Extreme Events and Climate Change. Группа авторов
Читать онлайн книгу.
but we may consider fairly mundane mid‐latitude storms even if they are not all that rare. An “impact event” is something like a flood (hydrological event), wildfire (ecological event), pest outbreak (agricultural event), or stock market crash (economic event), also being episodic and far from average, but occurring outside of the climate system.
Why care about this syntax? Just as an extreme weather event need not necessarily result in an extreme impact event, an extreme impact event may happen regardless of what the weather is doing. For example, in warmer climates (i.e., where snowmelt is not a factor) inland floods usually occur under conditions of heavy rainfall over some period of time. But it is also possible for floods to occur for other reasons unrelated to rainfall, such as under a controlled dam release for downstream ecological support or when urban water mains or sewer systems fail. Note also that an extreme weather event (or series thereof) may have long‐term consequences beyond an immediate impact due to destruction of infrastructure. Is it more appropriate then to focus on weather events or impact events? It depends on the purpose. For instance, although Cramer et al. (2014) generally considered their remit to focus on impact events, the assessment with regards to the extreme RFC was explicitly focused on weather events (and the risk implied by their occurrence). This chapter is motivated by the effects of extreme weather, and so the focus will be on that, but we will keep in mind that extreme weather events do not necessarily equate to extreme impact events.
1.2.2. Detection and Attribution
We should clarify a few points about using detection and attribution for understanding before continuing further, even if the term has little to do with extremes or synthesizing per se. Detection and attribution is used to describe the process of comparing predictions of what should have happened in the past and observations of what has actually happened in order to develop a comprehensive documentation of cause and effect (Hegerl et al., 2010; Stone et al., 2013). The predictions should be made based on some understanding of how the relevant systems operate, perhaps based on explicit numerical modeling of the component processes or through extrapolation of empirical relationships. Importantly, the demand on monitoring and modeling is high, such that conclusions are supported by a full wealth of information. However, the flip side is that confident conclusions are not always possible for any of a variety of reasons, including that a specific impact may not have been monitored. Hence, although confident detection of a climate change influence on something can be taken to mean that indeed climate change is having an influence, the lack of a confident detection does not necessarily mean the opposite (Hansen & Cramer, 2015).
As a case study, we will explore the application of detection and attribution analysis using data on the occurrence and impacts of tornadoes in the United States of America. The data are from the Storm Events Database, Version 3.0 (https://www.ncdc.noaa.gov/stormevents/, downloaded May 24, 2018), and is to our knowledge a unique documentation of extreme weather and its impacts. This database is produced by the US National Oceanic and Atmospheric Administration to document the occurrence of extreme weather events and their effects over the United States. Coverage depends on the type of weather event, with the earliest tornado record noted in January 1950. Data include the type of weather event, the county in which it occurred, the intensity of the event, and quantified impacts. We exclude Alaska and US‐dependent territories (e.g., Guam, Puerto Rico, and the US Virgin Islands) from analyses here because of incomplete records or complications from changes in county/borough boundaries. It is important to note that this product is not advertised as being a reliable documentation of trends in extreme weather and their impacts over the past 68 years. We will consider possible issues relating to that later in this section. Nevertheless, the product’s focus on extreme weather events, and its documentation of the weather type, location, and impacts, makes it ideal for the demonstrative analyses to be conducted in this chapter.
Figure 1.1 shows a simple way of diagnosing the contributors to the year‐to‐year variability and long‐term trends of two impacts of tornadoes in the United States. The black lines indicate direct injuries to humans and direct human deaths attributed to tornadoes over the 1950–2017 period according to the NOAA database. The colored lines (other than red) indicate variations in various other factors that may also contribute to the variations and trends in deaths and injuries, all adjusted to the same scale as the historical impact data: the tornado frequency (count of segments, which counts twice if an individual tornado crosses a county boundary or touches down twice), the tornado intensity (approximated by the ratio of the counts of F4 over F1 intensity tornadoes), the national human population (for the states included in the analysis), and the projection of the spatial pattern of tornado incidence onto human population (labeled “spatial pattern” in the figure, reflecting both spatial shifts in human population and shifts in tornado location). A multiple linear regression of observed impacts onto these four driving factors is shown in red.
Figure 1.1 Annual variations in fatality and injury impacts from tornadoes in the United States between 1950 and 2017. Documented fatality and injury impacts are shown in black. Tornado frequency and a measure of average tornado intensity (the ratio of the frequencies of F‐scale 4 to F‐scale 1 tornadoes) are also plotted as measures of the climate hazard, while the total US population and the spatial projection of tornado frequency onto population (at the county scale) are plotted as measures of exposure (Manson et al., 2017). A regression of the documented impacts against the measures of hazard and exposure is plotted in red. The uncertainty ranges of the contributed trends from the various regressed measures of hazard and exposure are estimated by removing the linear least‐squares trends from all regressed time series, resampling the residuals using 1,000 bootstrap samples, adding the linear trends back to these samples, calculating their linear trends, and then taking the 5th to 95th percentile range of the trends. All time series are scaled to the same units as the documented fatality and injury data. Tornado data are from the NOAA Storm Events Database (https://www.ncdc.noaa.gov/stormevents/).
The regression is dominated by the tornado intensity index for both impacts. Visually, the intensity peaks in 1953, 1965, and 1974 closely match the injury and death peaks in those years. However, the decline in injuries since 1980, and the lack of a long‐term trend in deaths, is not matched by the large(r) decline in intensity, which is mainly compensated for by the long‐term trends in event frequency and (in a nonsensical negative sense) by population. Note though that the long‐term behavior of the impacts and hazard data should be treated with caution because of long‐term changes in reporting practice and technology (Gall et al., 2009). For example, the widespread deployment of weather radar in the early 1990s corresponds to an increase in event counts; if radar increased the detection rate of weaker tornadoes, that would also have induced a downward shift in our intensity measure.
There are, however, some broad conclusions we can still take from this analysis. First, tornado intensity is the dominant factor influencing year‐to‐year variations in injuries and fatality risk. Second, year‐to‐year causal relationships may not be the major determinant of long‐term trends in risk; at the very least, population has little short‐term variability but could have doubled the impacts over this period. Finally, the missing driving factor in these plots, namely vulnerability, has likely decreased substantially over this period. Given that suspected biases in the underlying data might have induced a bias toward increasing trends, that population has approximately doubled, and that there is no upward trend in either impact, it stands to reason that a decrease in vulnerability has also played a role. From this cursory analysis we might conclude that there is evidence that trends in tornado behavior have not been a major factor in driving long‐term trends in tornado‐related fatality and injury.
1.2.3. Finding a Common Currency
If