Drought Status Assessment Report

[rburghol]

Drought Report @julieshortridge @COBrogan @ilonah22 @willprokopik @benhdye
See task list in #24

Report Components

• Intro
• Estimated baseflow parameters for gage in question.
• Characteristic baseflow events for the gage in question.
• Projections for the current timespan.
• Appendices:
  • Geology
  • Model calibration / methodology (we are using the HSPF paradigm to describe USGS flows and also evaluate the parameterization of the model)

Sample Report

Introduction

We intend to use hydraulic principles and numerical models to quantify baseflow storage characteristics and recession during periods of drought, for the purpose of estimating low flows over a period of 1 to 18 months into the future. The concept of baseflow in an unconfined surficial aquifer is governed by principles of hydraulic head flow through a porous media, and the terms below describe the relationships used in this analysis:

• "Baseflow" is the flow in a stream that comes from the slow discharge of water that has infiltrated deep into the surficial aquifer, and is released slowly over time. The rate of base flow is governed by a deterministic relationship between the head of stored water in the aquifer, and the discharge properties in the porous medium.
• "Baseflow recession" is a term that describes how the flow rate will naturally decrease during dry periods, as the head in the subsurface storage compartment decreases due to discharge of water without being replenished due to infiltration. In essence, each day that the aquifer discharges without replenishment will result in a drop in hydraulic head, which will in turn reduce the outflow by some measurable amount.
• "AGWRC" (Active Groundwater Recession Coefficient) is a decimal value that can describe a simple linear storage/outflow relationship in a watershed. We will be using a simple linear relationship wherein AGWRC varies between 0.0 < 1.0, with the majority of streams in Virginia having a coefficient of between 0.90 and 0.99. Figure 1 below shows the recession from a starting flow of 100 cfs, with the number of days for flow to reduce by 50% ("half-flow") indicated on the curve. In general, the higher the value for AGWRC, the more persistent the base flow discharge rate (Note: a watershed with an AGWRC of 0.999 would take 693 days for base flows to reduce from 100 to 50 cfs).
• "Simple Recession" The daily reduction in flow can be represented by a simple coefficient provided that hydraulic characteristics over the range of storage head are uniform.
• "Complex Recession" In watersheds with more complex geology and hydrology, multiple recession coefficients may be required to represent baseflow characteristics accurately.
• "2 point analysis" is a simple calculation that estimates the recession coefficient based on a starting flow rate (point 1), an ending flow rate (point 2), and a time in days between the 2 points. This method is simplistic in that it assumes zero recharge or storm flow influence during the intervening period.

Image 1: Theoretical baseflow recession curves for example AGWRC values from 0.99 to 0.90. Number of days for 50% reduction in flow to occur are indicated along the plot line.

We will combine analysis of observed stream flow, with numerical models of baseflow recession, and an analysis of the geological properties in the sub-surface aquifer to help us develop predictive relationships between current base flow rates and future baseflow rates in the absence of baseflow recharge. The main focus will be to determine baseflow recession coefficients under low baseflow storage conditions, however, attention will also be given to the potential for variations in baseflow recession rates at different levels of active groundwater storage. These relationships will then be used to describe worst case scenario flows into the future, should dry conditions persist for an arbitrary duration into the future.

Results/Analysis of Baseflow Parameters

Example 1: Baseflow Event Summary at Strasburg USGS gage.

