Figure 5

Extrapolated projection in predicted values (gy) by traditional models and projection by random forests compared to the training sample under the temperature rise scenarios of 1°C and 6°C warming relative to 2005 levels. (a)-(c) display the projected per capita GDP growth (gy) of cells by different methods. (d) displays the temperature distribution of cells with histograms under the corresponding scenarios. In the plots, black represents gy or temperature in the training sample (\(N=2560\)); gray represents values under the temperature rise scenario of 1°C warming for all cells (\(N=1078\)). Purple represents values under the temperature rise scenario of 6°C warming for all cells (\(N=1078\)). We set the training sample (\(N=2560\)) as the baseline for comparison since extrapolation assumes that prediction cannot exceed the range of training values in gy for random forests. Thus, the min and max values in gy determine the distribution range for random forests. Values beyond the range of the training sample gy present the direct extrapolation ability in traditional regressions