Policymaking under deep uncertainty implies policymakers will encounter difficulties when asked to design and implement a new policy. Predicting the future, and deciding on the most probable evolution before applying a policy, has often been unsuccessful, thus the notion of adaptive policies has emerged. By adaptive policies we mean those that focus on short-term planning, and present a robust framework of potential adjustive interventions to choose from when the initial policy is no longer expected to lead to success. Furthermore, within complex systems where environmental policies affect metrics in other sectors, policymaking has become harder, even for the short-term planning. In this report two toolboxes – an adaptation and a mitigation policy toolbox – capable of addressing these aforementioned problems and facilitating robust and successful policymaking are presented.
The central concept of the adaptation toolbox is to allow stakeholders to design adaptive policies, meaning to focus on short-term policy actions followed by a set of adaptive interventions. It produces intuitive visualisations of alternative pathways (adaptation maps – mixture of policies) leading to a desired target, by evaluating a large number of scenarios, created by diverse stakeholders’ perspectives acquired through large-scale consultation. The basic functionality of the adaptation toolbox is based on the notion of model-emulation, by which a statistical model is trained using sets of inputs and outputs produced by a (heavy) model. The resulting emulator can consequently be used for fast and accurate model estimations. This is achieved using a Gaussian Process-based machine learning algorithm achieving: (i) Fast/interactive model approximations and (ii) Quantification of the uncertainties governing the modelling assumptions.
To produce adaptation maps, the adaptation toolbox performs a “cause and effect” analysis to identify triggers that signpost imminent deviations from the set targets. This is done using the Patient Rule Induction Method (PRIM), clustering algorithm which identifies parameters that caused the success or failure of a specific policy instrument. An adaptation map is produced, and two sample pathways are generated according to two different hypothetical perspectives.
The emulation results show that the output parameter trend is correctly predicted by the emulator. More importantly, the estimation of results required only 0.2 seconds of execution time per 10-year scenario, while the per 10-year scenario execution time in the original, bottom-up model was 4 hours. These two observations reveal the capabilities of the adaptation toolbox to produce quick and accurate simulation results.
Currently, the application is equipped with macroeconomic models for Chile, Greece and Poland and allows simulation of policies such as environmental taxes, expansion plans in the electricity generation sector and sector specific mitigation actions, including changes in use of various fuels and products by the household. Thanks to the toolbox policy makers will be able to quickly obtain an initial assessment of the macroeconomic consequences of their proposed policies without the need to engage economists to perform the simulations.