• USGS Gage ID: 01634000
• River Segment: PS3_5100_5080
• Deep dive on 2002 February case study - high AGWRC when base flow stores are low
  • Define a function to calculate AGWS when Qout and AGWRC are known
  • What is estimated storage under different apparent AGWRC?
  • Leverage the usgs basin precip timeseries generated in HARP2024 for each basin to provide context for event analyses (already living in CSV files)
  • Identify other low AGWS events
• Current estimate of AGWRC under drought conditions: 0.975 (28 days for baseflow to recede 50%)
  • Analysis of computed AGWRC yields 0.975 as median for flows under 200 cfs (eyeballed, need to verify).
  • "2-point" analysis of drought of 2002 indicates flow dropped from 206 cfs on 6/20/2002 to 70 cfs on 9/6/2002, 67% decrease in 78 days, with an estimated AGWRC of 0.985.
  • "2-point" analysis of drought of 2023 indicates flow dropped from 110 cfs on 7/15/2002 to 55 cfs on 8/29/2023, 50% decrease in 45 days, with an estimated AGWRC of 0.985.
  • Note: While the estimated "2-point" recession coefficients for droughts in 2002 and 2023 are considerably higher than those estimated with the event analysis shown below, there are considerable small storm pulses evident in the hydrographs (and missing data in 2002), such that taking the 2-point estimate at face value could lead to overestimation of the recession coefficient. This caution is supported by the chart in Figure 2c, where it can be seen that the AGWRC=0.975 line follows base flow shape more closely than 0.985, and that there are clearly multiple small recharge events occurring during the summer that are shifting the starting point of recession higher.

Image 2a: Drought of 2002 in Strasburg. Note: gage adjustments required the use of estimated data (code: A e) from 7/23/2002 to 8/12/2002, so inferences about baseflow, and potential recharge in the intervening period cannot be made with certainty. Minimum observed flow approximately 70 cfs (after attempting to factor out water withdrawal influences).

Image 2b: Drought of 2023 in Strasburg. Minimum observed flow approximately 55 cfs (estimated after attempting to factor out water withdrawal influences).

Image 2c: Drought of 2023 in Strasburg with projected baseflow lines from the overall gage assessment (0.975) and the 2023 drought "2-point" assessment (0.985).

Image 2d: Minimum projected flow for fall 2025, in the absence of recharge events.

Image 2e: USGS BFS forecast of minimum projected flow for fall 2025, in the absence of recharge events. (from https://wa.water.usgs.gov/projects/baseflows/BFS_forecast_index.html )

Image 3: Scatterplot of computed AGWRC as a function of flow for all flows in the Strasburg VA USGS gage record. 8 events with numerical relationships indicative of baseflow conditions are presented, over a wide range of antecedent flows. (for more detail see Strasburg events 7, 8 142, 141, 194, , 196, 203, and 204 in Appendix B).

Image 4a-d: Baseflow event, by season, $AGWRC_{season} = f(Q)$

Image 5a: MLLR assessment in winter of 2002 in Mount Jackson Watershed (obtained from mapserver with this URL )

Appendices

Geology

Drought conditions are closely related to low flow conditions in rivers. When the rivers are in these low flow conditions, they are dominated by baseflow. Baseflow is the steady input of water into a river from the groundwater source that feeds the river. Because this flow is influenced by groundwater, it is important to understand the local area's geology. This project analyzed geology through multiple viewpoints, including geologic maps, previous USGS reports on the area, and raster data.
Using the Virginia Department of Energy’s Geology and Mineral Resources online geologic map, the geologic formation that each USGS gage of interest is in was recorded. This allowed the general geology of each gage location to be analyzed. Both Mount Jackson and Coote’s Store are in the Conococheague formation, which is primarily composed of limestone and secondarily composed of dolostone. Strasburg is in the Martinsburg and Orando formations, which are composed primarily of shale, and secondarily composed of sandstone.
Thus, the geology between Cootes Store and Mount Jackson is quite different from Strasburg. Cootes Store and Mount Jackson are in a landscape that is karst influenced, and limestone typically has a high hydraulic conductivity for rock. Strasburg is on a shale basin, which has a much lower hydraulic conductivity than other sedimentary rocks. Therefore, the porosity at which water permeates in and out of soil and rock may be different at each site, creating different interactions between runoff, bedrock, and baseflow.
However, this only accounts for the physical location of each USGS gage. The actual drainage basin of the watershed may have different underlying geology. For example, Cootes Store’s drainage basin is to the west of the gage, which is an area of composed of a few different geologic formations that are primarily made of shale, which has very different hydraulic properties than the limestone underneath Cootes Store’s gage. The difference between the geology of the location of the gage and the drainage basin of the river needed to be investigated.
To better understand the interactions here, raster data was used to study the watersheds as a whole.

