For the first question, the first step in the UC Santa Cruz prediction strategy, after they gathered the data, was to perform delay-coordinate embedding of the data
It should be noted that they also used several other prediction strategies
But, the one were referring to here: use delay-coordinate embedding as the first preprocessing step of the data
For Question 2, a neural net-based method worked better than a nonlinear dynamics-based method on SFI dataset A in the long term
And this is false
While Wans method, which was the neural net method, worked great in the short term, Tim Sauers nonlinear dynamics-based method did far better in the long term
So this question is false
The third question is simply, How does Lorenzs method of analogues work?
And this is near-neighbor prediction on the trajectory of a dynamical system
So given a known trajectory of a dynamical system, and a point in that state space, it looks for the points nearest neighbor and the known trajectory, then uses the next point in the trajectory as the forecast of where that point will go next
So this first answer is correct
The third answer is a simple modification of Lorenzs method of analogues using k nearest neighbors
However, the original Lorenzs method of analogues used one neighbor
So the first answer is correct