applications in hydrology

class: center, middle, inverse, title-slide

# <img src='https://www.tidyverse.org/rstudio-logo.svg' alt='RStudio' width='180' height='190' style='border:5px;margin:5px;clear:both;'><img src='https://www.freeiconspng.com/uploads/blue-plus-icon-12.png' width='60' height='70' align='top' style='border:5px;margin:25px;clear:both; position:relative;top:35px;'><img src='https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png' width='180' height='190' alt='R Shiny' style='border:5px;margin:5px;clear:both;'> applications in hydrology
### Josh Erickson
### USDA-Forest Service

---

## About Me

- Hydrologist (day job)
- Remote Sensing Analyst (detail)
- Geoscience and Math background
- Focus is leveraging computing platforms (R, Python, GEE) for resource management work.

<img src='rshiny_talk_files/images/me.png' height='300px' class='center'>
---

## Goals
--

- Leave having an idea of what <img src="https://www.tidyverse.org/rstudio-logo.svg" height="30px"> and <img src="https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png" height="30px"> are and what they can do.

- What/why, how.

- Most importantly, some applications in hydrology

---

## What is R + Shiny?
--

- R is a free open source software .blue[(FOSS)] programming language

- R was initially set up for exploring research data fast (S, S+); but has moved into reactive programming in the last decade {.blue[shiny]}

- R shiny <img src="https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png" height="30px"> is built on this reactive programming.

- It leverages the functional programming in R with reactive programming under the hood.

- This allows .blue[(users)] to explore complex problems fast and easy but also create automated/reproducible reports.

---

## Why R?

(disclaimer - I'm biased; other platforms are just as good!)

1. R has a lot of hydrology (USGS, acadamia) and statistics (extreme value, machine learning, etc) packages.

2. Well organized community (ask for help it shall be recieved).

3. It has {.blue[shiny]} and {.blue[rmarkdown]}, which are great for communicating projects.

4. Great visualization {.blue[ggplot]}

5. The {.blue[tidyverse]}, which are packages built on an analysis framework.

---

## Why is that important?

- Having packages built on this '_framework_' really helps with the **process**.

- Which ultimately makes **communicating** clearer and easier!

- Shiny is just one piece of this framework but IMO it can be go a long way in communicating and exploring/visualizing.

- Below is where .blue[Shiny] fits in to this workflow.

--
<img src='https://josiahparry.com/slides/tidymodels/images/tidyv-flow.png' height="300px" class='center'>
<center>
<figcaption>Fig from: https://josiahparry.com/slides/tidymodels/images/tidyv-flow.png</figcaption>
</center>

---

## Caveat! Warning! It's a steep climb!

To get to full-blown shiny apps!

<img src='https://www.business-science.io/img/curriculum/bsu_the_goal.jpg' height='400' class='center'>
<center>
<figcaption>Fig from: https://www.business-science.io/about_program.html</figcaption>
</center>
--

---
## Caveat! Warning! It's a steep climb!

- It is a programming language so you can't get around code...

---
## So...

If you're overoptimistic, impractical and want to see rapid progression over a short period of time. Then <img src="https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png" height="30px"> is for you!

---
## Else if your sane...

- There are some people willing to help you go down the rabbit hole!

.pull-left[<img src='https://media.giphy.com/media/ceK5e0DXc7Zh6/giphy.gif' height='300' class='left'>]

---

## very encouraging people...
--

<img src='https://media.giphy.com/media/bwLowbhUWm2lO/giphy.gif' height='400px' class='center'>
--

---

## that will help you along the way...

---

## Agenda for 'how & apply'

- Look at some non-reactive examples and then convert to reactive

- Then go over some examples in hydrology

- Then wrap up for questions/discussion

---
## Let's dive in

Shiny is a way to take a bunch of tasks that are really similar except for a few things (names, metrics, styles, etc) and react to those inputs.

- before we get started with shiny let's dive into the _non-reactive_ approach.

- again totally fine and a recommended approach.

- Let's explore some __peak__ flows at USGS gauging stations!

--
  
---
## non-reactive workflow

Use the {.blue[dataRetrieval]} from USGS to get __quality controlled__ peak flows and then plot using {.blue[ggplot2]}.

```r
library(tidyverse)
library(ggtext)
library(dataRetrieval)

# get peak flow data
peak_yaak <- dataRetrieval::readNWISpeak(12304500)

# graph data
ggplot(peak_yaak, aes(peak_dt, peak_va)) + 
 geom_point() + 
 geom_line() +
 labs(title = paste0(attr(peak_yaak, which = 'siteInfo')$station_nm), 
 subtitle = paste0("Drainage Area: ", attr(peak_yaak, which = 'siteInfo')$drain_area_va, ' sq.miles'),
 x = 'Date',
 y = "Q (ft3/sec)") +
 theme_bw() + theme(axis.title.y = element_markdown())
```

---

---

## But what about the Fisher River

and so on, and so on...