In addition, current USGS reports that have already studied the geology of the area referenced. A USGS SIR report has already studied the bedrock structure of the Shenandoah Valley. This paper has detailed information on the engineering properties of the bedrock in the Shenandoah Valley and how it was derived, as well as information on the influence of the karst landscape, and models to estimate groundwater recharge.

Model Methodology

• Model calibration / methodology (we are using the HSPF paradigm to describe USGS flows and also evaluate the parameterization of the model)
  • Recession "Event" identification: based on some period of successive days where mean change in AGWRC is near 0.0
  • Recession "Event" isolation: isolate points from event that pass strict criteria
    • $|dC_{AGWR}| &lt;= 0.03$; minimal change in computed recession coefficient ($C_{AGWR}$
    • $C_{AGWR} &lt; 1.0$; must be true to qualify as "recession"
  • Recession "Event" analysis:
    • Ask if there is a trend in $C_{AGWR}$ over time. (there should ne NO trend, if we really have found base flow)
      • Do lm() of valid points isolated, $y = mx + b$
    • 2 options:
      • Flow Decay Linearization
        • Apply log transformation to get decay event in linearized form
        • $Q=Q_0 * AGWR^t$ -> $log(Q) = log(Q_0) + t* log(AGWR)$
        • $m=log(AGWR)$
      • Linear analysis of computed AGWRC
        • $y = c_{AGWR}$; y = computed recession coefficient for a given day
        • $m ~ 0.0$; m is effectively 0.0, showing no significant slope through valid points
        • $b = C_{AGWR}$ C = effective AGWRC for the event in question.
        • this gives us 95-99% confidence interval -- maybe a reason to prefer/supplement to log(Q) approach?
  • ANOVA of different events
    • Group by flow range
    • Group by season
    • Are they significantly different?
• Analysis of baseflow parameters from USGS during diverse hydrologic periods
  • "Cloud" vs "Event": majority of previous analyses have lumped all data into a "cloud", to determine a single $C_{AGWR}$.
  • Our event approach enables us to identify potential storage dependent variability in $C_{AGWR}$ (Schultz, and others).
  • Event approach also allows us to more effectively filter out false positives for recession
  • Model Validation:
    • Do forecasts from computed $C_{AGWR}$
    • Ask question, what % of future flows are over/under-predicted?
    • What is margin?
• Forecast of drought flow out to 30-480 days. (see: Future Drought Predictions Using Model and USGS Data HARParchive#1435)
  • How much baseflow storage do we have? (in inches, BG)
    • Can we indicate this estimated AGWS on a 2nd y-axis in Image 1
  • Use results from regression and ANOVA to set MOS for 30-480 day flow prediction.

Questions to Answer

• How do we discuss modeled versus USGS?
  • Like, really, our main deliverable is a projection based on USGS
  • Can we also do a projection based on drought mashup model?
  • Do they agree?

@COBrogan

[benhdye]

Draft link:
https://virginiatech-my.sharepoint.com/:w:/g/personal/benjamindye_vt_edu/EWl1OAYOwCNAveL7gjMliScBDpMeDsALdbp-fZfFiDVwLg?e=AMqnn6

@rburghol @COBrogan
Just want to make sure this is the style and direction we are going for

[willprokopik]

Mount Jackson Case Studies (work in progress, will continue to update)

Mount Jackson 2002 (year of interest) event 94
This event occurs in mid-end February, lasting 15 days. As this is a late winter event, ET is expected to be very low so streamflow changes are dominated by groundwater discharge. The R-squared values of 0.933 for the log(Q) fit indicates very strong exponential behavior. The predicted AGWR value of 0.975 is also very strong. The flow plot shows a consistent monotonic decline with no storm pulses, which helps validate the claim that this would be a "good baseflow recession event." AGWR values are consistently < 1.0, and dAGWR values fluctuate around 1.0 and are mostly within the 0.97-1.03 range with a few exceptions. It is also interesting to note that this is the only event that was identified for this year.
Still need to perform projections for this event.

