A high-resolution crop fertilisation database for Germany under SSP/RCP scenarios, 1960-2100

Agriculture faces the challenge of securing a food supply for the growing global population while producing it in a sustainable manner, reducing the impacts of crop management on natural resources, air quality, climate change, and biodiversity. Robust representations of crop processes in agroecosystem modelling allow a better understanding of the interactions and feedbacks between agriculture, climate, environment and society, increasing the likelihood of meeting these challenges. In addition, improved agricultural modelling enhances the simulation of carbon cycling and natural vegetation in Dynamic Global Vegetation Models (DGVM). One main limitation for mechanistic agricultural representation in models is the low availability of high-resolution and long-term management datasets at regional or national scales, such as the application rates of nitrogen (N), the most important nutrient for plant growth. Here, we estimated a crop-specific N fertilisation dataset at 3-arc-min resolution for 1961–2015 and also projected future N applications at the same resolution under SSP1-RCP2.6, SSP3-RCP7.0 and SSP5-RCP8.5 for 2016-2100 in Germany. We included the crop groups C3 and C4 cereals, oil crops, starch crops, and fruit and vegetables under high and low-intensity management. Historical estimates were based on HILDA+ land cover data, harmonised by hindcasting the CRAFTY-GERMANY 2020 baseline and using an existing global N fertilisation database. The estimations were bias-corrected to match yearly FAO country-level statistics of fertiliser consumption.

For future projections, a baseline map for Germany of average fertilisation by state for 2005-2015 was generated, as well as a spatial deviation map, filling missing grid cells by ordinary kriging interpolation. In grid cells where a given crop was projected in the future according to the CRAFTY-GERMANY land-cover data obtained for each scenario, the fertilisation was calculated by weighting the average fertilisation baseline according to the future trends, adding the corresponding spatial deviation for that grid cell, plus a random value from the kriging interpolation error, assuming a normal distribution. The resulting historical spatiotemporal N fertilisation dataset was consistent with statistics of the International Fertilizer Association and can be used as an input for crop models and DGVMs. The same approach can be applied at the regional and global scale to improve modelling inputs.