Prediction-powered inference for time series across space
This work provides reliable point estimates and confidence intervals given: short labeled time series (across spatial locations), a longer unlabeled time series, and an imperfect predictor of labels given covariates, and shows the method outperforms natural alternatives.