Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2503.16118 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912498811142144 |
|---|---|
| author | Berk, Richard A. Braverman, Amy |
| author_facet | Berk, Richard A. Braverman, Amy |
| contents | In this paper, we step back from a variety of competing heat wave definitions and forecast directly unusually high temperatures. Our testbed is the Russian Far East in the summers of 2022 and 2023. Remotely sensed data from NASA's Aqua spacecraft are organized into a within-subject design that can reduce nuisance variation in forecasted temperatures. Spatial grid cells are the study units. Each is exposed to precursors of a faux heat wave in 2022 and to precursors of a reported heat wave in 2023. The precursors are used to forecast temperatures two weeks in the future for each of 31 consecutive days. Algorithmic fitting procedures produce forecasts with promise and relatively small conformal prediction regions having a coverage probability of at least .75. Spatial and temporal dependence are manageable. At worst, there is weak dependence such that conformal prediction inference is only asymptotically valid. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_16118 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Forecasting Extreme Temperatures in Siberia Using Supervised Learning and Conformal Prediction Regions Berk, Richard A. Braverman, Amy Applications In this paper, we step back from a variety of competing heat wave definitions and forecast directly unusually high temperatures. Our testbed is the Russian Far East in the summers of 2022 and 2023. Remotely sensed data from NASA's Aqua spacecraft are organized into a within-subject design that can reduce nuisance variation in forecasted temperatures. Spatial grid cells are the study units. Each is exposed to precursors of a faux heat wave in 2022 and to precursors of a reported heat wave in 2023. The precursors are used to forecast temperatures two weeks in the future for each of 31 consecutive days. Algorithmic fitting procedures produce forecasts with promise and relatively small conformal prediction regions having a coverage probability of at least .75. Spatial and temporal dependence are manageable. At worst, there is weak dependence such that conformal prediction inference is only asymptotically valid. |
| title | Forecasting Extreme Temperatures in Siberia Using Supervised Learning and Conformal Prediction Regions |
| topic | Applications |
| url | https://arxiv.org/abs/2503.16118 |