Talk, 3rd Global Workshop on Proximal Soil Sensing 2013, Potsdam (Germany): 2013-05-26 - 2013-05-29
Abstract:
We investigated the feasibility of VIS-NIR reflectance spectra to predict clay content for different land-use forms in situ. We used partial least square regression on an independent validation dataset and root mean squared error and the Akaike information criterion to evaluate our model. With this model we predicted clay content in four soil profiles on a 3 x 3 cm scale. Models performed well for spectra taken in the laboratory (RPD > 2; R² > 0.76). The accuracy for in situ predictions however varies between the land-use forms and predictions are preliminary.