Wednesday, May 19, 2021

Heavy Precipitation Trends

By it's very definition, a warming world will be, well warmer. Surprise! What about precipitation? What about heavy precipitation events? There is a strong theoretical framework to suggest that a warmer world will be wetter overall and that more frequent heavy precipitation events will occur. Of course the world is warmer now than it was 10 or 20 or 50 years ago. We do not need to speculate about the wetter world hypothesis, we can test it. 

NCA4

The 4th National Climate Assessment (NCA4) looked at past and future trends in heavy precipitation by looking at the change in 99th percentile events. Fig. 1 shows a 59-year trend in these 99th percentile events.


Fig 1. Regional change in 99th percentile precipitation events from the 4th National Climate Assessment during the 1958-2016 period. This is Figure 2.6 in the report. 

The methodology of the NCA4 map is shown above is as follows:
  1. Stations were chosen with less than 10% missing precipitation data for 1901-2016
  2. For each station, the 99th percentile threshold of daily precipitation was determined from the 1901-2016 data using only days with at least 0.5 mm of precipitation.
  3. For each station, for each year, the total amount of precipitation falling on days when the daily precipitation exceeds the 99th percentile threshold was calculated.
  4. For each one-degree by one-degree grid box, for each year with available data, the average amount of precipitation exceeding the 99th percentile was calculated for all stations in a grid box.
  5. For each region, for each year, we calculated the average amount of precipitation exceeding the 99th percentile threshold was calculated by averaging all of the grid box values in each region.
  6. For each region, the trend over 1901-2016 using ordinary least squares regression.
  7. The change was calculated as the percentage difference between the end points of the trend line. The end points are 1901 and 2016.
Fig. 1 shows a notable increase of heavy precipitation events in all regions, with the largest increases the farther east you go. The NCA4 map is functionally showing different recurrence intervals at different locations though. Since they are looking at a top 1% precipitation event, if precipitation fell on 100 days per year, it's a value you expect to occur once per year. If precipitation falls 50 days per year, it's a value you expect to occur every other year. Now you are comparing a 1-year event at one location with a 2-year event at another location. This problem is exacerbated the farther west you go. 

If the NCA4 analysis looked at lower percentile events (e.g., 90th or 75th percentile), the temporal problem would go away since practically every location receives measurable precipitation 4+ or 10+ times per year.

Alternate Methodology

A slightly different way of thinking about it is in terms of the days per year that certain precipitation values are expected to occur. If you find the value that occurs on average 5 days per year over a baseline time period, you can count the annual frequency of that event and build a time series to assess whether it is increasing or decreasing. If you do the same thing at a different location 1,000 km away, the trends (not the values) are directly comparable since the time components are the same. 

For this study, I did just that by extracting daily precipitation values for 310 midnight-to-midnight stations in the U.S. and 20 stations in Canada. The period of record in this study is 1951-2020. A 70-year period is more than long enough to make statements about long-term trends. For inclusion in this study, a station must have 65+ complete years during the 70-year period. A complete year means no more than seven missing days.

The establishment of a baseline period is necessary to not only identify an overall trend, but to specifically identify whether the current period is experiencing more of these days than in prior periods. The 1951-1990 period was used to compute a simple average value for the precipitation amount that occurred 3 days per year, 5 days per year, 10 days per year, and 25 days per year.

Example


Fig 2. Sample data for Dallas-Ft. Worth, TX, International Airport.


Fig. 2 shows a sample of how the data for Dallas-Ft. Worth, TX, was computed and evaluated for the precipitation event that occurs 5 days per year. During the 1951-1990 baseline period, every day's precipitation total (including 0.00" and Trace) were collected and sorted. Assuming no missing days, the 200th largest precipitation total during that 40-year period is the value that occurs 5 days per year on average. (Note: adjustments were made for missing days.) In the case of Dallas-Ft. Worth, this precipitation value was 1.41".

Next, we count the number of days in all years that had at least 1.41" of precipitation on a calendar day during the 1951-2020 period. The green line in Fig. 2 shows the annual time series. Finally, we fit a linear regression line to the 70-year time series and measure the difference (as a percentage) between the starting point of the regression line in 1951 and the ending point of the regression line in 2020. In the case of Dallas-Ft. Worth, the dashed black (regression) line goes from 4.15 to 6.72. This ending value is 63% greater than the beginning value.

What I like about this methodology (of course I like my methodology the best!) is that it lets you do a meaningful comparison across both space and time. The NCA4 methodology of using pure percentiles has a time problem. Other studies that look at changes in days per year with 1" or 2" of precipitation have a space problem since those values are rare in some areas and common in others. 

Maps and Interpretation

Now that we have the methodology out of the way, let's look at the maps. Specifically, there are maps for the trend change in the recurrence of the event that historically occurred 3 days per year, 5 days per year, 10 days per year, and 25 days per year. The first four maps (Figs 3-6) show the changes as a percentage at the station level. The next four maps (Figs 7-10) show the changes as a percentage at the state level by interpolating a weighted surface based on station values. The final four maps (Figs 11-14) show the precipitation values that were being evaluated for each time period for each station.

The obvious conclusion to draw is that the medium- and low-frequency precipitation events are becoming more frequent. This is not surprising. Unlike the NCA4 map, which looked at the very low frequency events, this study looks at the rate that slightly more common precipitation events occur and confirms that they are increasing as well.

Precipitation Change Maps (Stations)

Fig 3. Change in the frequency of precipitation events that historically (1951-1990) occurred 3 days per year by station. U.S. average is +30% increase. 


Fig 4. Change in the frequency of precipitation events that historically (1951-1990) occurred 5 days per year by station. U.S. average is +26% increase. 

Fig 5. Change in the frequency of precipitation events that historically (1951-1990) occurred 10 days per year by station. U.S. average is +20% increase

Fig 6. Change in the frequency of precipitation events that historically (1951-1990) occurred 25 days per year by station. U.S. average is +14% increase

Precipitation Change Maps (States)

Fig. 7. Change in the frequency of precipitation events that historically (1951-1990) occurred 3 days per year by state. U.S. average is +30% increase.


Fig. 8. Change in the frequency of precipitation events that historically (1951-1990) occurred 5 days per year by state. U.S. average is +26% increase.


Fig. 9. Change in the frequency of precipitation events that historically (1951-1990) occurred 10 days per year by state. U.S. average is +20% increase.


Fig. 10. Change in the frequency of precipitation events that historically (1951-1990) occurred 25 days per year by state. U.S. average is +14% increase.


Precipitation Amount Maps

Fig 11. Amount of precipitation that historically (1951-1990) occurred 3 days per year.

Fig 12. Amount of precipitation that historically (1951-1990) occurred 5 days per year.


Fig 13. Amount of precipitation that historically (1951-1990) occurred 10 days per year.

Fig 14. Amount of precipitation that historically (1951-1990) occurred 25 days per year.


***********************************

For the Alaska folks, here are charts for several major stations around the state. Stations will be in alphabetical order. No figure captions are given. They are similar to Fig. 2, except that each of the four precipitation categories are shown.


























***********************************************************************

Here are 4-panel plots of stations outside of Alaska. The 45 plots are in alphabetical order by city name.