Ensemble modeling improves projections of climate impact on potato

Potato-Blog2 Potato farm in Washington State. Photo: Ashok Alva, Kuwait Institute for Scientific Research

By Carolyn Mutter

In a recent Global Change Biology contribution, David H. Fleisher (USDA-ARS) and 25 others provide the first reported results quantifying uncertainty in simulated yields for tuber/root crops for both low- and high-input growing environments.

The report, entitled ‘A potato model intercomparison across varying climates and productivity levels’ also demonstrates that modeling assessments of climate change impact on potato may be improved using ensemble approaches such as those being advanced by the Agricultural Model Intercomparison and Improvement Project (AgMIP) community.

The potato crop multi-model assessment was conducted to quantify variation among models and evaluate responses to climate change. Nine modeling groups simulated agronomic and climatic responses at low- and high-input management sites. Two calibration stages were explored, partial (P1), where experimental dry matter data were not provided, and full (P2).

Fleisher_et_al-2017-Global_Change_Biology (fig1) Variation in simulated end of season tuber yield dry mass among nine potato crop models at four sites including low-input (Bolivia, Burundi) and high-input (Denmark, United States) management. Partial (P1) and full (P2) calibration phases are shown along with observed (Obs) tuber yield values and standard deviations. Observed (open circles) and simulated (filled circles) means and median values (solid line) as indicated. (Fleshier et al., 2017)

The median model ensemble response (see also figure right) outperformed any single model in terms of replicating observed yield in both low- and high-input growing sites.

Uncertainty in simulated yield decreased by 18% with full calibration. Uncertainty averaged 15% higher for low- vs. high-input sites, with larger differences observed for evapotranspiration, nitrogen uptake, and water use efficiency as compared to dry matter.

A minimum of five partial, or three full, calibrated models was required for an ensemble approach to keep variability below that of common field variation. Model variation was not influenced by change in carbon dioxide, but increased as much as 41% and 23% for yield and evapotranspiration, respectively, as temperature or rainfall moved away from historical levels.

“Collaborative, international activities help to foster a sense of community and common purpose. Although we all have our day-jobs, my sense as project leader was that the participating modelers were willing to put in additional work in order to improve our food-security related research tools. The strong buy-in from all participants from start to finish of the project is evidence of this dedication. The AgMIP potato pilot continues and has expanded its international base,” commented lead author David H. Fleisher.

The study also identified specific areas for future potato model development. One area regards the methods by which models implement temperature effects on developmental rates and growth processes. The observation that increases in temperature accounted for the highest amount of uncertainty suggests that methods and parameters for temperature sensitivity represent a considerable unknown among models. Similarly, the higher levels of uncertainty observed in low- versus high-input sites indicates methods for simulating water and/or nitrogen stress need to be evaluated. In addition, the observation that simulated harvest index is scarcely influenced by rising C levels (in contrast to experimental results) suggests carbon allocation is an additional modeling area that can be addressed to improve future forecasts.

See the AgMIP Potato Pilot for more information on potato intercomparison and improvement or to get involved with the potato pilot modeling activities.