Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Probabilistic bias adjustment of seasonal predictions of Arctic Sea Ice Concentration

Parsa Gooya, Reinel Sospedra-Alfonso

TL;DR

This work tackles the challenge of uncertainty in seasonal Arctic SIC forecasts by introducing a probabilistic bias correction using a conditional variational autoencoder (cVAE) to learn p(Y|x̄). The method enables generating large, calibrated ensembles of corrected forecasts and provides metrics (rank histograms, SOE, QQ plots, RMSE, SIA, SIE, IIEE) to assess calibration and accuracy. Results show that the nadj cVAE correction yields well-calibrated ensembles that closely match the observational distribution and improve both probabilistic and deterministic skill over a climatological bias correction baseline, especially at shorter lead times. The approach offers a scalable, uncertainty-aware post-processing framework for seasonal Arctic SIC forecasts with potential to support decision-making under risk and extremes.

Abstract

Seasonal forecast of Arctic sea ice concentration is key to mitigate the negative impact and assess potential opportunities posed by the rapid decline of sea ice coverage. Seasonal prediction systems based on climate models often show systematic biases and complex spatio-temporal errors that grow with the forecasts. Consequently, operational predictions are routinely bias corrected and calibrated using retrospective forecasts. For predictions of Arctic sea ice concentration, error corrections are mainly based on one-to-one post-processing methods including climatological mean or linear regression correction and, more recently, machine learning. Such deterministic adjustments are confined at best to the limited number of costly-to-run ensemble members of the raw forecast. However, decision-making requires proper quantification of uncertainty and likelihood of events, particularly of extremes. We introduce a probabilistic error correction framework based on a conditional Variational Autoencoder model to map the conditional distribution of observations given the biased model prediction. This method naturally allows for generating large ensembles of adjusted forecasts. We evaluate our model using deterministic and probabilistic metrics and show that the adjusted forecasts are better calibrated, closer to the observational distribution, and have smaller errors than climatological mean adjusted forecasts.

Probabilistic bias adjustment of seasonal predictions of Arctic Sea Ice Concentration

TL;DR

This work tackles the challenge of uncertainty in seasonal Arctic SIC forecasts by introducing a probabilistic bias correction using a conditional variational autoencoder (cVAE) to learn p(Y|x̄). The method enables generating large, calibrated ensembles of corrected forecasts and provides metrics (rank histograms, SOE, QQ plots, RMSE, SIA, SIE, IIEE) to assess calibration and accuracy. Results show that the nadj cVAE correction yields well-calibrated ensembles that closely match the observational distribution and improve both probabilistic and deterministic skill over a climatological bias correction baseline, especially at shorter lead times. The approach offers a scalable, uncertainty-aware post-processing framework for seasonal Arctic SIC forecasts with potential to support decision-making under risk and extremes.

Abstract

Seasonal forecast of Arctic sea ice concentration is key to mitigate the negative impact and assess potential opportunities posed by the rapid decline of sea ice coverage. Seasonal prediction systems based on climate models often show systematic biases and complex spatio-temporal errors that grow with the forecasts. Consequently, operational predictions are routinely bias corrected and calibrated using retrospective forecasts. For predictions of Arctic sea ice concentration, error corrections are mainly based on one-to-one post-processing methods including climatological mean or linear regression correction and, more recently, machine learning. Such deterministic adjustments are confined at best to the limited number of costly-to-run ensemble members of the raw forecast. However, decision-making requires proper quantification of uncertainty and likelihood of events, particularly of extremes. We introduce a probabilistic error correction framework based on a conditional Variational Autoencoder model to map the conditional distribution of observations given the biased model prediction. This method naturally allows for generating large ensembles of adjusted forecasts. We evaluate our model using deterministic and probabilistic metrics and show that the adjusted forecasts are better calibrated, closer to the observational distribution, and have smaller errors than climatological mean adjusted forecasts.

Paper Structure

This paper contains 12 sections, 8 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Data and Methods
  3. Generative Framework
  4. Architecture and Inference
  5. Data
  6. Evaluation metrics
  7. Evaluation metrics
  8. Results and Discussions
  9. Appendix
  10. Architecture and Training
  11. Evaluation Metrics
  12. Scaling the standard deviation of the prior distribution

Figures (2)

  • Figure 1: a) CDF of rank histograms of the nadj/badj versus lead times measured at marginal ice grid cells. Only three lead times are plotted for visibility. b) SOE versus lead time showing reliability. c) QQ plots at three lead times comparing the distribution of SIC at marginal ice grid cells with obs. d) RMSE (solid) over initialization time between the ensemble mean nadj/badj compared to obs at grid cells level averaged over the entire region. The dashed line shows the global mean ensemble spread averaged over initialization time. e) For each lead time, RMSE of SIA (solid line) and SIE (dashed line) over initialization time, and average IIEE (dotted line) over intialization time is compared between ensemble mean nadj/badj and obs. f) same as (e) but for pattern correlation relative to obs.
  • Figure A1: Same as Figure 1 but for the cVAE without scaling the standard deviation of the prior distributions at inference time. a) CDF of rank histograms of the nadj/badj versus lead times measured at marginal ice grid cells. b) SOE versus lead time showing reliability. c) QQ plots at three lead times comparing the distribution of SIC at marginal ice grid cells with obs. d) RMSE (solid) over initialization time between the ensemble mean nadj/badj compared to obs at grid cells level averaged over the entire region. The dashed line shows the global mean ensemble spread averaged over initialization time. e) For each lead time, RMSE of SIA (solid line) and SIE (dashed line) over initialization time, and average IIEE (dotted line) over intialization time is compared between ensemble mean nadj/badj and obs. f) same as (e) but for pattern correlation relative to obs.