09_summaries.qmd

---
title: "Summaries"
---

In this script we will finalize the data sets by adding metadata 
and summarizing as appropriate. Each data set will be saved to CSV file for future
analysis 
(see [Appendix - Data](XX_data.html) for descriptions of the data values). 

## Setup
```{r}
#| message: false
#| cache: false
source("XX_setup.R")
runs <- load_runs()
meta <- runs |>
  select(tagDeployID, speciesID) |>
  distinct() |>
  left_join(tbl(dbs[[1]], "species") |> select(english, id) |> collect(), 
            by = c("speciesID" = "id"))

bouts <- read_rds("Data/02_Datasets/bouts_cleaned.rds") |>
  left_join(meta, by = "tagDeployID")
trans <- read_rds("Data/02_Datasets/transitions_cleaned.rds") |>
  left_join(meta, by = "tagDeployID")
```


## Adding metadata

Add appropriate meta data for summaries as well as for creating final datasets.
Ensure datasets now include all the meta data and have been converted to flat
files (no list columns, no specially column formats, such as durations or difftimes), 
so can now be saved to csv.

> **Datasets created**:  
> 
> - [`Data/03_Final/bouts_final.csv`](XX_data.html#finalbouts_final.csv)
> - [`Data/03_Final/transitions_final.csv`](XX_data.html#finaltransitions_final.csv)

```{r}
trans <- trans |>
  # Omit weird metrics
  select(-"migration", -"connected") |> 
  # Flatten units
  mutate(min_time_hrs = flatten_units(min_time, "hours"),
         time_diff_hrs = flatten_units(time_diff, "hours"),
         next_dist_km = flatten_units(next_dist, "km"),
         lag1_min = flatten_units(lag1, "min"),
         lag2_min = flatten_units(lag2, "min"),
         speed_m_s = flatten_units(speed, "m/s"))|>
  select(-min_time, -time_diff, -lag1, -lag2, -speed)

tz <- bouts |>
  select(dateBegin, recvDeployLat, recvDeployLon) |>
  distinct() |>
  mutate(tz = tz_lookup_coords(recvDeployLat, recvDeployLon, warn = FALSE),
         offset = map2(dateBegin, tz, tz_offset)) |>
  unnest(offset) |>
  select("dateBegin", "recvDeployLat", "recvDeployLon", "utc_offset_h")

bouts <- left_join(bouts, tz, by = c("dateBegin", "recvDeployLat", "recvDeployLon")) |>
  # Add local times
  mutate(timeBeginLocal = timeBegin + hours(utc_offset_h),
         timeEndLocal = timeEnd + hours(utc_offset_h)) |>
  select(-"runID", -"len", -"ant") |>  # Omit list columns
  # Flatten units
  mutate(total_time = flatten_units(total_time, "min")) |>
  rename(total_time_min = total_time)

write_csv(bouts, "Data/03_Final/bouts_final.csv")
write_csv(trans, "Data/03_Final/transitions_final.csv")
```


## Daily summaries

Summarize how much time each individual spent at a particular station on a given
day. 

Here we will work in local times to determine when a day rolls over at midnight.
We will then split each bout that crosses midnight into multiple bouts, each 
starting and stopping at the day's limits (midnight).

Then we'll summarize these data into amount of time spent per day at a spectific
station.


