Arkwright Mutual Insurance Company A Preliminary Investigation of the Trend of Flood Events in The United States
A Research Position Paper of Applied Underwriting Research
Dr. Lixin Zeng Peter J. Kelly
March 27, 1997
Abstract
Property insurers in the United States have experienced significant increases in flood claims last year. While some of these events follow normal seasonal traditional patterns (such as winter rains in California), other events have seemed unprecedented from an historical perspective. An important question for insurers to wrestle with is whether or not these events are statistical aberrations, evidence of a greater than previously understood variability in the peril of flood, or evidence of a trend.
By examining the historical data available from the US Geological Survey, we conclude that indeed, there is a trend toward greater flood risk during extreme events. Put in the vernacular of the insurance industry, there appears to be a trend toward greater severity, despite a lack of trend toward greater frequency.
The implications for insured and insurer are rather obvious. Insureds that heretofore had thought that their facilities were free of flood exposure must now realize that they are exposed to flood losses. These customers must now take seriously the issues of loss prevention and control because insurers will either now or in the future be taking stock of the growing, emerging peril of flood and seek adequate prices for the risk transfer inherent in the insurance contract. Correspondingly, insurers must carefully evaluate their use of large limits as well as other terms and conditions while at the same time prudently using reinsurance protection where surplus positions dictate its use. Presented at the National Association of Real Estate Investment Managers Senior Officer Property Insurance Forum, Boston, March 18-19, 1997.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 1
1. Introduction The impact of human activities on the natural environment has become a great concern among the scientific community as well as the general public. Studies have shown (IPCC, 1995) that the global atmospheric chemical composition and energy balance have significantly changed. Does this mean that the climate is becoming more violent (with more frequent severe weather events) in response to such changes? Are large losses due to weather-related disaster in the recent years an indication of such a trend? Flood, of course, is the focus in this paper. There has been mounting evidence that the global hydrological cycle will be more intense in a warmer climate, leading to more heavy rain and flood events. Research has been devoted to study the frequency of intense rainfall, which shows an upward trend in many parts of the world including the United States (Iwashima and Yamamoto, 1993; Karl et al., 1995). In addition to intense rainfalls, flooding can be caused by other factors such as snow melt and continuous precipitation. Studies have been done to investigate the changes in flood occurrence in a warmer climate on regional scales (Gellens, 1991, for Belgium; Lettenmaier and Gan, 1990, for Sacramento-San Joaquin basin in California). It is found that more flood will occur in a warming climate. The focus of this paper is to study the trends of flood occurrence in the US on a national scale. We analyze the flooding records from over twenty-two thousand river gauging stations in the contiguous USA maintained by the United States Geological Survey (USGS). Many of the records extend back to early 1900’s, with some even to the late 1800’s. The flooding frequency and trends are investigated on national as well as regional scales.
2. Data analysis The flood data are retrieved from the USGS Water Resources Division’s peak flow data set. This data set contains over 22 thousand river gauging stations in the US1. The stations cover different periods of time. In order to perform a valid analysis of temporal trends in flooding, we first screen the data so that all stations cover the same period. After we require that all stations cover the period from 1940 to 1993, only 2432 stations qualify. They nevertheless provide a satisfactory coverage of the mainland US (Figure 1).