## what about Tobacco River? Or Thompson? etc, etc.

---

## Next logical step

Next logical step might be to make a function. This is a little better. But you still need to manually add the siteID.

```r
peak_plots <- function(siteID){
 
 
 peak <- dataRetrieval::readNWISpeak(siteID)

# graph data
ggplot(peak, aes(peak_dt, peak_va)) + 
 geom_point() + 
 geom_line() +
 labs(title = paste0("Peak Flow: ",attr(peak, which = 'siteInfo')$station_nm), 
 subtitle = paste0("Drainage Area: ", attr(peak, which = 'siteInfo')$drain_area_va, ' sq.miles'),
 x = 'Date',
 y = "Q (ft3/sec)") +
 theme_bw() + theme(axis.title.y = element_markdown())
 
}

peak_plots(12390700)
```

---

---

## What about <img src="https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png" height="100px">

Definitely some leg work but once you get it going it can be really nice.

Work through a simple example:

1. We want to get annual peak flow for USGS gauging stations.

2. We want them to be displayed on a map.

3. When we click the location a interactive graph appears.
   
--

---

## Shiny code (look similar?)

```r
library(shiny)

usgs_sites <- dataRetrieval::whatNWISdata(stateCd = 'MT', parameterCd='00060')
usgs_sites <- usgs_sites %>% filter(count_nu >= 365*10)

ui <- fluidPage(
 selectInput('station', 'Select Station Name', 
 choices = usgs_sites$station_nm), 
 plotOutput('peak_plot')
 )

server <- function(input, output, session) {
 output$peak_plot <- renderPlot({
 withProgress(message = 'loading data', value = 1/2,{
 site_no <- usgs_sites[usgs_sites$station_nm %in% input$station,]$site_no
 peak_plots(site_no)})})
}
shinyApp(ui, server)
```

---

---
## Step 2 & 3

What's nice with Shiny is we can add a map and click the locations from which a graph will pop-up. We can also make that graph interactive!

<video width="320" height="240" controls>
 <source src="rshiny_talk_files/images/plotly_peak.mp4" type="video/mp4">
</video>
--

---

## Why is this important?
--

- Again, this allows .blue[(users)] to explore complex problems fast and easy.

- We spend less time coding and more time looking at trends.

- We can share with others that might not have the time to code but would like to use.

<center>
.bg-washed-green.b--dark-green.ba.bw2.br3.shadow-5.ph4.mt5[Now we can start applying .blue['in the wild'] to help with resource management work.]
</center>
--

---
## In the wild?

- What else can we apply this to?

- Culvert sizing? Flood Frequency? Hydraulics?

## Yes, Yes, Yes Yes

---

## Culvert Sizing

- How could we apply this culvert sizing?

- Take a common approach on the Kootenai and apply to shiny.

Reasons I like this

1. It's reproducible.

2. Take away Excel, ArcGIS or Stream Stats and Microsoft Word.

3. It's efficient.

---

## Example (poc)

---
## Flood Frequency
 
- Could we get numerous univariate distributions on the fly using the peak function we created?

- Sounds like a good idea but how would that look?

- Let's take the peak function and then use those values to calculate numerous univariate distributions (GEV, Logpearson, Pearson, Lognormal, etc).

- Then let's visualize those distributions graphically (density, Q-Q, CDF, PP) but also run some goodness of fit tests (Kolmogroc-Smirnov, Anderson-Darling, AIC/BIC, etc)

- Then we can use this information (gof) to select a distribution to use...

---

## Example (poc)

---

## Hydraulics?

- How could we do this?

- We could make it easier to calculate some common metrics (shear stress and boundary shear stress)

- Let's take some inputs (shields D50, particle of interest D50, hydraulic radius, slope, D50) and use widgets to calculate the shear stress and boundary shear stress for us!

---

## Example (poc)

---
## Wrapping up

--
- You're probably sick of me saying this...

- This allows .blue[(users)] to explore complex problems fast and easy while also creating automated/reproducible reports.

- There's so so much more you can do!

- Have ideas? Let's create them!

- I hope that you can leave having an idea of what <img src="https://www.tidyverse.org/rstudio-logo.svg" height="30px"> and <img src="https://raw.githubusercontent.com/rstudio/shiny/master/man/figures/logo.png" height="30px"> are and get the creative juices going for projects on your forest!

---

# Questions?