This work was part of a project to better understand and characterize the risk extreme weather events pose to global deltas, particularly in light of climate change. Because more than half a billion people live in river deltas and more than 80% of the worlds mega-cites are located near river deltas, it is critical to understand the risk of damage, injury and death due to extreme flooding events. This project aimed to develop a set of metrics to better understand this kind of risk. We looked at 48 delta regions-with a combined population of over 340 million-around the world. We initially looked at many variables, including human population size, upstream and downstream levels of development, land use, river discharge, impervious surface fraction, ground water extraction, precipitation rates, inundation, and normalized difference vegetation index. Using un-supervised learning techniques, we were able to cluster deltas and explore the patterns. Some of this clustering work is in a second paper which has been conditionally accepted and is currently under revision.

Our analysis quickly revealed that variables could be grouped based on how they contributed to risk to a delta system. Some variables, such as the ,contribute to the probability of a hazardous event occurring and its severity. These variables were aggregated into a hazard probability “H”. Some variables like population were aggregated into an exposure variable “E”, which measures the number of people exposed to the hazard. Finally, a “V” variable captures the probability that the exposed population will be harmed using factors such as GDP to estimate, for example, whether a system is in place to evacuate an at risk population. The product of these variables R(h) =H(h)E(h)V(h) is considered to be the delta risk for a hazardous event h. This is averaged over possible hazardous events to get the expected risk.

Unfortunately, computing calibrated probabilities for H, E, and V is currently impossible as modeling these events precisely is still well beyond current knowledge and technology, given both the complexity of the events, and the complexity of the delta environment. The data itself is of variable quality. There are very few large scale extreme events, making it nearly impossible to fit accurate probabilities from empirical data. One point the paper makes is that while the absolute risk is very challenging, the change in risk is also of great interest and the factors are potentially easier to identify. For example factors which contribute to a relative sea level rise (RSLR) also increase the population potentially exposed (E’) to a flood event and can more easily be estimated.

As a proxy for probabilities an index was derived H, E', and V. These indices aggregate variables known to be contributing factors to each component of risk by summing using a set of weights, ranking the sums, and then normalizing the rank to obtain the index The weights were perturbed and subjected to a sensitivity analysis to determine the uncertainty in the index. The Investment Deficit Index (IDI) is used as a proxy for V and is derived from GDP and estimates of the ability of the local region to use engineering and investment to protect the population from harm. In the analysis section, the impact of this investment is demonstrated by showing how the index would change with and without this component. Some river deltas-like that of the Yangtze, Mississippi and the Rhine-become relatively less at risk because of investment, while other rivers lacking this investment-like the Ganges-become relatively more at risk. Finally it is supposed that in the future these large investments into event protection may become more difficult due to rising energy costs, increase in interest rates, or other factors which increase their costs relative to other components of the economy. Under such a scenario, deltas such as the Rhine, Mississippi, Han, Tone, and Chao Phraya experience the highest relative increase in risk.