Re-implement pixelwise linear regression

[Aariq]

I'm slightly hesitant to do this, but convinced myself that it'll be valuable. Hesitant because it's not like the pixelwise linear regression is something we necessarily aim to replicate—the GAMs are doing something different and I wouldn't expect the results to be identical. However, if there are particular regions or thresholds where results are consistently different, it may be valuable to investigate. For example, by choosing a lat lon and plotting predicted values from both linear regression and GAM along with observed data we could see where the linear regression is perhaps pulled by an extreme value or where the GAM isn't flexible enough to fit the trend.