|
| |
|
| |
|
| |
|
| |
|
|
Estimation of winter wheat yield using machine learning from airborne hyperspectral data
Švik, Marian ; Pikl, Miroslav ; Janoutová, Růžena ; Veselá, Barbora ; Slezák, Lukáš ; Klem, Karel ; Homolová, Lucie
Methods based on optical remote sensing allow nowadays to assess crop conditions over larger areas. The assessment of crop conditions and potential estimation of crop yields in the early growth\nstages can help farmers to better target their management practice such as application of fertilizers. In this study we analysed airborne hyperspectral images acquired several times during the growing season over two experimental sites in the Czech Republic (Ivanovice and Lukavec). The field experiments on winter wheat included 12 levels of fertilisation (combination of organic and mineral fertilisers). Such an experiment design and the possibility of combining the data from two sites together increased the variability in our wheat yield dataset, which varied between 2.8 and 10.0 t/ha. Further, we used a machine learning method – namely gaussian process regression from the ARTMO toolbox to train two variants of models: a) combining the spectral data from both sites and from the multiple acquisition days and b) combining the spectral data from both sites for individual acquisition days.The results showed that it was feasible to predict wheat yield already at the beginning of April with R2 > 0.85. This promising result, however, requires more thorough validation and therefore we plan to include more data from other sites in the next steps.
|
|
| |
|
| |
|
| |
|
| |
|
| |
host ::
RSS.