Two examples of using tidyhydat

Sam Albers

2019-03-27

Introduction

The real power of tidyhydat comes with its ability to efficiently import hydrometric data into R which then can be used in conjunction with R’s vast array of packages; tidyhydat gets hydrometric data into R and allows the user to start conducting data analysis as quickly as possible. This vignette is designed to take a user through two short examples of importing data using tidyhydat for use in hydrological visualization and analysis and will therefore use additional tools beyond simply the functions in tidyhydat.

Familiarity with the tidyverse

tidyhydat is designed to exist as a “tidy tool” that is complementary to the tidyverse suite of packages. This vignette assumes some familiarity with basic tidy tools particularly dplyr, tidyr, ggplot2 and lubridate. The R for Data Science book by Garrett Grolemund and Hadley Wickham is an enormously useful book both for general R use and for tidyhydat. The following is a list of useful links to begin learning particular aspect of each tidyverse package:

Example 1: Basic Data Extraction and Plotting

In this section, we will be using dplyr, a data manipulation package, ggplot2, an extremely simple yet powerful data visualization tool, lubridate a package that aids significantly in working with date and times in R and of course tidyhydat. Take a deep breathe. Though this might seem like an overwhelming onslaught of information on packages, each of these packages will save you considerable time with relatively minimal learning investment.

library(tidyhydat)
library(dplyr)
library(ggplot2)
library(lubridate)

Objective

Use tidyhydat to find the longest flow record in the Canadian hydrometric network then illustrate the use and ease of ggplot2 to create some visualizations of the discharge data.

Your first step is to download the HYDAT database which tidyhydat facilitates for you:

download_hydat()

You can see where the database was downloaded by running hy_dir(). This should be the only instance where you will need to interact directly with HYDAT. Each tidyhydat function prefixed with hy will automatically know where to look for HYDAT saving you the trouble. When using tidyhydat, often your first task is to find the station(s) that you are interested in. Because we are interested in the longest record we can extract that information with hy_stn_data_range() and then feed that information to hy_daily_flows() like this:

longest_record_data <- hy_stn_data_range() %>%
  filter(DATA_TYPE == "Q", RECORD_LENGTH == max(RECORD_LENGTH)) %>%
  pull_station_number() %>%
  hy_daily_flows()

Let’s break this down line by line to understand how tidyhydat uses tidy tools. First we are interested in getting data on record length:

hy_stn_data_range()
##   Queried from version of HYDAT released on 2019-01-17
##    Observations:                      11,923
##    Station(s) returned:               7,831
##    Stations requested but not returned: 
##     All stations returned.
## # A tibble: 11,923 x 6
##    STATION_NUMBER DATA_TYPE SED_DATA_TYPE Year_from Year_to RECORD_LENGTH
##    <chr>          <chr>     <chr>             <int>   <int>         <int>
##  1 01AA002        Q         <NA>               1967    1977            11
##  2 01AD001        Q         <NA>               1918    1997            80
##  3 01AD002        Q         <NA>               1926    2014            89
##  4 01AD003        H         <NA>               2011    2017             7
##  5 01AD003        Q         <NA>               1951    2017            67
##  6 01AD004        H         <NA>               1980    2016            32
##  7 01AD004        Q         <NA>               1968    1979            12
##  8 01AD005        H         <NA>               1966    1974             9
##  9 01AD008        H         <NA>               1972    1974             3
## 10 01AD009        H         <NA>               1973    1982            10
## # ... with 11,913 more rows

