California Outcome Likelihood Tool
Download as PPTX, PDF2 likes979 views
This document describes the California Climate Outcome Likelihood Tool, which provides probabilistic seasonal climate forecasts to help water managers. The tool analyzes historical climate data from weather stations to estimate the likelihood that a location will receive a specified amount of precipitation over a future period. It offers different data and analysis options, and provides graphical outputs like probability distributions and spaghetti plots. Limitations include relying on historical records and point-based data rather than areal forecasts, and future work aims to address feedback and expand the tool.
California Outcome Likelihood Tool
- 1. California Climate Outcome Likelihood Tool Nina Oakley, Kelly Redmond, Grant Kelly, Britta Daudert Western Regional Climate Center USDM Forum April 22 2015 – Reno, NV
- 2. Motivations How might this year turn out? NCDC “Amelioration” tool (PHDI) 1 M. Dettinger “ESP” analysis P. Iñiguez PRISM simulations What water managers want: Better seasonal forecasts, monitoring, communication2 Constrain decision-making Tool that is customizable, daily resolution Tool that is graphical, explains caveats3 1ncdc.noaa.gov/temp-and-precip/drought/recovery/ 2Steinemann 2014, BAMS 3Pulwarty 2003, Water Resources
- 3. Data Source NWS COOP via ACIS Daily, long record Decision makers often state climate forecasts unreliable1 Suggest stations with record 1920 or better-present Any station, any record in US Provisions for how much missing data acceptable COOP at Manzanita Lake, CA1Pulwarty 2003, Water Resources
- 6. 3 Data Options “Observed” data in station record Distribution built from all past periods equal to “recovery period” in station record “Analog” Periods Identify periods in station record of date range equal to observed period that are within +/- 1, 2, or 3 deciles of current observed total Take their corresponding future “recovery” periods and create distribution Random sampling of station record Sample each day of “recovery period” from station record, build “synthetic accumulations”, repeat 1000x Apr 13 2001 + Apr 14 1934 + Apr 15 1901 + ... Sept 30 1996 Doesn’t account for multiple day precipitation streaks
- 7. 2 “Analysis” Options Amelioration Likelihood of reaching normal, recovering deficit by end of “recovery period” Custom Threshold Likelihood of receiving a user-specified amount of precipitation by end of “recovery period”
- 9. At RENO 2014-10-01 to 2015-04-12 2.93 in accum. 0 missing days 2.43 in deficit 11.87 in needed by 09-30-2016 to recover deficit, reach normal 15.6% chance of recovering based on 77 of 78 possible periods in record 1937-03-01 to 2015-04-12 “Observed” option Deciles Amount needed Normal for period Probability of occurrence
- 11. “Analog” option At RENO 2014-10-01 to 2015-04-12 2.93 in accum., falls between 10th - 20th percentile 2.43 in deficit 11.87 in needed by 09-30-2016 to recover deficit, reach normal 13% chance of recovering using +/- 1 decile analog (10th-30th percentile), 23 periods Probability of occurrence Amount needed Normal for period
- 12. Example graph + table for Sacramento, CA Likelihood of exceeding threshold of 3 inches by end of WY 2015 (20% chance)
- 13. Spaghetti Plot (coming soon) Observed Precipitation 10-01-2014 through 04-12-2015 + Selected precipitation traces 04-13-2015 through 09-30-2016 = Range of outcomes based on climatology Hypothetical Station Observed period 10-01-2014 to 04-12-2015, Future period ending 09-30-2016 10 traces shown precipitation time
- 14. Limitations and Future Work Using historic record, does not account for future trends Point rather than areal data Test on target audience, implement feedback http://wrcc.dri.edu/col/ Expand to other accumulating quantities snowfall, HDD, CDD, GDD
- 15. Thank you! South Yuba River, Sierra Nevada, CA Feb 7/8 2015 Atmospheric River event Bad for skiing, good for kayaking Photo credit: Ben Hatchett nina.oakley@dri.edu @WRCCclimate Connect with WRCC!
- 17. Random Sampling 2014-10-01 to 2015-03-18 3.09 in 0 missing 0.15 in deficit 6.65 in needed to recover deficit by 09-30 28.6% chance of recovering based on 1000 random daily samples from record 1892-01-01 to 2015-03-19
Editor's Notes
- #3: We are often asked question how might this year turn out. A few tools address this at monthly, clim-div level. Managers want seasonal forecasts. We can’t deliver this but can work off climatology to help constrain decision making Felt it is important to make it flexible and at daily resolution and automated so people can run it as they need rather than wait for updates Pulwarty 2003 suggest graphical interfaces are good for communication of climate info, but need to be usable by managers and contain enough info to communicate caveats. Will work with target user groups to assess these needs are met.
- #4: Uses NWS COOP. Prism-daily doesn’t have long enough record (1981-present), other gridded data sets not good in West or for precipitation Avoided modeled data to keep simple approach, use climatology. Anticipate this will appeal to decision-makers, but has limitations such as future climate trends Suggest stations with record 1920 or better so several major CA droughts in 29-34, 76-77, 1987-1992, current Intended to run for any climate, any station but focus on CA
- #6: Usability features- questionmarks generate text box with useful info in them. Textboxes resizable, moveable. Invalid entries give you highlight of where they occur and description at top of page. A couple options hidden, don’t appear until selected. Rest of presentation we will be talking about NMSU station and current water year.
- #7: Any ideas to create a meaningful random sampling method? Please share. Doesn’t account for consecutive days of precipitation which occur frequently in CA (target area of tool).
- #8: Option I will be showing results for is “amelioration”
- #10: Bug in graphing line in CDF doesn’t extend all the way across
- #12: 3.24 normal 122 observations used in “observed period” observed value 2014-10-01 to 03-18-2015 greater than 78 values in the data/less than 44 values in the data– therefore greater than 64% of values “70th percentile” is values that are greater than 60-70 percent of the observations “80th percentile would be values that are greater than 70-80% of observations and so on
- #14: For those that are curious max year in “recovery” period– 1941, 14.9 in. Lowest 1953 1.95 in Will be able to turn years on and off