1
This data set does not include Alaska and Hawaii.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 2
45 40 30
35
lat.sta
•• •• • •••••• •• •••• ••• •••••• • •• • •• • ••••••• ••• • • • •• • ••••••• ••••••• ••••• • •••• •• ••••• • •••• • ••••• • • • •• • • •• • • •• •• ••••••• • ••• • • •• • • ••••• • •• • ••• • ••••••••••• • ••• •• • ••• •• • •••• • • • •• • • • ••• •• • • •• • • • • • • • • • •• ••• •••• • • • •• ••••••• •• • ••• •••• •••• ••• • • ••• • •• • ••• • • • • • • • • •• • • •••• •••••••••• • •• • •••••••• •• •••••• • • •• • • •• • • • •• • ••••••• ••• •••• • •• •••••••••• • • •••••• •• • • • •• •• • •• • •• • ••••••••••••••••••••••••• ••• •••••••••• ••• ••••••••• • •••• • •••••••••• •••• •• • •••• • • •• • • • •• ••••• • •• ••• ••••••••• • ••• • • • • • • • • • • •• • ••••••• • ••• ••••••••••••••••••••••••••• ••••••• ••••• •• • •• •• •• • •• • •• • • •••••••• • ••• •• • •• •• • • •• •• • • ••• •• •• • •• • • ••••• ••••••••••••••••••••••••• •••••••••••••• ••••••• • • ••••••••• • •• • ••••• • • •• ••• • • • • ••••••• •••• •• • •••••••••••••••••• •••• •••••• • •• ••• •• • • • • • • • •• ••••• •••••• ••• ••••••• •••••••• ••• • • ••• • •• • •••• •••• •• •••••• ••••••••••••••• •••• •••••••••••••• •••••••••••• •• ••• •• • •• • ••••• •••••• ••••••••••••••••••• ••• ••••••••• ••••••••••••••••••••••••••••••••••••••••••••••••••••••• • • • • • • • • • • • • ••••••••••• • • • • • • • • ••• • • •• •••••• • • ••• • •• • •• ••• • • •• • ••• •••••••••• •• • ••• • •• • ••• •• ••• •••••• •• ••••••••••••• ••••• ••• •••• • •• • ••••••• ••••• •• •••••••••••••••••••••••• • • •• • •••••••••••••• •• •••••• •••••• •• •• •• • • • •• ••• ••• ••••• • •••••• •••• • • • •••••••• •••••••••••••• • •• • • •• • • ••• •••••• • •• • •• ••••• ••• •••••••••••••••••••• •••••••••••••••••••••••• ••••• • • ••••••••••• • •••• • • • • • ••• •••••• • •••••• •• •• • •••• ••••••• •••• •••••• ••• • • • • • ••••• ••••••••••••••••• • •• ••••• •• • ••• • • •••• • • •• • •••••••• ••• • •• •••••••••••• • • • • •••••••• ••• ••• • ••••••••••••• • ••••••• • ••••••••••••••••••••• • • • •••• •• • •• •••• ••• •• ••• ••• •••••••••• • •••••• •• •• •• ••••• • • • ••• •••• ••••••••••• • •• ••• • •• •• • •• • • • • • • • • • • • • • • • • • • • • • • •• • •• • •• •••• ••••••••• •••••• ••• • •• ••• • •••• • • • • • • • • • • • •••• • • • •••• ••••• •• • ••• • • • • • •• ••••• • •••••••••••••• • • •• • • •••••• ••••••• •• ••••••••••••• •••••• •• ••••••••• ••• • •• • •••• •• • • • •• •••••• •••••• ••••• ••••••• ••••••• • ••• • •• •• • •• • ••• • •••• •••• •• • •• •••••••••••• •• • •• ••• • ••• •••• ••• • •• •• • • •• • • ••••••••••• • •••• • •• ••• • • ••••• • • -120
-110
-100
-90
-80
-70
lon.sta
Figure 1. The locations of the 2432 USGS river gauging stations surveyed in this study.
6000
Recorded flood events2 are grouped into the years during which they occurred. Figure 2 shows the total number of flood events at all the gauges each year. To detect the existence of any trends in the data, a linear fit of the time series is calculated in a least square sense, which is shown as the dashed line in Figure 2. A slightly upward trend of 4.5 events per year is identified. This number is only 0.6% of the inter-annual variability of number of flood events as measured by a standard deviation of 738 events.
•
•
•
5000
• •
•
•
•
••
•
• •
•
•
• 4000
•
•• •
• •
•
•
•
•
• •
•
•
•
• •
•
•
••
•
• 3000
number of flood events per year
•
•
•
•
• •
•
•
•
•
• •
•
• •
1940
1950
1960
1970
1980
1990
year
Figure 2. Total number of flood events recorded by all of the gauging stations surveyed. The dashed line is a linear least square fit to the data.
2
Defined as the event during which the water level is above the flood stage.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 3
The key question to address next is whether or not there is any trend in the intensity of flood events. To answer this question, we analyze the average amount of stream flow per flood event recorded at the same stations during the flood events each year3. The time series of average flood intensity is shown in Figure 3. Similar approach is applied to detect trends in this data. The best linear fit demonstrates an increase of 330 cfs4 per year. The trend is 5.0% of the inter-annual variability of the average stream flow as measured by a standard deviation5 of 6600 cfs Compared with the trend in flood frequency above, this is a much more significant trend.
•
50000
60000
•
•
• •
•
•
•• ••• • ••• • • • • • ••• ••• • •• ••• •• • • • • • ••• • • •• • • • • • •• • • • • •• • • • • ••• ••• • • • • • • • • • • ••• •• • ••• • • • • •
40000
average flood stream flow (cfs)
70000
• • •
•
•
• •
1940
1950
1960
1970
1980
1990
year
Figure 3. Average stream flow (cfs) per flood event recorded by the gauging stations surveyed. The dashed line is a linear least square fit to the data.
