8/2/2021 HARP Weekly Work Session

[rburghol]

• discuss the example length of drought and ET > P as f(t)
• Is the no-precip days chart using year - min(year) as x value, so our coefficients aren't made smaller
• Should we do year - min(year) ?? (RWB Thinks so, but wants other perspectives)
• How do we judge the power of these charts? R^2m, p-value, other?
• Other statistical alternatives?
• In addition to north and south, what other cases do we want?
• Let's map this sucker.
• How can we have less days with ET > Precip than consecutive days without precip? Did the chart on the right have total # of days without precip?
• Did USGS report (that we cited in SWRP) not find trends in drier conditions because they were looking at monthly totals for precip, and that is essentially limited by the number of days in a month?

[katealbi11]

I just fixed the graphs of no precip days. I accidentally graphed total no precipitation days instead of max consecutive no precip days!

Judging Power of Charts:

• the R^2 value should be as close to 1 as possible
• t-value should be MUCH greater than 2 ( like in the 100s)
• p-value should be generally less than 0.05
• can also use further analysis measures such as a Cook's Distance plot (single data point has a significant influence on the data and should be removed) and tests for homoscedasticity (constant residuals), bias, and normality
  • Cooks Distance plot should show no outliers or points with significant influence
  • test for homoscedasticity should show a graph with little to no pattern IF LINEAR RELATIONSHIP
  • test for bias should show little to no pattern IF LINEAR RELATIONSHIP
  • test for normality should show a linear relationship to guarantee normality of residuals IF LINEAR RELATIONSHIP
  • Presentation shows examples of each of these plots

[rburghol]

I would agree with all of those with the exception of R^2. Literally, R^2 says what % of variability in dependent variable can be explained by changes in independent variable, and thus the only way for R^2 to be close to 1.0 would be if the independent variable, in this case year (time), is the only thing that influences #of dry days.

Thus, I think that our expectation of possible maximum R^2 has to be scaled according to our sense of what other factors may play a role in causing the number of consecutive dry days to increase or decrease from year to year. Just based on natural observed variability, and knowing that things like el niño/la niña are believed to cause wetter/drier conditions in our area I would say that the max R^2 here might be very small.

As for the follow up tests -- that looks good, I would also like to see ~~the consecutive days~~ the total rainless days versions -- sometimes errors can reveal things of interest!

Thanks! Talk soon, rb

On Mon, Aug 2, 2021 at 12:09 PM katealbi11 @.***> wrote:

Judging Power of Charts:

• the R^2 value should be as close to 1 as possible
• t-value should be MUCH greater than 2 ( like in the 100s)
• p-value should be generally less than 0.05
• can also use further analysis measures such as a Cook's Distance plot (single data point has a significant influence on the data and should be removed) and tests for homoscedasticity (constant residuals), bias, and normality
  • Cooks Distance plot should show no outliers or points with significant influence
  • test for homoscedasticity should show a graph with little to no pattern IF LINEAR RELATIONSHIP
  • test for bias should show little to no pattern IF LINEAR RELATIONSHIP
  • test for normality should show a linear relationship to guarantee normality of residuals IF LINEAR RELATIONSHIP
    • Presentation https://docs.google.com/presentation/d/1St1jl4Rfzc_PJdXWYGzqSwLhOf5Oat0jsNjyjFcyoB4/edit#slide=id.ge558a2b86f_0_26 shows examples of each of these plots

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/HARPgroup/HARParchive/issues/109#issuecomment-891148604, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABC4AIWA46NGGFRBXBS6JZTT227LJANCNFSM5BI2L6JA .

--

Robert W. Burgholzer 'Making the simple complicated is commonplace; making the complicated simple, awesomely simple, that's creativity.' - Charles Mingus

[snhandelsman]

What constitutes a drought? meteorological drought - lack of precipitation agricultural drought - lack of soil moisture hydrologic drought -reduced streamflow or groundwater levels (USGS)

https://www.ncdc.noaa.gov/extremes/cei/graph/se/01-12/5 <-- extremes in days with no precipitation in southeast yearly trends (NOAA) https://www.ncdc.noaa.gov/extremes/streaks/mapping/VA/500/all-time <-- most consecutive days w no precipitation (by state) https://www.ncdc.noaa.gov/temp-and-precip/drought/historical-palmers/psi/198901-202012 <-- animation of droughts through years (whole US)

• 2012 US drought was analyzed using NLDAS data
• trends showed drought precedents 6 weeks before event occurred (soil moisture NLDAS and crop conditions ESI)
• https://www.sciencedirect.com/science/article/abs/pii/S0168192315300265?via%3Dihub

[snhandelsman]

• mean evapotranspiration averages (yearly) in North America
• 45 in per year in puerto rico
• 7.6 inches per year in alaska
• southeast --> about 35 inches per year, 70% of total precipitation
• hot weather can cause less evapotranspiration in some cases -> plant stomata close to keep water in
• agricultural ET significantly less than native plants ET

https://www.ntsg.umt.edu/project/modis/mod16.php

• uses satellite data to track global evapotranspiration
• penman-monteith equation

[katealbi11]

Produces code for 7-day, 30-day, and 90-day rolling averages for precipitation and temperature. Do we want to add PET to this as well?

[rburghol]

PET yes, but most importantly the difference between PRECIP and PET, i.e., PRECIP - PET

On Tue, Aug 3, 2021 at 2:10 PM katealbi11 @.***> wrote:

Produces code https://github.com/HARPgroup/HARParchive/blob/master/HARP-2021-2022/day_avgs%20(1).Rfor 7-day, 30-day, and 90-day rolling averages for precipitation and temperature. Do we want to add PET to this as well?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/HARPgroup/HARParchive/issues/109#issuecomment-892056149, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABC4AIWLX6SVBW2W7IPAH5TT3AWLDANCNFSM5BI2L6JA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .

--

Robert W. Burgholzer 'Making the simple complicated is commonplace; making the complicated simple, awesomely simple, that's creativity.' - Charles Mingus

[katealbi11]

Updated rolling averages code with PET and water deficit (PET-Precip)

[katealbi11]

Linear Regression w/ Confidence Intervals Graph for Annual Precip: