. In this case move v.overlay up to the first position.


Be aware of correct computation region, don’t forget to align region bounds to input raster data (g.region with align option).

Reclassification of floating point raster maps can be done in GRASS by r.recode. Here is reclassification table for our case:


Beside predefined color tables r.colors (see Color table section) also allows to use user-defined color table by rules option. In our case color table can be quite simple:

1 grey
2 yellow
3 green


Reclassification and color table is recommended to store into files otherwise it can be lost when opening model next time: reclass.txt and colors.txt


Fig. 67 Extended model.

Sample model to download: ndvi-v2.gxm (note: don’t forget to fix path to reclass and colors file for r.recode and r.colors modules)


Till now our models have all parameters hard-coded, there is nothing which can be influenced by user when launching the model.

In Graphical Modeler user input can be defined by two mechanisms:

  • parametrization of module options
  • using self-defined variables (ideal when more modules are sharing the same user input value)

Let’s start with parametrization of module options, it’s simple. We would like to change our model in order to provide the user ability to:

  • define own city region area (option ainput in v.overlay)
  • set threshold for small areas (option threshold in v.clean)

For each command that we want to parameterize let’s open proprieties dialog by double-click on command item in the model. Then you find a option to be parameterized and enable Parameterized in model checkbox below. That’s all.


Fig. 68 Parametrization of ainput option in v.overlay command.


Parameterized commands are highlighted in the model by bold border.

After pressing grass-execute Run model the model is not run automatically. Instead of that a GUI dialog is shown to allow user defining inputs.


Fig. 69 Model is run after defining user input parameters. Parameterized options are organized into tabs based on the modules.

After setting the input parameters the model can be Run.


Saved models can be run directly from Layer Manager File ‣ Run model without opening Graphical Model itself.

Let’s test our model with various settings.


Fig. 70 NDVI vector class without small area reduction.


Fig. 71 NDVI classes smaller than 2000m 2 (so 20 pixel) removed.

Now we can change region, eg. by buffering Jena city region (v.buffer).

v.buffer input=jena_boundary output=jena_boundary_5km distance=5000

Fig. 72 NDVI vector classes computed in 5km buffer around Jena city region.

Sample model to download: ndvi-v3.gxm