The impact of historical land-use change on the simulated global yield of wheat, maize, and rice

Description

Considering land-use change in dynamic global vegetation models has improved the simulation of vegetation state and carbon cycling since the legacies of past land-use changes persist in soil carbon and nitrogen concentrations for several years. However, the influence of land-use and management history on crop yields is not well explored in large-scale modelling. This study assessed the effect of land-use changes on global crop yield estimates with LPJ-GUESS for wheat, maize and rice and their interaction with climatic and management drivers. A total of 56 global simulations were performed by combining three different factors: (1) Four global land-use setups before the crop simulation: cropland, natural vegetation, conversion of natural vegetation to pasture before conversion to cropland, and a historical land-use change reconstruction; (2) The individual contributions of five drivers, atmospheric CO2, precipitation, radiation, temperature, and fertilisation; (3) Two different climate forcing datasets, CRU-NCEP and AgMERRA to assess the relative size of uncertainty from two different climate models. Yields simulated based on both climate-forcing datasets showed a similar relative response in trends and interannual variability. Simulations with previous land-use of natural vegetation and pastures caused higher soil nitrogen and carbon pools, increasing yields at the beginning of the simulations. Similarly, fertilisation was the main driver impacting the trend in yields and interannual variability. Simulations in which land was assumed to have always been cropland had similar trends to those using the land-use change database. All the land-use setups tended to converge over time to give similar yields, but convergence could take several decades. The main results highlight the critical role of historical land-use in simulating crop yield at subregional to local scales, particularly in locations with low fertilisation. At global to regional scales, the assumption of previous cropland cover simplifies simulations without affecting accuracy significantly.

Steps to reproduce

Fifty-six global simulations were performed by combining two different factors. Four different setups of land-use change (LC) during spin-up and simulations and seven crop yield driver setups as described below. Finally, all simulations were carried out for two different climate datasets. Land-use change setups • LCcrop: Five hundred years of spin-up with cropland cover and simulation from 1901 to 2010 for CRU-NCEP. For AgMERRA, the spin-up is until 1980, and then the simulation is undertaken. This spin-up was previously used for crop-focused papers (Olin et al. 2015b; Camargo-Alvarez et al. 2022) and is comparable to approaches taken in the majority of GGCMs (Franke et al., 2020). • LCnat: Five hundred years of natural vegetation spin-up and land-use change to cropland in 1960. Land-use change to cropland in 1980 for AgMERRA. • LCnatpas: Five hundred years of natural vegetation spin-up and land-use change to grassland in 1920, then to cropland in 1960. Land-use change to cropland in 1980 for AgMERRA. • LCLUH: Land-use history from Land Use Harmonization 2 (Hurtt et al. 2020), specifying net land transitions between natural vegetation, pasture, and cropland annually starting from 1901, i.e., the best estimate of actual LC history. Land-use fractions of 1901 are used for previous years. Driver combinations • All drivers fixed (DFIX). • [CO2] is time varying, other drivers fixed (DCO2). • Fertilisation is time varying, other drivers fixed (DFER). • Precipitation is time varying, other drivers fixed (DPRE). • Radiation is time varying, other drivers fixed (DRAD). • Temperature is time varying, other drivers fixed (DTEMP). • All drivers time varying (DALL). Forcing Climates All the simulations were performed using CRU-NCEP and AgMERRA forcing climates. A direct quantitative comparison between simulations forced by the two climate datasets was not performed since the simulations have different timespans and spin-up periods. Rather, the datasets are compared qualitatively. A different spin-up protocol was used for CRU-NCEP, reflecting the longer climate time series available. In the case of CRU-NCEP, the first historical year is 1901 and then observed climate and [CO2] data were used during the LC setups before 1960. In AgMERRA, the first historical year is 1981, meaning LC setups occurred during the spin-up with cycling 1980-2010 climate and [CO2] from 1980. Fixed driver setups recycle the values from 1960 for CRU-NCEP and 1980 for AgMERRA.

Institutions

• University of Birmingham

  England, Birmingham
• Lund University

  Skåne, Lund
• Karlsruhe Institute of Technology

  Baden-Wurttemberg, Karlsruhe

Categories

Funders

• University of Birmingham

  Birmingham

Licence