I am currently working on separately-funded research projects on (i) statistically modeling changes in floods, and (ii) providing decision support to communities making adaptation plans for uncertain future climate change. These two separate projects have piqued my interest in how trend model selection affects climate change adaptation decisions.
My first project focuses solely on modeling how floods have changed over time. For instance, this model can describe how much the magnitude of a flood with a given frequency (e.g., once every ten years on average) has changed due to environmental perturbations, such as climate or land use change, at a given location on a river. One major preliminary finding from my research is that accounting for trends in the variability of annual peak flows can yield substantially high flood estimates downstream of urbanizing areas than more commonly used models that only take into account trends in the central tendency of annual flood probability distributions. This modeling decision can substantially affect estimates of extreme floods, such as the 100-year flood, that are often used in community planning and engineering design. Determining whether or not the modeled trends may persist into the future and adjusting models to reflect different plausible change trajectories further complicates these efforts.
Next, I am designing a decision-support system (DSS) to help communities determine the implications of this kind of trend modeling decisions. There are different ways by which they can decide to use flood trend scenarios. One approach is to determine which scenario is the most likely and then base their climate adaptation decisions on that one scenario. However, many climate change projections vary substantially and many experts have a low degree of confidence in them as well. Due to these limitations, identifying flood protection measures that are robust to a wide range of climate futures becomes preferable. Furthermore, stakeholders may have different model preferences, and the identification of solutions that are robust to a model choice can offer a compromise between stakeholders with different interests.
I am applying my DSS to a quasi-hypothetical community inspired by the town of Exeter, New Hampshire. One objective is to minimize the sum of flood control and flood damage costs incurred during the 21st century. Adaptation options include protective barriers along river reaches, property-scale floodproofing, flood insurance and permanent retreat. Estimates of future extreme flood events that would cause damage are sensitive to the global climate model outputs. (Note that a statistical approach like the one I am using in my other project is more difficult to implement in this idea due to limited streamflow data.) Different model projections of future flood hazards diverge substantially around 2070. However, instead of debating which of the two projections is better, stakeholders with different model preferences might seek a compromise solution that is robust to a wide range of model projections. However, additional optimization modeling may be needed to inform stakeholders, including a satisficing solution that adequately satisfies the stakeholder objectives. These two models do not necessarily yield the same solution, so it is important to consider both perspectives.
One place where trend model selection is currently controversial is coastal North Carolina. Incorrectly choosing a high sea level rise projection may overly reduce economic opportunities while an excessive low projection may result in greater property damage and put more people at risk during storm surges. In 2016, the state will determine the type(s) of statistical or physically-based model(s) that provide a sea level rise projection(s). Previously, the state almost banned all types of models except for linear models fit to tidal observations in 2012. However, some scientists predict that the sea level rise will at a faster rate over the rest of the 21st century than it did during the previous century. Hopefully, there is not only a debate about which projection is the most accurate, but also a discussion about planning approaches that take into account different scientifically defensible model preferences that stakeholders may have.