Mount Jackson 2023 (year of interest) event 151
This event primarily occurs throughout the month of April, so we would expect ET to normally start increasing around this time (vs winter). The flow plot is interesting, as it has a clear peak just before the start of the identified event. This may be a good example of stormflow recession transitioning into baseflow recession. The decline is mostly smooth (especially after possible "stormflow recession"), with a small bump towards the end indicating possible slight disturbance. The early AGWR/dAGWR values show some noticeable swings, and there's fluctuation at that later bump as well. Otherwise, these values are fairly consistently within the proposed range. The R-squared on log(Q) is very strong, aligning with the stable evidence in the flow chart (excluding the start of the event), and the predicted AGWR value is also strong at 0.958. The central ~3 week timeframe of this event demonstrates a solid baseflow recession while the early/late days show non-stationary behavior. This is one of two identified events for this year.
Still need to perform projections for this event.

Mount Jackson 2023 (year of interest) event 152
This event is 19 days long and occurs in the first half of the month of June. With ET on the rise, we might expect to see more day-to-day disturbances in the data. The flow chart shows an overall decline with a couple of very small disturbances (slight bump, slight flattening out). The AGWR values are largely < 1.0 (or right around), but the dAGWR values see a lot of swings sending them out of the 1.03-0.97 band. The R-squared value is slightly lower than event 151, but it still sits strong at 0.930. The predicted AGWR value of 0.966 is logical for the given context. The majority of this event demonstrates evidence of clean baseflow recession with the exception of the dAGWR spikes (given the time of year, this might be OK). This is the second of two total events for this year.
Still need to perform projections for this event.

Mount Jackson event 99 (interesting event)
This event occurs during the first 24 days of the month of February. This is an event that stood out to me earlier too, because it shows absolutely no consistent "recession." I think this can definitively be grouped as "not baseflow recession." Maybe we could call this baseflow, or a baseflow plateau, or even just a disturbed segment. It's R-squared for log(Q) is almost 0, which is incredibly weak. Additionally, the predicted AGWR is > 1.0 and the dAGWR values are very volatile. A couple of potential pattern causes: Rob mentioned something about ice previously, this event starts in early February, maybe there's a connection there with a related measurement issue. could also just be a case of event segmentation, our system latched it onto a baseflow plateau as opposed to a decay.
Still need to perform projections for this event.

Mount Jackson 2025 (year of interest) event 156
First of two events in 2025, occurring mostly in late February (expect low ET). The flow plot demonstrates a strong monotonic decline, this event is probably a stormflow recession -> baseflow recession example given the initial sharp decline followed by a more mild decline. The AGWR is in range every day apart from one with no real pattern. dAGWR begins low and rises closer to 1.0 as the flow values begin to drop at a slower rate. The R-squared on log(Q) is pretty strong at 0.972, and the predicted AGWR is steep at 0.926. Based on these characteristics, this event is definitely a recession, it's just a matter of what we categorize it (stormflow recession vs baseflow recession).
Still need to perform projections for this event.

Mount Jackson 2025 (year of interest) event 157
This is a very convoluted event that was grabbed. This is the second and final event identified in 2025, and there's a lot going on. It occurs in April, so we can expect ET to be increasing (can cause noise). The flow plot has a strong early drop characteristic of stormflow recession, with a fairly stable middle section, ending with a bump on the last few days. AGWR values are mostly < 1.0, but dAGWR experiences frequent out of band fluctuations, especially at the beginning and end. The predicted AGWR number is solid (0.969), but the R-squared of only 0.701 is a little concerning. Generally speaking, that number isn't alarmingly low, but in the context of other events, it's noticeably lower. I'm unsure of what category this event would fall into.
Still need to perform projections for this event.

[ilonah22]

Strasburg Case Studies

The 8 case studies that I chose for our years of interest are in red below:

The one with a very low AGWRC is event 203 from 2025

1930s

Event 7
Duration: 30 days
Season: Spring
AGWRC: 0.964
R-Squared: 0.921

Event 7 seems to be influenced by storm recession until about March 27th. After that there is no clear trend in AGWRC and d_AGWRC for the rest of the event, which is ideal.