The results in Figures 2 and 3 show that, although there is not a significant trend toward more flood events in the past 50 years, there does exist a trend toward more severe flooding when an event does occur. Regional analyses illustrate that there is great amount of geographical variability in the trends, as demonstrated in Figure 4. The gauging stations with upward trends are plotted as “x” while those with downward trends are plotted in “o”. In fact, the number of stations that demonstrate upward trends out-numbers those with “downward” trends (1426 vs. 987). Regions dominated by “upward” ones include Pacific Northwest and Mississippi River Basin. The persistence of the trend is tested on a longer period of time (1990 - 1993), during which data are available at only 260 stations. Based on the limited amount of data, an trend toward more intense flood still presents. As expected, oscillations with longer periods that do not appear in the 53 year analysis are identified. 3
To account the geographical uneven distribution of stations, we weigh the records on each station based on the density of stations in its neighborhood. The weight function is chosen to be inversely proportional to the number of stations within a 500 km neighborhood. This avoids over-representation of areas with dense gauging stations and vice versa. 4 Cubic feet per second. 5 The standard deviation is calculate after the trend is removed. Same is done for the standard deviation of the number of events.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 4
45 40 30
35
lat
xo oxxoxox ooxooxo xox xxooxx ooo oox xxxx xo o x x xo x oxo o x oo o xxxo oxxoxxxox oo o o xxxxxxx xxxo xx ooox oo x xo x x oxx xoxo o xo xo x oo o xx xxo o x xxx x x x x xx xxxx o xo o oxx o xxxxxx oxox o oxoo ooxx xxxx x o xoxx o x o xxoo x x xx oo ooxoxo o xo o x xx xoxxoxoxxxxoxo xoo oxox o oxooxxo xooxxo x o o x o o xx oxxo o x x o o oo x x o o x x x o x x x x x o o xxoo x xxxoxxo oox o oooo o xxxoxxx oo ooxo oo x o o xoo xxxxxoxxxo oxxxo oo xxx x xxxoxoxx x oo x ooxxxox o o o xox oxo xx o xxxo oooxoo xxxxxxooooxoxxoxoxxoxxoxxxxxoxxxox xxxxo oxx xxo xo o ooxxoxx x x o xo o oo xxoox oxxoo o xo xoxo xoo x xxx o o x oxxxo xxo x o o o o oo o x x x x x o x o x x x x o o o x o o x x ox o oo o xxo x xxxxxoxx oooxox xx oxxo xxxxxxo o oxxoxooxoxoxx xoo ox xo xx o xoxx ooxxxoxoxoo xo oxooooxoxo xoxooxoxxoxoxxxo xx xoxxxoxxxoo xx xxxxx xxxoxxox xooxo xo x ox ox o xxo xooxxxxxxxxoo x xooo x xx oxo ooox xoxooxoxoxxxooxooxxoo xxxxxxxxo xxxxxo o x xo o xxxxo x x xx xxx x xxxoxoxoooo xoxx xxx o xxxo xx x o xxxooxoxxoxxxxxoxxxoxxxooo oxoo ox x xxxoxxxxoxxxoxxxxxxxxxxxxoxoxxxx oxxo xxxxxxoxxx x xxo xxxx oxxxx x xo oo xxxxxoxxxo xo xoxoxxx o o x o x o x x x x x o x x x x o x x x x x x x x x x o o o xox xoxxxxxxxoxo xox xoooo oxxxx xxxxxxxoxxxoxxx x xoo o x xoo xx o ooooox xxxx o xo xoooxxo xoo x oo xxoxxooxoxxo xoo o xo xoxoxxoo xoxooooo ooo xxxxxoxxoxoox oxxo o o oxo oo ox xo o xx x oox oxxxoxxxxx xxxxoxxo x o o x x oo x xx x x x o o x x o o x o xxxxooo x o xxo xxxxo o o x o x x xo xox x x oxxxoo oxoo o oo xxxxxoxxooo oo xxoxxxxoxxoo o oxo ooo xxoo xxxoo xoooxooxooo xoxo xooxx o oxxxxxxoxxo xoxxoxxoxxxxxxxxxxo ooxox xxxx xxx xoxo xoo xxo xxx xxo ooo oxxxxxxxoxxxxxo xxooooxooooxoooooooo oxoo xxxo o xxo xoxx x oxoooox oxxxoxxoxo xoxx ooo oo xo xxoo o oo oooo oo o o o ooo x o xoxxox ooxoxxxxxxxo oxxxxxxoxo o x xxo xxoxxxxxoxo xxo xxxooxoo oo o o x o o xo x o x xxoo ox o o xx xxxxo o x x o o xxoo o x x o o o o x x o x x x o o o o oxx xxx xxxxxx xoxxxooxoxx x xxxx xxxxxoo oxoxxxxooxxo xo oo oooxoox o x x x x o x o o xxxox ooxxoo ooo oooo oxooo xxx x x xxxxoxxx xxxxxoxxo x x o o o x x x x o x o x x x x o xo xo x x xo oooooxoo oo oxx ox x xx xxxoxoxxoxxoxx oo o xxxoxx x x xxxx xxxoxxxxoooxxo xxo x x xx x xx o xxx xxxooxooxxo o oxxoxxxooooo o o xo xxxoxoox x xooxxxxo o xoooxoxo oxx xooxx xx xxxx xxooo o o o o xoxo xxooxooxo x xoo xxxx o xoxxo o xx oxxxxxxxx xo o x x x o o o x x o o o o o o o o o o x x x x x x x x x x o x o xxoxxxo xxoxoxoxxxo ooo x xox xx xx o x x x x x o x o x xxo x x x x ooooo oo oxxxxxx xooxxx oxoo x xxx x x oo oo xo xooxoxxo xooxxx xxxxox xxoxxxoxo ox o oxxoxxxoooo oo xxxxoxx oxoxxoxxxxxxxx ooxxoxo o xx ooo xxxxo o x x ooxo o o xxxxoo o xxoo x x xxo o oo xxoo x o o -120
