Climate processes

Abstract:

Aerosol–cloud interactions (ACIs) are considered to be the most uncertain driver of present-day radiative forcing due to human activities. The nonlinearity of cloud-state changes to aerosol perturbations make it challenging to attribute causality in observed relationships of aerosol radiative forcing. Using correlations to infer causality can be challenging when meteorological variability also drives both aerosol and cloud changes independently. Natural and anthropogenic aerosol perturbations from well-defined sources provide “opportunistic experiments” (also known as natural experiments) to investigate ACI in cases where causality may be more confidently inferred. These perturbations cover a wide range of locations and spatiotemporal scales, including point sources such as volcanic eruptions or industrial sources, plumes from biomass burning or forest fires, and tracks from individual ships or shipping corridors. We review the different experimental conditions and conduct a synthesis of the available satellite datasets and field campaigns to place these opportunistic experiments on a common footing, facilitating new insights and a clearer understanding of key uncertainties in aerosol radiative forcing. Cloud albedo perturbations are strongly sensitive to background meteorological conditions. Strong liquid water path increases due to aerosol perturbations are largely ruled out by averaging across experiments. Opportunistic experiments have significantly improved process-level understanding of ACI, but it remains unclear how reliably the relationships found can be scaled to the global level, thus demonstrating a need for deeper investigation in order to improve assessments of aerosol radiative forcing and climate change.

Abstract:

We show that the standard computational pipeline of probabilistic programming systems (PPSs) can be inefficient for estimating expectations and introduce the concept of expectation programming to address this. In expectation programming, the aim of the backend inference engine is to directly estimate expected return values of programs, as opposed to approximating their conditional distributions. This distinction, while subtle, allows us to achieve substantial performance improvements over the standard PPS computational pipeline by tailoring computation to the expectation we care about. We realize a particular instance of our expectation programming concept, Expectation Programming in Turing (EPT), by extending the PPS Turing to allow so-called target-aware inference to be run automatically. We then verify the statistical soundness of EPT theoretically, and show that it provides substantial empirical gains in practice.