Our objective here is to filter from this data for the station that has the longest record of flow (DATA_TYPE == "Q"). You’ll also notice this symbol %>% which in R is called a pipe. In code, read it as the word then. So for the data_range data we want to grab the data then filter it by flow (“Q”) in DATA_TYPE and then by the maximum value of RECORD_LENGTH:

hy_stn_data_range() %>%
  filter(DATA_TYPE == "Q", RECORD_LENGTH == max(RECORD_LENGTH))
##   Queried from version of HYDAT released on 2019-01-17
##    Observations:                      1
##    Station(s) returned:               1
##    Stations requested but not returned: 
##     All stations returned.
## # A tibble: 1 x 6
##   STATION_NUMBER DATA_TYPE SED_DATA_TYPE Year_from Year_to RECORD_LENGTH
##   <chr>          <chr>     <chr>             <int>   <int>         <int>
## 1 02HA003        Q         <NA>               1860    2017           158

then pull the STATION_NUMBER that has the longest record:

hy_stn_data_range() %>%
  filter(DATA_TYPE == "Q", RECORD_LENGTH == max(RECORD_LENGTH)) %>%
  pull_station_number()
## [1] "02HA003"

then feed that number to hy_daily_flows()

longest_record_data <- hy_stn_data_range() %>%
  filter(DATA_TYPE == "Q", RECORD_LENGTH == max(RECORD_LENGTH)) %>%
  pull_station_number() %>%
  hy_daily_flows()

The result of this collection of simple functions is that we’ve extracted data the entire daily flow dataset (from 1860-2016) from station 02HA003.If we wanted more information on this station we could query other tables in HYDAT for more information. The hy_stations() function is very useful and outputs considerable metadata on a given station (converted here to a list for viewing purposes):

hy_stations(station_number = unique(longest_record_data$STATION_NUMBER)) %>%
  as.list()
## $STATION_NUMBER
## [1] "02HA003"
## 
## $STATION_NAME
## [1] "NIAGARA RIVER AT QUEENSTON"
## 
## $PROV_TERR_STATE_LOC
## [1] "ON"
## 
## $REGIONAL_OFFICE_ID
## [1] 5
## 
## $HYD_STATUS
## [1] "ACTIVE"
## 
## $SED_STATUS
## [1] "NA"
## 
## $LATITUDE
## [1] 43.15694
## 
## $LONGITUDE
## [1] -79.04722
## 
## $DRAINAGE_AREA_GROSS
## [1] 686000
## 
## $DRAINAGE_AREA_EFFECT
## [1] NA
## 
## $RHBN
## [1] FALSE
## 
## $REAL_TIME
## [1] FALSE
## 
## $CONTRIBUTOR_ID
## [1] 647
## 
## $OPERATOR_ID
## [1] 337
## 
## $DATUM_ID
## [1] 10
## 
## attr(,"missed_stns")
## character(0)

We now know that this station is actually NIAGARA RIVER AT QUEENSTON in Ontario and has an upstream drainage basin area of 686,000 square kilometres. As a first step toward visualization, let’s simply plot the time series for the entire record with a smoother added:

longest_record_data %>%
  ggplot(aes(x = Date, y = Value)) +
  geom_line() +
  geom_point() +
  geom_smooth() +
  labs(y = "Discharge (m)") +
  theme_minimal()

You can see very clearly where continuous monitoring was established. However, this type of plot obscures much of the data that we are interested in. We could plot all the years on the same axis and separate year by line colour. That requires a little bit of manipulation ahead of plotting using both dplyr and lubridate. Remember you can run a pipe adding one line at a time and comparing the outputs to get a sense of what each step is doing:

longest_record_data %>%
  mutate(dayofyear = yday(Date), Year = year(Date)) %>%
  mutate(dayofyear_formatted = as.Date(dayofyear - 1, origin = "2016-01-01")) %>% ## leap year as placeholder
  ggplot(aes(x = dayofyear_formatted, y = Value, colour = Year)) +
  geom_line() +
  scale_x_date(date_labels = "%b %d") +
  labs(y = "Discharge (m)") +
  theme_minimal()

This still is not a very useful plot mostly because our colour range still doesn’t resolve year very well and any intra-annual is not visible with this plot. Consider, for example, if rather than a line plot we use a tile plot and modified our colour scale to include more colours:

longest_record_data %>%
  mutate(dayofyear = yday(Date), Year = year(Date)) %>%
  mutate(dayofyear_formatted = as.Date(dayofyear - 1, origin = "2016-01-01")) %>% 
  ggplot(aes(x = dayofyear_formatted, y = Year, fill = Value)) +
  geom_tile() +
  scale_x_date(date_labels = "%b") +
  scale_y_reverse(expand = c(0, 0)) +
  scale_fill_gradientn(name = "Discharge (m^3/s) ", colours = rainbow(10)) +
  labs(y = "Year", x = "Date") +
  theme_minimal() +
  theme(legend.position="bottom")

This type of plot provides a clear indicator of discharge patterns over time. Our analysis ends here but one can imagine drawing in climate indices and variation to help explain periods of extreme discharge.

