Logo image
Back
Conformal Prediction Strategies for the Cherry Blossom Day Problem
Conference proceeding   Peer reviewed

Conformal Prediction Strategies for the Cherry Blossom Day Problem

Fnu Shrishti, Nitul Singha, Shusen Pu and Achraf Cohen
Proceedings of IEEE Southeastcon
SoutheastCon 2026 (Huntsville, Alabama, USA, 02/20/2026–03/15/2026)
02/20/2026
DOI: https://doi.org/10.1109/SoutheastCon63549.2026.11476689

Metrics

1 Record Views

Abstract

The Peak Bloom Day (PBD) of cherry blossoms has cultural and economic importance in cities such as Washington, D.C., Kyoto, and Liestal. Climate change has shifted the PBD earlier, increasing uncertainty and complicating event planning. We compare three conformal prediction strategies: strict full conformal, rolling-window conformal, and split conformal for phenological forecasting. Using climate and bloom data from 1981-2022, with 2023 held out for evaluation, we pair strict full conformal with multiple linear regression (MLR), rollingwindow conformal with ARIMAX, and split conformal with long short-term memory (LSTM) networks. We evaluate point accuracy (MAE, RMSE) and interval reliability (width, empirical coverage). For the observed 2023 PBD, MLR and ARIMAX achieve valid 95% coverage, whereas LSTM yields sharper but under-covering intervals (61.5%), highlighting the tightnessvalidity trade-off. Conformal prediction intervals provide actionable bloom windows rather than single-date estimates.

Details

Logo image