---

Event 8
Duration: 20 days
Season: Spring
AGWRC: 0.967
R-Squared: 0.941

Event 8 does not have an obvious influence from storm recession. Some days at the end of the event have and AGWRC over 1, which we have discussed filtering out. This is probably the cause for the slightly lower R-squared.

---

2002

2002 only had 2 events, one in January and one in February.

Event 141
Duration: 22 days
Season: Winter
AGWRC: 0.977
R-Squared: 0.966

Event 141 has no obvious trends in AGWRC or d_AGWRC. This appears to be decent event capturing despite some days with am AGWRC over 1.

---

Event 142
Duration: 16 days
Season: Winter
AGWRC: 0.993
R-Squared: 0.857

The AGWRC for event 142 seems to alternate between positive and negative (compared to 1). Since there are quite a few days with a higher AGWRC, the R-squared is lower.

---

2023

2023 had 6 different events from January through November over a total of 114 days.

Event 194
Duration: 25 days
Season: Winter
AGWRC: 0.974
R-Squared: 0.964

Event 194 includes a bit of storm recession. However, overall, there is a high proportion of days where both the AGWRC and d_AGWRC fall within the set range. There is only one day where the AGWRC is above one, which is pretty good for the types of events that are usually captured.

---

Event 196
Duration: 26 days
Season: Spring
AGWRC: 0.970
R-Squared: 0.978

Event 196 also has a lot of days where AGWRC and d_AGWRC are valid, which results in the higher R-squared.

---

2025

So far, 2025 has had 2 events, one in February and one in April.

Event 203
Duration: 14 days
Season: Winter
AGWRC: 0.923
R-Squared: 0.977

Event 203 has a few days of storm recession. It also seems to have a slight trend in AGWRC toward the end of the event, which we don't really want to see.

---

Event 204
Durations: 14 days
Season: Spring
AGWRC: 0.965
R-Squared: 0.990

Event 204 has a very high R-squared value and most of the days of the event have valid AGWRCs and d_AGWRCs.

[benhdye]

Coote's Store Case Studies:

1930s
event 3
R squared: 0.94
Calculated AGWR: 0.942
Selected due to being in a time of interest

---

1941
event 15
R squared: 0.998
Calculated AGWR: 0.893

Selected due to very low flow and very high R squared

---

1969
Event 47
R sqaured 0.999
Calculated AGWR: 0.948
selected due to essentially perfect R squared value

---

1976
Event 62
R Squared: 0.988
Calculated AGWR: 0.913
Selected because every point was in the threshold

---

1999
Event 115
R Squared 0.763
Calculated AGWR: 0.965
Selected due to long event that looks good but has lower R squared value

---

Interesting enough, I had no event recorded from 2002
2023
Event 187
R squared 0.993
Calculated AGWR: 0.897
Selected due to being highest R squared value in year of interest

---

2025
Event 190
R squared 0.973
Calculated AGWR: 0.902
Selected due to being highest R squared in year of interest

[rburghol]

He @ilonah22 @willprokopik @benhdye excellent case studies from all. Some quick thoughts individually and collectively:

• @willprokopik - excellent choice of "interesting" events in no. 99 -- I suspect that this may be a rainy wet period, but who knows, it could just be an example of how the aquifer might respond differently under different levels of "fullness". I would like to see this event (and the others for that matter) in the context of the AGWRC = f(Q) charts that @ilonah22 developed which I think are probably on their way to being functionalized.
• @ilonah22 - I think you made a very astute observation in that we don't want to see a slope in the AGWRC plot. I think that this could actually form the basis of filter criteria. That is, if we have events that have a non-zero slope for AGWRC, we might discard them from analysis on the basis that they are not actually baseflow. That said, for this particular event, the slope in question appears asymptotic to my eye, with the "true" AGWRC being the asymptote line. Food for thought in terms of filtering and analyzing events.
• @benhdye your observation that there were no events in 2002 is super important. This is a pretty legendary drought period (though it might not be the most severe in Shenandoah). It has been said that it is a funny drought to understand because it was characterized by super low base flow recharge in the winter and relatively frequent summer rains. Because winter base flow recharge was so low, the rains would boost flows temporarily, then it would fall right back down into base flow -- but if there were in fact frequent rains in Shenandoah, it might be difficult for our routines to detect them. I think that we crafted our event analysis such that we can pass in any fragment of flow timeseries (if not we need to do so asap), and I would like to have you use visual inspection to identify a period that looks like baseflow to you in 2002, and then pass in the timeseries for the dates you identify to the detailed analysis routine and see what it shows. You may also pay attention to late 2001, and December through early 2002 period as I said it was in fact a dry winter.

