How to separate orchards from natural vegetation with similar characteristics?

Working in a rather mountainous area is interesting but can be a challenge when trying to determine land use through satellite imagery. Cropping land is mainly located in the plains where fields are usually of a larger size, easy to distinguish from surrounding pastures and other area. Orchards, on the other hand, are located in the mountains, are small scale (due to the topography and the difficulties it brings for mechanized agriculture), and have a greening period at approximately the same time as the surrounding natural tree vegetation. I made a comparison of average NDVI per month to show the seasion.


NDVI season boxplot of shrubs based on MODIS 250 m data, from 2000-2013.


NDVI season boxplot of orchards based on MODIS 250 m data, from 2000-2013.

Both orchards and shrubs have a similar seasonal evolution, starting to green up in early spring, peak in May, and then slightly decreasing in greenness during the dry summer months. In the charts, orchards have a larger range between min and max than the shrubs, which could be explained by the coarseness of MODIS data. Orchards in Duhok are often smaller than 250 by 250 m (based on field observations) and this means that most MODIS pixels that include orchards also include other land covers that have both higher and lower NDVI than the median.

The spectral signature, i.e. the reflectance of light in different wavelength bands, for different land cover types are often used in supervised and unsupervised classification algorithms. Instead of using chart with the temporal dimension on the x axis, we use charts with the radiometric dimension, starting with blue reflectance, then green, red, Near Infrared (NIR) and sometimes also Short Wave Infrared. Based on the vegetation greenness, the presence of water, and perhaps the presence of soil, it can be possible to distinguish between vegetation types. With Object based Image Analysis, where pixels are grouped into objects based on their spectral properties, and then classified as objects, it is also possible to add texture and geometry as criteria for classification. In Landsat 8 bands 2-5, however, the spectral signature for shrubs and orchards are more or less the same, which makes it very difficult to distinguish between the two.



Spectral signature of shrubs based on Landsat 8 (image aquisition date: 157 julian calendar)


Spectral signature of orchards based on Landsat 8 (image acquisition date: 157 Julian calendar)

The 30 m resolution of Landsat makes it difficult to identify orchards, since it’s too coarse to show individual trees. The ability to detect individual trees would help discriminate orchards from natural tree vegetation, since fruit trees are often arranged in rows, while natural vegetation is more heterogeneous. These two images show an area with orchards and surrounding natural vegetation. The Google Earth image show clearly where the orchards are, while in the Landsat 8 panchromatic image (15 m, taken in June 2013) the orchards and the natural vegetation on the right side blend together. One thing that I’ve noticed when visually analyzing the area with Google Earth and Landsat images is that the orchards seem to be located in river valleys, often in ephemeral rivers and with rather sparse vegetation surrounding. This could be a way to go forward and try to distinguish orchards, the question is only if it will yield results that are accurate enough!


Orchards in Kurdistan with Google Earth (high resolution)


Orchards in Kurdistan with Landsat 8 Panchromatic (15 m resolution)



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s