Gaussian Process Model Settings and Length Scale

Hello,

So I have a dataset comprised of 40 runs of 52 yearly observations and 40 runs of 32 yearly observations (3360 observations in total). There are 18 simulated predictors (all with different ranges falling between 0-22 meters/second).Time (discrete values from 0-52) is my 19th predictor as well to represent the effect of the current state on the rate of change. I have researched the process a fair amount and am familiar with the syntax and effect of different kernels but I am having trouble figuring out if my general approach is correct. I'm fitting a Gaussian process model for each of the 10 components in a system that add together as such, reserving one of the runs for validation:

gpr.(fName) = fitrgp(COV(ABC,:),score(ABC,i),'FitMethod','sd',...
'PredictMethod','exact','KernelFunction','squaredexponential',...
'OptimizeHyperparameters','all','HyperparameterOptimizationOptions',...
    struct('AcquisitionFunctionName','expected-improvement-plus','UseParallel',true,'MaxObjectiveEvaluations',20));

and then planned on using the output of this to re-fit the model but this time with different length-scales for my predictors since scales are held constant during optimization.

What are the obvious pitfalls with this plan such as overfitting, and any ideas on how to work with the variety of ranges of values between my predictors?

Thank you!