> **Datasets created**:  
> 
> - [`Data/03_Final/bouts_final_split.csv`](XX_data.html#finalbouts_final_split.csv)
> - [`Data/03_Final/summary_daily_bouts.csv`](XX_data.html#finalsummary_daily_bouts.csv)

```{r}
plan(multisession, workers = 6) # Setup parallel

bouts_split <- bouts |>
  mutate(
    # Split bouts by days
    date_local = future_map2(timeBeginLocal, timeEndLocal, \(x, y) {
      if(as_date(x) != as_date(y)) {
        d <- seq(as_date(x), as_date(y), by = "1 day") 
        d <- d[d >= x & d <= y]
      } else d <- c()
      unique(c(x, d, y))
    }, .progress = interactive()),
    t1 = map(date_local, \(x) x[-length(x)]),
    t2 = map(date_local, \(x) x[-1]),
  ) |>
  unnest(cols = c(t1, t2)) |>
  select(-dateBegin, -timeBegin, -timeEnd, -timeBeginLocal, -timeEndLocal, -date_local) |>
  rename(timeBeginLocal = t1, timeEndLocal = t2) |>
  mutate(dateBeginLocal = as_date(timeBeginLocal))

daily_bouts <- bouts_split |>
  summarize(time_hrs = sum(difftime(timeEndLocal, timeBeginLocal, units = "hours")),
            time_hrs = as.numeric(time_hrs), 
            .by = c("english", "tagDeployID", "stn_group", "dateBeginLocal")) |>
  # Add in lat/lon (corresponds to one of the stations, not all in a group, but close enough)
  left_join(select(bouts, "stn_group", "recvDeployLat", "recvDeployLon") |> distinct(),
            by = "stn_group")

write_csv(daily_bouts, "Data/03_Final/bouts_final_split.csv")
write_csv(daily_bouts, "Data/03_Final/summary_daily_bouts.csv")
```


## Individual Summaries

Summarize the individual and overall patterns and number of samples in the data.

We'll categorize birds as those who we know 

- Did not travel
- `travelled` (moved at least 100 km)
- `migrated` (moved at least 1 latitude ~111km south/north)
- `migrated_far` (moved at least 5 latitudes ~555km south/north)
- Overall (across all birds)

This doesn't mean that birds *didn't* move farther, only that we have no evidence that they did.

> **Datasets created**:  
> 
> - [`Data/03_Final/summary_birds.csv`](XX_data.html#finalsummary_birds.csv)


```{r}
trans_dist <- trans |>
  summarize(total_dist = sum(next_dist), 
            travelled = total_dist > set_units(100, "km"),
            migrated = (max(lat1) - min(lat2)) > 1,
            migrated_far = (max(lat1) - min(lat2)) > 5,
            .by = c("english", "tagDeployID"))

sum_time <- daily_bouts |> 
  left_join(trans_dist, by = c("english", "tagDeployID")) |>
  mutate(total_dist = replace_na(total_dist, set_units(0, "km")),
         travelled = replace_na(travelled, FALSE),
         migrated = replace_na(migrated, FALSE),
         migrated_far = replace_na(migrated_far, FALSE)) |>
  arrange(tagDeployID, dateBeginLocal) |>
  summarize(n_stn = n_distinct(stn_group), 
            min_date = min(dateBeginLocal),
            mean_date = mean(dateBeginLocal),
            max_date = max(dateBeginLocal),
            total_time_hrs = sum(time_hrs),
            mean_time_hrs = mean(time_hrs),
            res_stn = stn_group[1],
            mean_time_no_resident_hrs = mean(time_hrs[stn_group != stn_group[1]]),
            first_time_hrs = sum(time_hrs[stn_group == stn_group[1]]),
            last_time_hrs = sum(time_hrs[stn_group == stn_group[n()]]),
            .by = c("english", "tagDeployID", "travelled", "migrated", "migrated_far")
  ) |>
  mutate(mean_time_no_resident_hrs = replace_na(mean_time_no_resident_hrs, 0)) |>
  mutate(across(where(is.difftime), as.numeric))

write_csv(sum_time, "Data/03_Final/summary_birds.csv")
```

