After some effort, I'm able to construct a reasonably accurate SPICE model for these LEDs, using measured I-V curves over temperature. I'd prefer a more sophisticated model that allows for dynamic temperature simulation, maybe next time?