The curves in the datasheet don't have enough resolution so it will be difficult to fit a Berkeley SPICE diode model to them. Best option is always a manufacturer supplied model - *IF* LTspice can run it!
If they don't offer a model, you can take one of the LEDs, run it at a very low current so you can count the number of dies, then make a best guess how many parallel strings, and how many dies per string to get instance parameter N for the simulated diode, then put it on a curve tracer (preferably one that can do low duty cycle pulsed measurements to avoid self-heating effects, divide the voltages measurements by N and apply the fitting procedure documented at the link below, and you can get a pretty decent model out of it.
https://hackaday.io/project/12874-automated-ledlaser-diode-analysis-and-modeling/log/48368-estimating-spice-diode-modelsFor a detailed explanation of the Berkeley SPICE diode model, and its parameters see:
http://www3.imperial.ac.uk/pls/portallive/docs/1/7292572.PDF