A Methodology for Robust Multiproxy Paleoclimate Reconstructions and Modeling of Temperature Conditional Quantiles

Lucas Janson, Bala Rajaratnam

Abstract

Great strides have been made in the field of reconstructing past temperatures based on models relating temperature to temperature-sensitive paleoclimate proxies. One of the goals of such reconstructions is to assess if current climate is anomalous in a millennial context. These regression based approaches model the conditional mean of the temperature distribution as a function of paleoclimate proxies (or vice versa). Some of the recent focus in the area has considered methods which help reduce the uncertainty inherent in such statistical paleoclimate reconstructions, with the ultimate goal of improving the confidence that can be attached to such endeavors. A second important scientific focus in the subject area is the area of forward models for proxies, the goal of which is to understand the way paleoclimate proxies are driven by temperature and other environmental variables. In this paper we introduce novel statistical methodology for (1) quantile regression with autoregressive residual structure, (2) estimation of corresponding model parameters, (3) development of a rigorous framework for specifying uncertainty estimates of quantities of interest, yielding (4) statistical byproducts that address the two scientific foci discussed above. Our statistical methodology demonstrably produces a more robust reconstruction than is possible by using conditional-mean-fitting methods. Our reconstruction shares some of the common features of past reconstructions, but also gains useful insights. More importantly, we are able to demonstrate a significantly smaller uncertainty than that from previous regression methods. In addition, the quantile regression component allows us to model, in a more complete and flexible way than least squares, the conditional distribution of temperature given proxies. This relationship can be used to inform forward models relating how proxies are driven by temperature.

Introduction

Understanding historical climate variability is essential for contextualizing modern anthropogenic warming. Paleoclimate reconstruction relies on "proxies"—natural archives such as tree rings, ice cores, and marine sediments—to infer temperatures prior to the instrumental record. Traditionally, these reconstructions utilize regression-based approaches, often variants of least squares, to model the conditional mean of the temperature distribution.

However, proxy data presents significant statistical challenges: it is often noisy, irregular, and characterized by strong temporal dependence. Standard least squares methods, which assume normally distributed and independent errors, are highly sensitive to outliers and heavy-tailed noise. This sensitivity often leads to high uncertainty in millennial-scale reconstructions. Furthermore, mean-based modeling fails to capture how proxies might respond differently to climate extremes.

Methodology: The QUARTS Framework

To address these limitations, this paper introduces QUARTS (QUAntile Regression with Time Series errors). This methodology integrates robust quantile regression with an autoregressive (AR) error structure to provide a more flexible and precise reconstruction.

Quantile Regression and the Check Function

Unlike traditional regression that fits the conditional mean, Quantile Regression (QR) estimates conditional quantiles (e.g., the median or specific percentiles). This is achieved by minimizing the sum of asymmetrically weighted absolute residuals, defined by the "check function" $\rho_\tau(x)$.

The check function $\rho_{\tau}(x)$ for $\tau = 0.75$, illustrating the asymmetric weighting of residuals in quantile regression.

By using the absolute deviation rather than the squared error, QR is inherently more robust to outliers. The asymmetry of the check function allows the model to focus on specific parts of the conditional distribution.

Likelihood and Distributional Assumptions

The statistical foundation of QR can be framed through the Asymmetric Laplace Distribution (ALD). Modeling the residuals with an ALD provides a likelihood-based approach to quantile estimation, where the parameter $\tau$ corresponds to the quantile of interest.

The Asymmetric Laplace Distribution (ALD) density, which provides the likelihood framework for quantile regression.

Handling Temporal Dependence

Climate data is inherently a time series. QUARTS explicitly models residuals as an AR process to account for temporal autocorrelation. Because the resulting optimization problem is non-convex, the authors developed a cyclical block coordinate descent algorithm. This algorithm alternates between estimating QR coefficients and AR parameters until convergence is achieved. To manage high-dimensional proxy data, Principal Components Analysis (PCA) is employed for dimensionality reduction, and a parametric bootstrap framework is used for rigorous uncertainty estimation.

Data and Experimental Results

The methodology was applied to the M08 Proxy Dataset (Mann et al., 2008), specifically a subset of 79 Northern Hemisphere proxies extending back to 998 AD. The model was calibrated against the Climatic Research Unit (CRU) instrumental temperature record (1850–2006).

Uncertainty Reduction

A key finding is that moving from mean-based Generalized Least Squares (GLS) to median-based QUARTS significantly reduces reconstruction uncertainty. The QUARTS reconstruction achieved an average out-of-sample prediction interval width of 0.61 °C, compared to 0.91 °C for the traditional GLS method—a 50% increase in uncertainty for the latter. Residual analysis formally rejected the normality assumption (p-value 2.0%), further justifying the robust QR approach.

Statistical Significance and Historical Clarity

QUARTS identified 43 significant proxy predictors at a 5% level, whereas the optimal GLS model identified only 17. This suggests that the robust nature of QUARTS better captures the underlying signal in noisy proxy data. The resulting reconstruction clearly delineates historical periods such as the "Little Ice Age" with greater precision than previous methods.

Conclusion and Future Directions

The QUARTS methodology demonstrates that modeling conditional quantiles offers a more robust and informative approach to paleoclimate reconstruction than traditional mean-fitting methods. By accounting for non-normal errors and temporal dependence, it provides a clearer picture of historical climate with substantially reduced uncertainty.

Future work may involve expanding the proxy subset through imputation techniques and exploring non-linear relationships between proxies and temperature. Additionally, the ability of QUARTS to model the full conditional distribution offers a promising path for improving "forward models" that describe the physical mechanisms of proxy formation during extreme climate events.

Summary updated

Abstract

Great strides have been made in the field of reconstructing past temperatures based on models relating temperature to temperature-sensitive paleoclimate proxies. One of the goals of such reconstructions is to assess if current climate is anomalous in a millennial context. These regression based approaches model the conditional mean of the temperature distribution as a function of paleoclimate proxies (or vice versa). Some of the recent focus in the area has considered methods which help reduce the uncertainty inherent in such statistical paleoclimate reconstructions, with the ultimate goal of improving the confidence that can be attached to such endeavors. A second important scientific focus in the subject area is the area of forward models for proxies, the goal of which is to understand the way paleoclimate proxies are driven by temperature and other environmental variables. In this paper we introduce novel statistical methodology for (1) quantile regression with autoregressive residual structure, (2) estimation of corresponding model parameters, (3) development of a rigorous framework for specifying uncertainty estimates of quantities of interest, yielding (4) statistical byproducts that address the two scientific foci discussed above. Our statistical methodology demonstrably produces a more robust reconstruction than is possible by using conditional-mean-fitting methods. Our reconstruction shares some of the common features of past reconstructions, but also gains useful insights. More importantly, we are able to demonstrate a significantly smaller uncertainty than that from previous regression methods. In addition, the quantile regression component allows us to model, in a more complete and flexible way than least squares, the conditional distribution of temperature given proxies. This relationship can be used to inform forward models relating how proxies are driven by temperature.
BibTeX
@article{janson2013methodology,
  author = {Lucas Janson and Bala Rajaratnam},
  title = {{A Methodology for Robust Multiproxy Paleoclimate Reconstructions and Modeling of Temperature Conditional Quantiles}},
  journal = {Journal of the American Statistical Association},
  year = {2013},
  doi = {10.1080/01621459.2013.848807},
}
Figure