Additional Verification Figures

Residuals

Distribution of the residuals between the true and predicted fluxes (i.e., the result of subtracting one from the other). The dotted black lines represent a factor of 5 error in the prediction. We observe that both models perform similarly, with the majority of data being within a factor of 5, especially for the lower energy channels. The higher energies are typically more variable, so it's not unexpected to see the model finds these predictions more difficult.

GOES + Indices

ind_Residuals.pdf

Goes + Solar Wind

sw_Residuals.pdf

Performance by Activity Level

Capturing quiet-time dynamics is important, but usually the crux of a good model is if it can perform well during active times. To this end, we split up our model predictions by deciles of SYM-H or Bz and present the respective performance over the energy channels.

Even in the tails of the distributions our model predictions typically fall within a factor of 6 (10^0.75). Generally, the performance is consistent across activity levels. Notable exceptions are the dips in performance at the highest deciles of SYM-H. We speculate that these data partially represent the results of shocks hitting the magnetosphere - such dynamic behaviour is always going to be difficult to capture.

ind_sw_symh_bz.pdf

Timeseries

Timeseries of data and model outputs averaged over single crossings of the boundary. The blue dots represent the mean flux measured by THEMIS as it crossed the boundary. The orange points with errorbars represent the mean model output and standard deviation. The grey envolope represents a factor of 5 error from the true value.

We again see that the model performs well, though see that the uncertainty may be underestimated for the higher energy channels (there is less overlap between the measured and predicted flux distribution).

GOES + Indices

ind_timeseries_test.pdf

Goes + Solar Wind

sw_timeseries_test.pdf

Page updated

Google Sites

Report abuse