Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Density correction for multivariate spatial fields of global climate model output using deep learning

Reetam Majumder, Shiqi Fang, A. Sankarasubramanian, Emily C. Hector, Brian J. Reich

TL;DR

A new semi-parametric estimation of conditional densities (SPECD) approach is proposed for density correction of the joint distribution of daily precipitation and maximum temperature data obtained from gridded GCM spatial fields and predictions obtained using SPECD are better calibrated compared to predictions using asynchronous quantile mapping and canonical correlation analysis.

Abstract

Global Climate Models (GCMs) are numerical models that simulate complex physical processes within the Earth's climate system and are essential for understanding and predicting climate change. However, GCMs suffer from systemic biases due to simplifications made to the underlying physical processes. GCM output therefore needs to be bias corrected before it can be used for future climate projections. Most common bias correction methods, however, cannot preserve spatial, temporal, or inter-variable dependencies. We propose a new semi-parametric estimation of conditional densities (SPECD) approach for density correction of the joint distribution of daily precipitation and maximum temperature data obtained from gridded GCM spatial fields. The Vecchia approximation is employed to preserve dependencies in the observed field during the density correction process, which is carried out using semi-parametric quantile regression. The ability to calibrate joint distributions of GCM projections has potential advantages not only in estimating extremes, but also in better estimating compound hazards, like heat waves and drought, under potential climate change. Illustration on historical data from 1951-2014 over two 5 x 5 spatial grids in the US indicate that SPECD can preserve key marginal and joint distribution properties of precipitation and maximum temperature, and predictions obtained using SPECD are better calibrated compared to predictions using asynchronous quantile mapping and canonical correlation analysis, two commonly used bias correction approaches.

Density correction for multivariate spatial fields of global climate model output using deep learning

TL;DR

A new semi-parametric estimation of conditional densities (SPECD) approach is proposed for density correction of the joint distribution of daily precipitation and maximum temperature data obtained from gridded GCM spatial fields and predictions obtained using SPECD are better calibrated compared to predictions using asynchronous quantile mapping and canonical correlation analysis.

Abstract

Global Climate Models (GCMs) are numerical models that simulate complex physical processes within the Earth's climate system and are essential for understanding and predicting climate change. However, GCMs suffer from systemic biases due to simplifications made to the underlying physical processes. GCM output therefore needs to be bias corrected before it can be used for future climate projections. Most common bias correction methods, however, cannot preserve spatial, temporal, or inter-variable dependencies. We propose a new semi-parametric estimation of conditional densities (SPECD) approach for density correction of the joint distribution of daily precipitation and maximum temperature data obtained from gridded GCM spatial fields. The Vecchia approximation is employed to preserve dependencies in the observed field during the density correction process, which is carried out using semi-parametric quantile regression. The ability to calibrate joint distributions of GCM projections has potential advantages not only in estimating extremes, but also in better estimating compound hazards, like heat waves and drought, under potential climate change. Illustration on historical data from 1951-2014 over two 5 x 5 spatial grids in the US indicate that SPECD can preserve key marginal and joint distribution properties of precipitation and maximum temperature, and predictions obtained using SPECD are better calibrated compared to predictions using asynchronous quantile mapping and canonical correlation analysis, two commonly used bias correction approaches.

Paper Structure

This paper contains 23 sections, 13 equations, 16 figures, 2 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Data and Methods
  3. Overview
  4. Temperature and precipitation datasets
  5. Density correction model for GCM projections at a single location
  6. Extension to spatial data
  7. Conditional density estimation using SPQR
  8. Model evaluation metrics for SPECD
  9. Numerical Study
  10. Overview
  11. Results
  12. Density correction of historical TMAX and PRCP
  13. Overview
  14. Evaluating density-corrected output based on SPECD
  15. Comparison of SPECD with CCA and QM
  16. ...and 8 more sections

Figures (16)

  • Figure 1: Locations in the Southwest (left) and Southeast (right) regions of the US, which form the basis of our bias-correction case study.
  • Figure 2: Monthly mean fields of nClimGrid anomalies (top) and GCM raw data (bottom) of TMAX and PRCP for the Southeast US study region.
  • Figure 3: Density curves of climate model (Mod) and observed (Obs) TMAX and PRCP data, based on a single replicate from the simulation study. PRCP curve is presented on the log scale.
  • Figure 4: Comparison of QM and CCE against two SPECD methods without (SPECD1) and with (SPECD2) spatial correlation. The left column reports Wasserstein distance; the remaining columns report MAE. Lower values indicate greater accuracy.
  • Figure 5: Marginal density curves of climate model (Mod), observed (Obs), and calibrated (Cal) TMAX and PRCP data in the Southwest (SW) and Southeast (SE), pooled across location and month. PRCP results are presented on the log scale.
  • ...and 11 more figures