All -- 2 crucial next steps:

• Generate the scatterplots of $AGWRC = f(Q)$ for your locations
  • Make sure the scatterplot funcdtions return an image object, so that you can add a point or points to highlight the specific event that you are analyzing in the context of the larger scatterplot so that we can see if the computed AGWRC is an outlier
• Make sure that the event analysis functions can be passed in an arbitrary timeseries period that you hand select (see my comments to @benhdye ) -- I will note that the paucity of events in 2002 is super interesting, and bears futher investigation
• Get the projected flows in there. It would be nice to see multiple projected flows overlayed over one another especially when you have multiple events in a single year. And/or it might be interesting to see all of your select events overlayed in a single chart (and also solo).

[benhdye]

@rburghol
You're right it was a dry winter and event 128 carries over into 2002, which can be seen here:

which has an R squared of 0.940
and AGWR of 0.956

I looked into the summer, and there definitely should be at least on event there. It's hard to see exactly because I just changed the review app to have a 365 day buffer and had to zoom in a lot so I lost the axis markers, but to the right in this screenshot you can see a period that is essentially a month from August to September where the flows are sub 5 cfs, which should be a drought event. Our current idea is that because the flow is so low, the changes from for example 1.4 to 1.8 is outside the AGWRC change threshold.

This is interesting because it presents a scenario where by our definition this period is not a drought, but by DEQ history, this period is a known drought. So maybe this means that when flow gets so low, the change in AGWRC is not super important anymore as we see in this scenario where flow is less than 2 cfs?

we're currently discussing on ways to work around this, but were not sure we can add a way to manually select and analyze dates without major changes to the code behind the review app because it is based on finding events first, which makes it difficult to select specific dates that haven't been flagged as being in an event.

Also Will and Ilona also didn't get any events in the summer of 2002, only early January/February.

[rburghol]

@benhdye @willprokopik @ilonah22 thansk for the review on this. I am going to address the manual data ask here, but bravo on the winter 2001-2002 analysis -- it is interesting and thought provoking.

To be clear, I am not asking for you to amend the app at this point, simply to leverage the functions that underly the app, by passing those functions a time series flow corresponding to your manually identified event.

This was the intent of this item in our task list "Accepts event data frame as argument, returns updated copy with new columns containing analysis metrics" -- which from my cursory review of your code linked at the top is done quite well. Ideally, our goal is such that the functions should be agnostic to the source of the input time series, provided that the timeseries that they are given is appropriately formatted. Again, I think for the most part what you all have given does that.
Here are my suggestions to making it more robust:

• summarize_event - this should be tweaked to simply accept the event data as it's input, making the passing of the entire dataframe and the event ID superfluous. Basically, this practice requires that the function find the data that it is looking for, and I think that a correctly abstracted function should simply handle the data that it is given, not search, then handle. In short:
  • Lines 30-44 should be excerpted into a separate function

    baseflow_storage/summarize_event.R

    Line 5 in 0ddb853

    summarize_event <- function(analysis_data,

  • Note: this separate function should be called before summarize_event() is called, and then summarize_event() should receive the output of that function.
• I am not 100% sure but perhaps the function analyze_recession() could be too reliant on some internal data structures(I am unclear if it is grabbing actual dataframe columns or sub-dataframes)?

[ilonah22]

@rburghol, we've been looking back through the functions and the scripts to try and figure out the best way to go about this and think we have found a solution, if I am understanding what you mean correctly.

We think that the simplest way to handle a manually determined recession timeseries is to make a small change within the event_identification script. We included another arguement which is a logical option for a manual timeseries input. If the argument is set to TRUE, the script will skip the flag_stable_baseflow step, which is the main filtering script. Then, the script will automatically assign he input data a GroupID of 1 and set RecessionDay to TRUE. This way, the rest of the script can be run on that manually input event and get the calculated AGWRC and R-squared stats added at the end.

