Conference Presentation (After Call)

FZJ-2016-02021

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Analysing data aggregation effects on large-scale yield simulations

Hoffmann, H. (Corresponding author) ; Zhao, G. ; Asseng, S. ; Bindi, M. ; Cammarano, D. ; Constantin, J. ; Coucheney, E. ; Dechow, R. ; Doro, L. ; Eckersten, H. ; Gaiser, T. ; Grosz, B. ; Haas, E. ; Kassie, B. ; , K. K. C. ; Kiese, R. ; Klatt, S. ; Kuhnert, M. ; Lewan, E. ; Moriondo, M. ; Nendel, C. ; Raynal, H. ; Roggero, P. P. ; Rötter, R. ; Siebert, S. ; Sosa, C. ; Specka, X. ; Tao, F. ; Teixeira, E. ; Trombi, G. ; Yeluripati, J. ; Vanuytrechten, E. ; Wallach, D. ; Wang, E. ; Weihermüller, L.FZJ* ; Zhao, Z. ; Ewert, F.

2016

Abstract: Large-scale yield simulations often use data of coarse spatial resolution as input for process-based models (Ewert et al., 2015; Zhao et al., 2015). However, using aggregat-ed data as input for process-based models entails the risks of introducing errors linked to aggregation effects (AE) such as: i) data modification; ii) missing the valid range of the model; iii) data inconsistencies between data types. While the regional crop yield bias is usually <5 % on average over all years, it may increase to more than 10 % under specific conditions, e.g. in single years (Hoffmann et al., 2015), depending on the mod-el. In order to assess these differences in detail, we present a model intercomparison on AE for a range of environmental conditions differing in climate and soil for two crops grown under three different production situations.Multi-model ensemble runs were conducted with soil and climate input data at resolu-tions from 1 to 100 km for the state of North Rhine-Westphalia, Germany. Climate data was spatially averaged. Soil data was aggregated by area majority. Winter wheat and silage maize yields of 1982-2011 were simulated with 11 models for potential, water-limited and water-nitrogen-limited production after calibration to average re-gional sowing date, harvest date and crop yield.Results and DiscussionRegional yields were reproduced by the models on average, regardless of input data type and resolution. However, AE were observed in dry years as well as due to soil aggregation. Large positive AE between coarser and 1 km resolutions were associated with low soil water holding capacity (SWHC) and low climatic water balance (CWB) at 1 km resolution in combination with increases in SWHC and CWB when aggregating to coarser resolutions. Consistently, the opposite was found for large negative AE. Large AE (larger/lower mean +/-2 standard deviations of AE) were due to changes in soil type. However, the lower 50 % of all AE also showed differences in the aggregation method: soil aggregation by majority led to about 20 % of grid cells with no AE at 100 km resolution whereas climate data aggregation by averaging always led to AE. Still, soil aggregation by majority led to a larger fraction of AE larger than 10 % as compared to climate averaging. Notably, this was further increased by the combined use of ag-gregated soil and climate data. Finally, models differed considerably in AE.ConclusionsThe results highlight the interactions between model, data and aggregation method with AE, emphasizing the importance of models intercomparison analyses.

---

Contributing Institute(s):

1. Agrosphäre (IBG-3)

Research Program(s):

1. 255 - Terrestrial Systems: From Observation to Prediction (POF3-255) (POF3-255)
2. MACSUR - Modelling European Agriculture with Climate Change for Food Security (2812-ERA-158) (2812-ERA-158)

---