Tytuł pozycji:
Attempt at an application of neural networks for assessment of the nitrogen content in meadow sward on the basis of long-term fertilizer experiments
- Tytuł:
-
Attempt at an application of neural networks for assessment of the nitrogen content in meadow sward on the basis of long-term fertilizer experiments
- Autorzy:
-
Kopec, M.
Gondek, K.
- Data publikacji:
-
2015
- Wydawca:
-
Uniwersytet Warmińsko-Mazurski w Olsztynie / Polskie Towarzystwo Magnezologiczne im. Prof. Juliana Aleksandrowicza
- Język:
-
angielski
- Prawa:
-
Wszystkie prawa zastrzeżone. Swoboda użytkownika ograniczona do ustawowego zakresu dozwolonego użytku
- Źródło:
-
Journal of Elementology; 2015, 20, 1
1644-2296
- Dostawca treści:
-
Biblioteka Nauki
-
Przejdź do źródła  Link otwiera się w nowym oknie
On the basis of long-term fertilizer experiment, conducted since 1968, an attempt was
made to verify the nitrogen content with the use of a neural network, in terms of yields from
subsequent cuts. The experiment is located at Czarny Potok village near Krynica (20°54′ E;
49°24′ N) on the altitude of about 720 m a.s.l., at the foot of Jaworzyna Krynicka Mt., in the
south-eastern Beskid Sądecki mountain range, on a slope with 7° inclination and NNE aspect.
The experiment was set up in 1968 on a natural type of mountain meadow of mat-grass (Nardus
stricta L.) and red fescue (Festuca rubra L.) with a large share of dicotyledonous plants. The
soil was classified to acid brown soils developed from the Magura sandstone with the texture of
light silt loam. Since autumn 1985, the experiment has been conducted in two series, with the
same level of fertilization: without liming (0 Ca) and limed (+Ca). Liming was repeated in 1995.
The first liming was conducted with a dose calculated on the basis of 0.5 Hh value, the second
one was established according to the total hydrolytic acidity. Mineral fertilization was discontinued
in 1974 - 1975 and in 1993 - 1994, when the experiment was limited to an assessment
of the sward yield and its chemical composition. The experiment comprises 8 treatments with
five replications, receiving either nitrogen or phosphorus fertilization (90 kg N or 39.24 kg P ha-1)
and (39.24 kg P ha-1 and 124.5 kg K ha-1) against the PK background, nitrogen in two forms
(ammonium nitrate and urea) and two doses (90 and 180 N ha-1). In 1968-1980, phosphorous
and potassium fertilizers were sown in autumn and since 1981 – in spring. However, potassium
(1/2 of the dose) was supplemented in summer after I cut. In 1968 - 1973, thermophosphate was
applied, but triple superphosphate has been used since 1976. Over the whole period of the experiment,
nitrogen fertilizers have been sown at two dates: 2/3 of the annual dose in the spring
at the onset of plant growth and 1/3 of the dose several days after the first cut. A single regenerative
treatment with copper (10 kg kg-1) and magnesium (8 kg ha-1) was applied once in 1994.
Foliar nutrition (2 dm3 ha-1 applied twice) with the microelement fertilizer Mikrovit-1 has been
used since 2000. The microelement fertilizer contains (per 1 dm3): 23.3 g Mg; 2.3 g Fe; 2.5 g Cu;
2.7 g Mn; 1.8 g Zn; 0.15 g B and 0.1 g Mo. The model was compared with a regression analysis.
Statistical analysis was applied for two data sets: the whole data set, i.e. 43 years and 8 treatments,
2 cuts and 2 series – data of the 1st set (n = 1376), and a narrow data set, comprising exclusively fertilization 90 kg N ha-1, irrespective of the form against the background of PK – 43
years and 2 objects, 2 cuts and 2 series – data of the 2nd set (n = 344). A neural network can
be applied in the assessment of the nitrogen content on the basis of yield including subsequent
years of nitrogen fertilization and cuts. Neural networks including quantitative and qualitative
features are useful for modelling the element content.