Fig 1.
We adopt county level weather data and yield data to model the weather impact on maize yields. We consider parameter uncertainty through a pre-calibration method and climate forcing uncertainty through an ensemble of downscaled climate products.
Fig 2.
Annual mean yield hindcasts and projections under different methodological choices.
The black dots represent the area-weighted average annual yield observations. The green line is the best estimate of yield hindcast based on the model with the least cross-validation error. The deep blue lines are best estimates of this model for 18 different climate projections. Adding the effects of considered parameter uncertainty allows a pre-calibration to cover 95% of the observed yield data (light green area). The total effects of the considered uncertainties (climate forcings and parameters) expand the projection with a much wider 95% uncertainty range (light blue area).
Table 1.
List of model estimates and parameter sampling range.
Fig 3.
Marginal distributions of 30-year mean yield projections for different considerations of uncertainties.
a: 2020–2049 b: 2070–2099 The point labels the point estimate without considering any uncertainty (the best estimate in a linear shifted climate projection); the three solid lines (red, green and blue) are the distributions when considering only parameter uncertainty, only climate forcing uncertainty and both uncertainty sources respectively. The distribution medians are labeled as vertical black lines on the box-whisker plots.
Fig 4.
Decomposition of the uncertainty in 30-year mean yield distributions.
The uncertainties are measured in yield anomaly standard deviations. Two panels are for two periods. a: 2020–2049 b: 2070–2099. The percentages are the proportion of the uncertainty cumulatived up to each stage.