I think this is the cleanest option: dt = (dt0-dt1)*exp(-(t-tstart)/td) + dt1 (or something like this).
Should address the short required timescale at the start, and the long timescale towards the end of an initialisation, ideally minimising the initial Thwaites runaway, and the later high frequency oscillations.
I think this is the cleanest option: dt = (dt0-dt1)*exp(-(t-tstart)/td) + dt1 (or something like this).
Should address the short required timescale at the start, and the long timescale towards the end of an initialisation, ideally minimising the initial Thwaites runaway, and the later high frequency oscillations.