Advice on the nitrogen (N) fertilisation of grassland is usually based on the N supply of the soil (SNS). Generally, a higher SNS results in advice for lower N fertilisation rate. Recent research showed that the recommended N fertiliser rate can deviate by more than 60 kg ha-1 y-1 from the optimum. Historic and recent grassland field trial data are, therefore, re-analysed to improve the accuracy of optimum fertiliser rates. Since 1950, over 250 trials were carried out on 140 different locations, resulting in a database with about 10,000 records. The main soil types were sand (46%), clay (21%), river clay (4%) and peat (28%). Analysis showed that the SNS increased over time and that the apparent nitrogen recovery (ANR) was constant over N doses varying between 0-350 kg ha-1 y-1. However ANR varied considerably between sites and years. We used machine learning algorithms (random forest models) to estimate the ANR in response to soil, weather and management. Initial results showed that it was possible to accurately distinguish fields with ANR.
Pagina's / pages: 3
Type: Congres bijdragen
Download (pdf, 0.54 MB)
Abstract / summary in English:
Keywords in English: soil N delivery, ANR, N fertiliser, random forest