Example 2: Distribution of Historical Data

Objective

Evaluate current conditions in the station from Nunavut with longest record that is also in the Reference Hydrometric Basin Network relative to historical values.

For our second example, we will illustrate how to calculate a common metric used in hydrology: percentiles. For this approach we will use another piped series of functions to zero in on the station that we want. The packages we need for this example are loaded first:

library(tidyhydat)
library(dplyr)
library(ggplot2)
library(lubridate)

The pipe below finds the active station that has realtime data in the Reference Hydrometric Basin Network (RHBN) in the territory of Nunavut that has the longest record then grabs all the daily flow information:

nunavut_stn_flows <- hy_stations() %>%
  filter(HYD_STATUS == "ACTIVE") %>%
  filter(REAL_TIME == TRUE) %>%
  filter(RHBN == TRUE) %>%
  filter(PROV_TERR_STATE_LOC == "NU") %>%
  pull_station_number() %>%
  hy_stn_data_range() %>%
  filter(DATA_TYPE == "Q") %>% 
  filter(RECORD_LENGTH == max(RECORD_LENGTH)) %>%
  pull_station_number() %>%
  hy_daily_flows()

Armed with our data, we can now evaluate the historical distribution of data then to put some real time data into historical context. First we need to calculate where an individual observation is distributed against all other observations on that day for every day. The prctile column tells us what percentage of values on that same day over the entire data record fall above and below the observations on that row. We are restricting our analysis to the last thirty days from when this vignette was last compiled. The ecdf() function creates an equation to calculate percentiles based on the Value column (i.e discharge) then takes each individual observation of Value and calculates the percentile.

pct_flow <- nunavut_stn_flows %>%
  mutate(dayofyear = yday(Date), Year = year(Date)) %>%
  filter(dayofyear %in% yday(seq.Date(from = (Sys.Date()-30), 
                                      to = Sys.Date(), by = "day"))) %>%
  group_by(dayofyear) %>%
  mutate(prctile = ecdf(Value)(Value)) %>%
  mutate(Date_no_year = dmy(paste0(day(Date),"-",month(Date),"-",year(Sys.Date())))) %>%
  ungroup()

To collect real time data, we can use the realtime_dd() function in tidyhydat. Because real time data is collected in hourly (or higher) intervals and we are operating on a daily basis, we first need to take a daily mean of discharge. Again this is accomplished in a pipe:

nunavut_realtime <- realtime_dd(unique(nunavut_stn_flows$STATION_NUMBER)) %>%
  mutate(Date_day = as.Date(Date)) %>%
  group_by(Date_day) %>%
  summarise(Value = mean(Value, na.rm = TRUE)) %>%
  ungroup()

Finally we can plot all of this on to one figure. This type of plot is useful to combine historical data sources with real time data and provide a visual assessment of how typical flows in a given system are.

CAUTION: All real time data are presented AS IS and represents unapproved data.

ggplot(pct_flow, aes(x = Date_no_year, y = Value)) +
  geom_point(aes(colour = prctile)) +
  geom_line(data = nunavut_realtime, aes(x = Date_day), colour = "black") +
  geom_point(data = nunavut_realtime, aes(x = Date_day, shape = factor(year(Date_day))), colour = "black") +
  scale_colour_gradientn(name = "Discharge Percentile", colours = rainbow(10)) +
  scale_shape_discrete(name = "Year") +
  theme_minimal() +
  labs(title = "Historical flow relative to current year",
       subtitle = "Current year flows are displayed in black",
       caption = "Real time data is presents AS IS and represents unapproved data",
       x = "Date", y = "Discharge (m^3/s)")