Dependence of radiation belt simulations to assumed radial diffusion rates tested for two empirical models of radial transport

Abstract

Radial diffusion is one of the dominant physical mechanisms that drives acceleration and loss of the radiation belt electrons, which makes it very important for nowcasting and forecasting space weather models. We investigate the sensitivity of the two parameterizations of the radial diffusion of Brautigam and Albert (2000) and Ozeke et al. (2014) on long‐term radiation belt modeling using the Versatile Electron Radiation Belt (VERB). Following Brautigam and Albert (2000) and Ozeke et al. (2014), we first perform 1‐D radial diffusion simulations. Comparison of the simulation results with observations shows that the difference between simulations with either radial diffusion parameterization is small. To take into account effects of local acceleration and loss, we perform 3‐D simulations, including pitch angle, energy, and mixed diffusion. We found that the results of 3‐D simulations are even less sensitive to the choice of parameterization of radial diffusion rates than the results of 1‐D simulations at various energies (from 0.59 to 1.80 MeV). This result demonstrates that the inclusion of local acceleration and pitch angle diffusion can provide a negative feedback effect, such that the result is largely indistinguishable simulations conducted with different radial diffusion parameterizations. We also perform a number of sensitivity tests by multiplying radial diffusion rates by constant factors and show that such an approach leads to unrealistic predictions of radiation belt dynamics.

The Effect of Plasma Boundaries on the Dynamic Evolution of Relativistic Radiation Belt Electrons

“…It can be seen that, except for the outer boundary at L * around 7, Ali et al () provide smaller radial diffusion coefficients than the other models. Similar comparisons of different radial diffusion coefficients have been made by other studies (e.g., Drozdov et al, ; Huang et al, ; Li et al, , ). In these comparisons, radial diffusion coefficients from Brautigam and Albert () are usually higher compared to other models during storm times.…”

“…In these comparisons, radial diffusion coefficients from Brautigam and Albert () are usually higher compared to other models during storm times. In particular, the radial diffusion coefficients from Brautigam and Albert () and Ozeke et al () were tested in Drozdov et al () by performing long‐term (1 year) simulation, and the results were similar. Note that in Drozdov et al () the outer boundary of the simulations was set up by using Van Allen Probe data at L * =5.5.…”

“…In particular, the radial diffusion coefficients from Brautigam and Albert () and Ozeke et al () were tested in Drozdov et al () by performing long‐term (1 year) simulation, and the results were similar. Note that in Drozdov et al () the outer boundary of the simulations was set up by using Van Allen Probe data at L * =5.5. In this study we set up the outer L * boundary condition at L * =6.6.…”

Local and Statistical Maps of Lightning‐Generated Wave Power Density Estimated at the Van Allen Probes Footprints From the World‐Wide Lightning Location Network Database

“…An early attempt to model the effect of LGW on trapped electrons is Abel and Thorne (, Table 1), who assumed a constant LGW amplitude of B m = 10 pT and an occurrence rate of 3% to produce a drift LGW intensity of 3 pT 2 that is independent of location and geomagnetic activity. Other authors adjusted the LGW occurrence rate based on literature and/or empirical knowledge of observed electron flux decay (e.g., Drozdov et al, ; Kim et al, ; Ripoll, Chen, et al, ; Subbotin et al, ). An empirical database of LGW intensity has been produced as a function of L shell, magnetic local time, and geomagnetic activity (Figure 1 of Meredith et al, ).…”

Measures of Model Performance Based On the Log Accuracy Ratio

“…This metric is similar to sMAPE in that the normalization uses the mean of x and y , but the normalization factor is constant for any given time and is given by max( y i ( f ) + x i ( f ))/2 where the maximum value is taken over all L * at a given time. An additional example of the ND metric being applied to characterize model performance over a 2‐D domain was given by Drozdov et al (), who compared Van Allen Probes electron flux data (binned in L * and time) with simulation output. They note that they use ND for this as “[ i ] t emphasizes how well the simulation can reproduce the flux peaks and flux profiles around the maximum.…”