| Abstract: |
Climate is a long-term issue, and as such, climate forecasts should be
designed with a long-term perspective. These forecasts are critical for
crafting mitigation policies aimed at achieving one of the primary objectives
of the Paris Climate Agreement (PCA) and for designing adaptation strategies
to alleviate the adverse effects of climate change. Furthermore, they serve as
indispensable tools for assessing climate risks and guiding the green
transition effectively. This paper introduces a straightforward method for
generating long-term temperature density forecasts using observational data,
leveraging the realized quantile methodology developed by Gadea and Gonzalo
(JoE, 2020). This methodology transforms unconditional quantiles into time
series objects. The resulting forecasts complement those produced by physical
climate models, which primarily focus on average temperature values. By
contrast, our density forecasts capture broader distributional
characteristics, including spatial disparities that are often obscured in
mean-based projections. The proposed approach involves conducting an
outof-sample forecast model competition and integrating the forecasts from the
resulting Pareto-superior models. This method reduces dependency on any single
forecast model, enhancing the robustness of the results. Additionally,
recognizing climate change as a non-uniform phenomenon, our approach
emphasizes the importance of analyzing climate data from a regional
perspective, providing differentiated predictions to address the complexities
of a heterogeneous future. This regional focus underscores the necessity of
accounting for spatial disparities to better assess risks and develop
effective policies for mitigation, adaptation, and compensation. Finally, this
paper advocates that future climate agreements and policymakers should
prioritize analyzing the entire temperature distribution rather than focusing
solely on average values. |