Now we'll create an overall summary of these individual-level summaries. 
This is a summary table looking at the sample sizes (No. XXX) as well as 
averages of individual means (Avg of mean etc.). 

```{r}
#| code-fold: true
sum_time |>
  bind_rows(mutate(sum_time, travelled = FALSE, migrated = FALSE, 
                   migrated_far = FALSE, overall = TRUE)) |>
  summarize(n = n(),
            n_species = n_distinct(english),
            across(-c("tagDeployID", "res_stn", "english", "min_date", "max_date"), mean), 
            min_date = min(min_date), max_date = max(max_date),
            .by = c("travelled", "migrated", "migrated_far", "overall")) |>
  mutate(type = case_when(overall ~ "Overall", 
                          migrated_far ~ "Migrated Far (>5 Latitudes)", 
                          migrated ~ "Migrated (>1 Latitude)",
                          travelled ~ "Travelled (>100km)",
                          TRUE ~ "No large movements")) |>
  select(-travelled, -migrated, -migrated_far, -overall) |>
  select(`No. Individuals` = "n", 
         `No. Species` = "n_species",
         `Avg. No. Stations visited` = "n_stn",
         `Min date` = "min_date",
         `Avg date` = "mean_date",
         `Max date` = "max_date",
         `Avg total time detected` = "total_time_hrs",
         `Avg mean time detected` = "mean_time_hrs",
         `Avg mean time detected (not at resident station)` = "mean_time_no_resident_hrs",
         `Avg total time detected by resident station` = "first_time_hrs",
         `Avg total time detected by final station` = "last_time_hrs",
         type) |>
  mutate(across(where(is.numeric), \(x) round(x, 2))) |>
  pivot_longer(-type, names_to = "Measure", values_transform = as.character) |>
  pivot_wider(names_from = "type") |>
  select("Measure", contains("No large"), contains("Trave"), contains(">1"), contains(">5"), "Overall") |>
  gt() |>
  gt_theme() |>
  tab_header(title = "Overall summary")
```


## Looking for stopovers

**Are there any individuals that actually hang out around the receiver?**

We'll look for two different indicators of a stopover

1. Actually being detected around a receiver for longer than half an hour
    - In the above summaries, we see that the average amount of time detected at
      a receiver which is not the first (or home) receiver, is about 9min (0.15 hours)
  
2. Being detected around a single receiver (or receiver group) on more than one
   day each detection within a week of another (and no other stations detected 
   in the meanwhile).

So looking only at October through April (months 10-4), we'll define some 
stopover metrics. 

> **Datasets created**:  
> 
> - [`Data/03_Final/summary_stopover_time.csv`](XX_data.html#finalsummary_stopover_time.csv)
> - [`Data/03_Final/summary_stopover_days.csv`](XX_data.html#finalsummary_stopover_days.csv)



```{r}
stopovers <- daily_bouts |>
  arrange(tagDeployID, dateBeginLocal) |>
  filter(stn_group != stn_group[1], 
         month(dateBeginLocal) %in% c(10, 11, 12, 1, 2, 3, 4), .by = "tagDeployID") |>
  mutate(
    # Detected at the same station across days?
    same_stn = stn_group == lead(stn_group, default = "x"),
    same_stn = same_stn | lag(same_stn, default = FALSE),
    # Detected within 7 days?
    close_time = dateBeginLocal >= (lead(dateBeginLocal, default = ymd("2999-01-01")) - days(7)), 
    close_time = close_time | lag(close_time, default = FALSE),
    # If both, call this a multi-day stopover
    stopover = same_stn & close_time,
    .by = "tagDeployID")
```


### Over 30 min at a receiver
Here we see different individuals spending time around receivers during
migration.
```{r}
stopover_time <- stopovers |>
  group_by(english, tagDeployID) |>
  select(-same_stn, -close_time, -stopover) |>
  arrange(english, tagDeployID) |>
  filter(time_hrs > 0.5)
write_csv(stopover_time, "Data/03_Final/summary_stopover_time.csv")

stopover_time |>
  select(-"recvDeployLat", -"recvDeployLon") |>
  gt() |>
  gt_theme() |>
  fmt_number(columns = time_hrs) |>
  cols_label_with(everything(), tools::toTitleCase) |>
  cols_label(time_hrs ~ "Time ({{Hours}})") |>
  tab_options(container.height = px(600),
              container.overflow.y = "auto") |>
  tab_caption("Stopovers >30min (see Time (Hours) column)")
```

**Examples**

```{r}
#| fig-asp: 2
#| out-width: 75%
plot_map(trans, 43250) + 
  labs(title = "Song Sparrow",
       caption = "Spent 1 hour at 2812-7446 on October 15th and 30min at 7338 on Oct 24th")

plot_map(trans, 51150) + 
  labs(title = "Hermit Thrush",
       caption = "Spent 10 hours at 8858 on November 6th")
```