Lines 30-44 should be excerpted into a separate function

    for(i in (1:max(data$GroupID))){
      
      # Set sqldf query for valid days since fn$ wasnt working
      sqldf_query <- paste0(
        "select * from data 
    where GroupID = ", i, " and
    AGWR < 1 and
    delta_AGWR < ", dAGWRmax," and
    delta_AGWR > ", dAGWRmin,"
    ")
      
      event_data <- sqldf(sqldf_query)
      
      # Create lm of event selected dates
      logFlow_lm <-lm(log(event_data$Flow) ~ event_data$Date)
      event_sum <- summary(logFlow_lm)
      
      # Assign AWGR and R-squared
      AGWR <- exp(event_sum$coefficients[[2,1]])
      R_squared <- event_sum$r.squared
      
      new_row <- data.frame(i, AGWR, R_squared)
      
      event_df <- rbind(event_df, new_row)
      
    }

Those lines are there to ensure that the lm() and summary() are only being done to one event at a time, rather than the entire input dataset (for cases where the input data contains data from multiple events). If we think it is better to remove those AGWR and d_AGWR filters, we can, but it won't change anything with the events, it will just impact how the $AGWR$ and $R^2$ are calculated.

Maybe I'm not understanding exactly what you mean by finding the data its looking for?

Note: Lines 30-44 are only run is there is no input event, making a dataframe with the AGWRs and R-squareds for all events. If the user inputs a specific event, only that event will be analyzed after line 62.

[rburghol]

Hey @ilonah22 thanks for the quick turnaround -- I see what you are saying about the event ID, so in that way, you have already built this capability in.

What you suggest looks to me like it will work, but long term these functions need to improve in terms of modularity. The actual analysis should be a separate function that receives a single event, then returns the analysis for that event, rather than being inside of a loop inside of the function. So, in that way, you can call it for any time series. This is something that will be needed later on.

[rburghol]

@benhdye

This is interesting because it presents a scenario where by our definition this period is not a drought, but by DEQ history, this period is a known drought. So maybe this means that when flow gets so low, the change in AGWRC is not super important anymore as we see in this scenario where flow is less than 2 cfs?

Great freakin question. I am inclined to say no, based on these reasons:

• Given what we see in the chart you posted, copied below (did you just update to include this?). The last 4 days where the AGWRC is steady may be a candidate for the final AGWRC under low storage conditions.
• After the last 4 days, I'd hypothesize there are some small thunderstorms influencing flows. (We can actually check this out, and may do so later in the school year to make our reports more grounded in the available data).
• Why this didn't get flagged as an event might have more to do with the accuracy of the gage than anything else.

Can you tell us what the actual AGWRC values are during that period?

[willprokopik]

Mount Jackson Scatterplots

Some observations:
All Events Plot
Event 99 stands out with high AGWR above 1.0 with a moderate mean flow (around 200 cfs). Definite outlier (in my opinion). Events 94, 152, 157 cluster toward lower flow ranges (somewhere between 50-150 cfs) but near the upper AGWR group (right around 0.97). Event 156 has the lowest AGWR among study events at about 0.92 at higher flow (430 cfs). Not necessarily a true outlier. From this, we might say that larger flows tend to have slightly lower AGWR, but the effect is minimal. Outliers like event 99 deviate from the trend.
Seasonal Plots
Spring: Similar negative slope to the all-events plot. The most study events occur in this season (94, 151, 156, 157). Event 156 again appears as a low AGWR/high-flow point.
Summer: Very few events overall and trend is negative but based on a small sample. Event 152 sits at low flow (about 50 cfs) and relatively high AGWR (around 0.97).
Fall: There were no case study events marked. The pattern is still a gradual decrease in AGWR with increasing flow.
Winter: Event 99 continues to stand out with the highest AGWR of any case study event. The overall slope appears close to flat so seasonally, winter AGWR seems less sensitive to flow magnitude.