Bridging the data gap: Decision tree models for complete pitch-angle distributions

Modelling the Earth’s radiation belts is a key tool in the modern space physics research arsenal. With current state-of-the-art radiation belt models, such as the British Antarctic Survey Radiation Belt Model (BAS-RBM), we can investigate wave-particle interactions and long-term radiation belt behaviour, as well as provide short-term forecasting of radiation belt fluxes. The quality of these model outputs can only ever be as good as the inputs, however.

As with any Fokker-Planck based radiation belt model, the BAS-RBM is driven and moderated by two key inputs: the initial conditions, and the boundary conditions. For both inputs, we require full knowledge of the radiation belt conditions, over the whole simulation space (initial) and over slices at the edges of the simulation space (boundary). Deriving these from in-situ satellite data gives us the best chance at reproducing real-world events and providing accurate predictions, however satellites are notoriously localised, proving only trace measurements throughout the simulation space. One of the biggest limitations of satellite measurements for these purposes is the lack of full electron equatorial pitch-angle distribution (PAD) measurements. Even with satellites such as RBSP and Arase, which purport to provide full PADs, any time they are off-equator we get only partial equatorial PADs. This necessitates some form of “filling” to allow for full pitch-angle distributions in simulation space. Traditionally this is done using regressive models, such as the Shi et al. (2016) PAD model; these models are limited, however.

We present a novel technique for deriving electron pitch-angle distributions using white-box machine-learning, allowing for the generation of full PADs from partial data, derived from electron measurements from the RBSP satellites. We demonstrate the benefits that this model has over traditional approaches, and the impacts that such “realistic” models have on the outputs of the BAS-RBM. Finally, we investigate the potential utility of this model in other areas of radiation belt science.