PREDICTION OF SOIL PROPERTIES WITH SPECTRORADIOMETRIC MEASUREMENTS

Vis-NIR spectroscopy for soil has gained relevance among
the development of soil technique analysis. Itatiaia National
Park (INP) located at the Southern border of Rio de Janeiro
and Minas Gerais states, has a fundamental role for water
storage and distribution in the surrounded watersheds. INP
is also one of the best exemplars of soil carbon stocks in
the tropical regions. The purpose of this study was to test
the Random Forest Algorithm for soil properties prediction.
The spectral readings were done on 116 soil samples, in
a completely dark room laboratory. Results show that the
R2 reached 0.57 for Hydrogen. Random Forest model
created a rank of importance for the variables used in the
prediction, in this stroke within this soil samples, the most
important wavelength band was 2165 nm among others. The
homogeneity and the limited number of samples suggest a
disadvantage of the prediction models.