We are going to classify a multitemporal image stack of MODIS NDVI time series (MOD13Q1). The stack consists of 23 bands (16-day composites) with a spatial resolution of 231m in sinusoidal projection. We want to classify the different land use types, especially to discriminate different crop types.

Install Python and required image processing and scientific programming packages:

For beginners, the basic Python IDLE is sufficient for scripting. However, another IDE, LiClipse is highly recommended: http://www.liclipse.com/

We assume that you already have created a bunch of training samples in 8bit TIFF format with distinct class labeling (1,2,3,4, etc). It is necessary that these single pixels are snapped to the pixel size of the original dataset and have the same dimensions and extent.

Now write your Training script:

And now write your classification script:

Multitemporal NDVI MODIS Stack with clipped AOI:

1

Training pixels with distinct classes are created from field sampling points and snapped to the original raster extent:

2

Classification results (left image: same extent as above):

3 4