Conceptual Models for the Role and Benefits of EO and Modelling
Project title: Conceptual Models for the Role and Benefits of EO and Modelling Background We investigate the role of future learning about the climate system (by global earth observation and modelling) and about climate thresholds in timing abatement policies. Learning plays a crucial role when irreversibilities or rigidities and large uncertainties are present in the system as in the case of climate change. For computational reasons most multi-stage models have simplified the learning process to an autonomous, perfect one-time learning. We focus on the implications of sequential, potentially active resolution of uncertainty in a simple multi-stage model with a climate threshold. Thereby, the concept of 'value of information' and 'future value of information' can be extended to the value of different sequential learning processes. Thereby, it is hoped to gain qualitative insights into questions like: How do first period optimal decisions with anticipated sequential learning compare to decisions for one-time learning? What is the benefit from resolving uncertainties over time in terms of improved abatement policy? What would the value of tipping point early warning systems be? How does this value depend on the flexibility in abatement and on other system properties? When is anticipation of learning essential? Method (Approximate) stochastic dynamic programming. Results In a first step, the probably simplest sequential decision model possible representing irreversibility, time-lags and a threshold (modified version of the model presented in Maddison(1995)) was used to gain first insights into the questions above. Learning was represented in a simple parametrized form taken from Kolstad (1996) and modified to represent active learning. Besides the sunk-cost – climate-irreversibility trade-off well known from the literature, additional tradeoffs are observed between learning velocity and threshold location and impacts, and between learning velocity and flexibility in abatement. Whether there is a clear correspondence between one-time learning and sequential learning in terms of first period recommendations is yet to be determined. Figure: Adaptive strategies and corresponding temperature increase for climate threshold at 2°C. Green: no learning; blue: most probable learning paths (leading to shown temperatures); black: improbable learning paths. Outlook Planned is a similar investigation in a more realistic model.