Bayesian Models in Population Projections and Climate Change Forecast

Abstract

The goal of this dissertation is to develop new methods for probabilistic projections in demography and climate science.In the first project, I propose new models for improving estimates and projections of total fertility rates (TFR) for most countries. I develop a new Bayesian hierarchical model for projection that incorporates an additional layer modeling bias and measurement error in estimates of past TFR. In this way, the uncertainty of past TFR estimates is evaluated and incorporated into the projections. The proposed method addresses the issue of data correction across different versions of the UN’s World Population Prospects. The resulting projections are of an appropriate form to be taken as alternatives to results from Alkema et al. [2011], the current default probabilistic projections of TFR produced every two years by the United Nations Population Division. Moreover, the UN Population Division has until now used five-year average TFR but plans to switch to annual values for greater precision. The five-year model is inadequate to represent the autocorrelation of the annual series. I extend the projection model to this situation by adding an autoregressive component on the random distortion to the fertility transition model. I implemented the new model computationally and issued as a major update of the R package bayesTFR, currently used by the UN. The package has been updated to be generic for different settings, and computational difficulties have been resolved by vectorization in the Markov Chain Monte Carlo sampling process. In a separate project, I propose a method for probabilistic forecasting of global carbon emissions, on a country-specific basis. These forecasts are further linked with atmosphericoceanic global circulation models to generate probabilistic forecasts of global mean temperature. By connecting representative concentration pathways to probabilistic forecasts of carbon emissions, I generated the first probabilistic forecasts of global mean temperature. In out-of-sample predictive experiments, I showed that the resulting method provides accurate forecasts and well-calibrated forecasting intervals. I also built a dataset summarizing the country-specific National Determined Contributions in carbon emission reductions signed on the Paris Agreement. By making reasonable assumptions on different scenarios, I compute the probability for major countries achieving their NDC promises, and the additional efforts need to achieve the Paris Agreement goals of limiting global warming by 2100 to 2C or 1.5C.