Bayesian Optimisation Against Climate Change: Applications and Benchmarks

TL;DR

This paper surveys Bayesian optimisation (BO) applications in climate-change contexts, organizing four main domains—material discovery, wind-farm layout, optimal renewable control, and environmental monitoring—and introduces LAQN-BO as a realistic, publicly available environmental-monitoring benchmark. It highlights public data sets and example code to facilitate adoption, argues for broader benchmarks beyond synthetic problems, and discusses challenges such as transfer learning and sample efficiency. By comparing benchmarks and outlining future directions, the work aims to standardise climate-themed BO evaluations and accelerate practical deployment. Overall, the paper provides concrete data sets, benchmarks, and guidelines to help researchers apply BO more effectively to real-world climate problems.

Abstract

Bayesian optimisation is a powerful method for optimising black-box functions, popular in settings where the true function is expensive to evaluate and no gradient information is available. Bayesian optimisation can improve responses to many optimisation problems within climate change for which simulator models are unavailable or expensive to sample from. While there have been several demonstrations of climate-related applications, there has been no unifying review of applications and benchmarks. We provide such a review here, to encourage the use of Bayesian optimisation for important and well-suited applications. We identify four main application domains: material discovery, wind farm layout, optimal renewable control and environmental monitoring. For each domain we identify a public benchmark or data set that is easy to use and evaluate systems against, while being representative of real-world problems. Due to the lack of a suitable benchmark for environmental monitoring, we propose LAQN-BO, based on air pollution data. Our contributions are: a) summarising Bayesian optimisation applications related to climate change; b) identifying a representative range of benchmarks, providing example code where necessary; and c) introducing a new benchmark, LAQN-BO.

Figures (5)

• Figure 1: Bayesian optimisation to maximise solar panel power generation.
• Figure 2: Benchmark illustrations. For detailed explanations see \ref{['sec-app-illust']}.
• Figure 3: Planning of wind turbines within limits shown in green. The wind is shown by the blue arrows: wider arrows indicate stronger wind and more undulating arrows indicate more turbulence. In the layout on the left one turbine is directly downwind of the other, resulting in weaker and more turbulent wind. To the right, both turbines get unhindered wind.
• Figure 4: Adjusting solar panels to optimise power generation as the sun moves across the sky. The benchmark optimises the sum of direct, diffuse and reflective sunlight.
• Figure 5: Example LAQN-BO problems from the training set (2015, top) and test set (2016, bottom). Not all sensors are available on every day. The overall pollution level varies, compare the top left and right plots. The locations of maxima also vary, compare the right top and bottom plots. Most of the sensors are clustered together in central London.