-110
-100
-90
-80
-70
lon
Figure 4. The trends of flood intensity changes at the surveyed gauging stations. Stations showing upward trends are marked as “x” whereas those showing downward trends are marked as “o”.
3. Summary Based on the analyses of flood records at over two thousand USGS river gauging stations for the period of 1940 to 1993, we conclude that there exists an upward trend for flood risk. This trend is apparently more associated with the increasing flood intensity than frequency. This is consistent with the research findings that although there is no significant overall increase of rainfall amount, frequency of intense precipitation have increased. In addition to the impact of climate change, the global hydrological system is directly feeling the consequences of human activities, including river regulations, land use, deforestation and large scale river diversions. At present, global hydrological processes are going through drastic changes which are accompanied by an increased anthropogenic impact on water balance by the strengthening of hydrologic exchange (Klige, 1990). These anthropogenic impacts on the hydrological system usually result in increased river runoff as shown by regional studies (for example: Sud et al.,1996; Bryukhanov, 1990). On the other hand, the advent of better land use planning and use of detention ponds have successfully reduced flood risk in some regions. In addition, the development of flood water retarding (control) dams and levees tend to constrain rivers and increase flow elevations. This might have reduced the occurrence of small and moderate floods but the potential consequence of possible failures caused by severe floods will be disastrous. Moreover, a strong regional variability is illustrated in our analysis. This makes necessary future studies to investigate regional data in great detail to fully exploit the information contained in the historical data. It also indicates that more robust flood modeling tools based on the state-of-the-art hydrological models are desirable to satisfy the increasing need to study flood on smaller scales.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 5
References Bryukhanov, V.A., Some aspects of anthropogenic impact on channel processes of rivers (by example of Altai), Meteorologiya i Gidrologiya. no.12. pp. 87-92. 1990. Gellens, D., Impact of a CO2-induced climate change on river flow variability in three rivers in Belgium, Earth Surface Processes Landforms, vol. 16, pp 619-625, 1991. IPCC 1995: Climate Change 1995: The science of climate change, Cambridge University Press, UK, 572pp, 1995 Iwashima-T. and Yamamoto-R., A statistical analysis of the extreme events: Long term trend of heavy daily precipitation, J. Met. Soc. Japan, vol.71, pp. 637-640, 1993. Karl-T-R., Knight-R-W. et al., Trends in high-frequency climate variability in the twentieth century, Nature, vol.377, pp. 217-220, 1995. Klige, R.K., Historical changes of the regional and global hydrological cycles, GeoJournal, vol.20, no.2. pp. 129-36, 1990. Lettenmaier, D.P. and T.Y Gan, Hydrological sensitivities of the Sacramento-San Joaquin River Basin, California, to global warming, Water Resources Research, vol. 26(1) pp. 69-86, 1990. Sud, Y.C., R. Yang, G.K. Walker, Impact of in situ deforestation in Amazonia on the regional climate: general circulation model simulation study, Journal of Geophysical Research. vol.101, no.D3. pp. 7095-109, 1996.
Trends in Flood Events
© 1997 Arkwright Mutual Insurance Co.
Page 6