Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Forecasting Arctic Temperatures with Temporally Dependent Data Using Quantile Gradient Boosting and Adaptive Conformal Prediction Regions

Richard Berk

Abstract

Using data from the Longyearbyen weather station, quantile gradient boosting (``small AI'') is applied to forecast daily 2023 temperatures in Svalbard, Norway. The 0.60 quantile loss weights underestimates about 1.5 times more than overestimates. Predictors include five routinely collected indicators of weather conditions, each lagged by 14~days, yielding temperature forecasts with a two-week lead time. Conformal prediction regions quantify forecasting uncertainty with provably valid coverage. Forecast accuracy is evaluated with attention to local stakeholder concerns, and implications for Arctic adaptation policy are discussed.

Forecasting Arctic Temperatures with Temporally Dependent Data Using Quantile Gradient Boosting and Adaptive Conformal Prediction Regions

Abstract

Using data from the Longyearbyen weather station, quantile gradient boosting (``small AI'') is applied to forecast daily 2023 temperatures in Svalbard, Norway. The 0.60 quantile loss weights underestimates about 1.5 times more than overestimates. Predictors include five routinely collected indicators of weather conditions, each lagged by 14~days, yielding temperature forecasts with a two-week lead time. Conformal prediction regions quantify forecasting uncertainty with provably valid coverage. Forecast accuracy is evaluated with attention to local stakeholder concerns, and implications for Arctic adaptation policy are discussed.

Paper Structure

This paper contains 17 sections, 2 equations, 8 figures.

Table of Contents

  1. Introduction
  2. Svalbard, Norway: The Forecasting Setting
  3. Some Implications for Forecasting Temperatures
  4. Data
  5. Statistical Methods
  6. Background
  7. Some Details
  8. Results
  9. Univariate Plots
  10. Time-Series Plot for Svalbard Temperatures in 2022, 2023, and 2024
  11. The Quantile GBM Fit of Temperature
  12. Predictor Impacts on the Fitted Values
  13. Functional Forms
  14. Forecasts and Estimated Uncertainty
  15. Uncertainty Results
  16. ...and 2 more sections

Figures (8)

  • Figure 1: 2 p.m. 2023 temperatures with the time-series plot in the left panel and the histogram with a GEV distribution overlay in the right panel.
  • Figure 2: Daily 2 p.m. temperatures for 2022, 2023, and 2024 using a color palette that is color-blind friendly.
  • Figure 3: Observed versus fitted 2 p.m. temperatures. Black dots are the data, the blue solid line is a loess smooth serving as a visual aid, and the red horizontal line marks the melting point at $0^{\circ}\mathrm{C}$.
  • Figure 4: Observed and Fitted 2 p.m. temperatures from quantile gradient boosting plotted against day of year (2023). Black dots show the observed temperatures over time, the jagged blue line is an interpolation of the fitted values, and the solid red line marks the melting point at $0^{\circ}\mathrm{C}$.
  • Figure 5: Relative contribution of each lagged predictor to the fitted 2 p.m. temperatures, computed as the standardized reduction in loss attributable to each predictor.
  • ...and 3 more figures