Parameterization-induced uncertainties and impacts of crop management harmonization in a global gridded crop model ensemble

Abstract

Global gridded crop models (GGCMs) combine agronomic or plant growth models with gridded spatial input data to estimate spatially explicit crop yields and agricultural externalities at the global scale. Differences in GGCM outputs arise from the use of different biophysical models, setups, and input data. GGCM ensembles are frequently employed to bracket uncertainties in impact studies without investigating the causes of divergence in outputs. This study explores differences in maize yield estimates from five GGCMs based on the public domain field-scale model Environmental Policy Integrated Climate (EPIC) that participate in the AgMIP Global Gridded Crop Model Intercomparison initiative. Albeit using the same crop model, the GGCMs differ in model version, input data, management assumptions, parameterization, and selection of subroutines affecting crop yield estimates via cultivar distributions, soil attributes, and hydrology among others. The analyses reveal inter-annual yield variability and absolute yield levels in the EPIC-based GGCMs to be highly sensitive to soil parameterization and crop management. All GGCMs show an intermediate performance in reproducing reported yields with a higher skill if a static soil profile is assumed or sufficient plant nutrients are supplied. An in-depth comparison of setup domains for two EPIC-based GGCMs shows that GGCM performance and plant stress responses depend substantially on soil parameters and soil process parameterization, i.e. hydrology and nutrient turnover, indicating that these often neglected domains deserve more scrutiny. For agricultural impact assessments, employing a GGCM ensemble with its widely varying assumptions in setups appears the best solution for coping with uncertainties from lack of comprehensive global data on crop management, cultivar distributions and coefficients for agro-environmental processes. However, the underlying assumptions require systematic specifications to cover representative agricultural systems and environmental conditions. Furthermore, the interlinkage of parameter sensitivity from various domains such as soil parameters, nutrient turnover coefficients, and cultivar specifications highlights that global sensitivity analyses and calibration need to be performed in an integrated manner to avoid bias resulting from disregarded core model domains. Finally, relating evaluations of the EPIC-based GGCMs to a wider ensemble based on individual core models shows that structural differences outweigh in general differences in configurations of GGCMs based on the same model, and that the ensemble mean gains higher skill from the inclusion of structurally different GGCMs. Although the members of the wider ensemble herein do not consider crop-soil-management interactions, their sensitivity to nutrient supply indicates that findings for the EPIC-based sub-ensemble will likely become relevant for other GGCMs with the progressing inclusion of such processes.

A regional nuclear conflict would compromise global food security

“…All EPIC-based models use the same core executable (EPICv0810) but differences arise from parameterizations of crop cultivars, soil attributes, soil nutrient cycling, hydrologic processes, and field management. See Folberth et al (72) for additional details and evaluations.…”

The GGCMI Phase 2 experiment: global gridded crop model simulations under uniform changes in CO<sub>2</sub>, temperature, water, and nitrogen levels (protocol version 1.0)

“…Understanding crop yield response to a changing climate is critically important, especially as the global food production system will face pressure from increased demand over the next century (Foley et al, 2005;Bodirsky et al, 2015). Climate-related reductions in supply could therefore have severe socioeconomic consequences (e.g., Stevanović et al, 2016;Wiebe et al, 2015).…”

A multi-model analysis of teleconnected crop yield variability in a range of cropping systems

“…Previous studies have shown that around 20 % to 45 % of IOD variability could be explained by ENSO depending on the data and the investigated time frame (Saji and Yamagata, 2003;Zhang et al 2015). The nature of this relationship is still debated (Hameed et al, 2018;Stuecker et al, 2017), and determining the influence of ENSO on the IOD and vice versa is not in the scope of this study. However, through the use of multivariate ridge regression, we aim to filter the influence of ENSO from the IOD patterns.…”