Efficiently Learning Nonstationary Gaussian Processes for Real World Impact