UCLA

In the following we unite Adaptive Least Squares (ALS) and Inverse Distance Weighting as a computationally frugal means of modeling very large space-time data. This technique, dubbed weighting by inverse distance with adaptive least squares (WIDALS) boasts several merits, including a small and readily interpretable

hyperparameter space, and relative ease of implementation. We include RMSE comparisons between WIDALS and various solutions including the Kalman solution on small simulated data sets. We culminate our work with a large scale imputation/model tting ritual (dubbed \phyning") using WIDALS on 6 contemporaneous climatological data set, provided by the National Climatological Data Center, possessing 6 11669 = 70014 covariates/responses over 1016 time points accompanied by about 24 million exogenous covariates for a total of about 96,000,000 scalar data.