Precisionsodling av vall
Cultivated grassland is an important crop, both considering crop sequence and live- stock keeping. It is cultivated on 45 % of the Swedish arable land, but it is a crop that differs from grains regarding time of harvest, number of harvests and composi- tion of species. In other crops e.g. wheat an...
| Autor principal: | |
|---|---|
| Formato: | First cycle, G2E |
| Lenguaje: | sueco sueco |
| Publicado: |
2017
|
| Materias: | |
| Acceso en línea: | https://stud.epsilon.slu.se/12831/ |
| Sumario: | Cultivated grassland is an important crop, both considering crop sequence and live-
stock keeping. It is cultivated on 45 % of the Swedish arable land, but it is a crop that differs from grains regarding time of harvest, number of harvests and composi-
tion of species. In other crops e.g. wheat and barley, it is possible to estimate bio-
mass and nitrogen demand since the early 00’s using remote sensing technology. But due to the unique properties of ley, the technology cannot be applied with ease.
There are a few ongoing Swedish projects where the correlation between remote sensing and ley properties is studied. One is at the experimental station in Rådde,
which is where the data analysed in this report come from. In this particular field experiment, four different ley compositions are considered: three strict, different grasslands and one containing clover and grass. Five different amounts of nitrogen fertilizer have been applied to one of the grasses and to the mixed ley field. In con-
nection to all three harvests, an N-sensor has been used to measure hyper spectral reflectance data. In this work the correlation between reflection and biomass, TS, raw protein content and protein amount have been examined by using two methods.
The first one looks into single linear regression between the measured data and four
vegetation indices (VI), and the second method uses Partial Least Squares regres-
sion (PLS) with hyperspectral data, 400-900nm. The multivariate PLS models were validated by cross validation.
The results show that there are relationships between remote sensing and meas-
ured data – where the strongest correlations can be found between hyperspectral
data and protein amount in grasses. Among the studied VI’s, the ones based on green wavelength bands yielded the highest correlations. Also, it can be noted that
the ley containing clover had low correlations regarding all examined properties.
When forming models with PLS, the highest correlations could be seen with the
protein amount and raw protein content. Besides, it gets clear that the sensor used in
this study makes a good estimation of the share of clover.
Despite the facts that hyperspectral data yields higher correlations than using VI’s, but also that the correlations were high in cultivated grass, more research has
to be done in order to make further general conclusions. |
|---|