### Several days at a receiver
What about individuals which are spotted several days in a row at a station, 
even if not for a long period of time?

```{r}
stopover_days <- stopovers |>
  group_by(english, tagDeployID) |>
  filter(stopover) |>
  select(-same_stn, -close_time, -stopover)
write_csv(stopover_days, "Data/03_Final/summary_stopover_days.csv")

stopover_days |>
  select(-"recvDeployLat", -"recvDeployLon") |>
  gt() |>
  gt_theme() |>
  fmt_number(columns = time_hrs) |>
  cols_label_with(everything(), tools::toTitleCase) |>
  cols_label(time_hrs ~ "Time ({{Hours}})") |>
  tab_options(container.height = px(600),
              container.overflow.y = "auto") |>
  tab_caption("Stopovers where detected over several days at a receiver")
```

**Examples**
```{r}
#| fig-asp: 2
#| out-width: 75%
plot_map(trans, 18901) + 
  labs(title = "Swainson's Thrush",
       caption = "Spent Oct 8 -  Oct 10 at station group 3495-4725-5240")

plot_map(trans, 34474) + 
  labs(title = "White-throated Sparrow",
       caption = "Spent Oct 19 - Oct 27th at station group 5398-7354")
```


## Looking at circadian patterns

### Directed exploration 

This is a quick look at circadian patterns of movement, especially for in BC birds in the fall.

We first do a quick look to pull out some ids of relevant birds to explore.
Then we create `start` and `end` columns to hold the hour of day values (i.e. 10.5
which would be 10:30 am, local time). 
```{r}
# To find birds in BC which have many records
b <- filter(bouts_split, recvDeployLon < -110, recvDeployLat > 48.97) |>
  summarize(n = n(), .by = c("tagDeployID", "english")) |> 
  arrange(desc(n))

# To convert start times to start/end fractional hours (i.e. time of day rather than date/time)
b0 <- bouts_split |>
  mutate(start = hour(timeBeginLocal) + minute(timeBeginLocal)/60,
         end = hour(timeEndLocal) + minute(timeEndLocal)/60)
```

Next we'll explore the activity (detections by a station over the course of the day).

In these figures we're looking at date along the Y axis and the time of day of bouts
detected by a receiver on each day *in local* time along the X axis. 

In both cases there is only one station that the birds are detected at (within 
the range of dates we're exploring).

It suggests that there are times when these birds spend mornings near the receiver, but
spend their nights or afternoons farther away before returning again for the evening (see the 
large patches of empty time between 10am and 8pm, especially in August and early
September in the second figure. You can even see the funnel shape of the changing
daylengths in the activity of these two birds.

```{r}
#| code-fold: true
#| fig-height: 10
#| fig-asp: 1
filter(b0, tagDeployID == 41224, dateBeginLocal > "2022-09-01", dateBeginLocal < "2023-01-01") |>
  ggplot(aes(x = start, xend = end, y = dateBeginLocal, colour = stn_group)) +
  geom_segment(linewidth = 1.5) +
  labs(x = "Hour (Local timezone)", title = "tagDeployID 41224 (Spotted Towhee)")

filter(b0, tagDeployID == 41209, dateBeginLocal < "2022-12-01", dateBeginLocal > "2022-07-30") |>
  ggplot(aes(x = start, xend = end, y = dateBeginLocal, colour = stn_group)) +
  geom_segment(linewidth = 1.5) +
  labs(x = "Hour (Local timezone)", title = "tagDeployID 41224 (Spotted Towhee)")
```

### Create data set for exploration

> **Datasets created**:  
> 
> - [`Data/03_Final/summary_circadian_bouts.csv`](XX_data.html#finalsummary_circadian_bouts.csv)

The above exploration revealed some interesting patterns. 
While we don't have time in this round of the project to get into them in 
detail, we can create a data set for exploration.

We'll filter the bout-level data to include only periods of time for individuals when 
they spent more than 30 days at a specific station (without being detected at
other stations in between). This may omit some which are hanging out in a 
local area and are being recorded by multiple stations, but at least gives
us a starting point. 

:::{.callout-tip}
#### Note 
We filter by ID and date, not by station, so figures will show birds at multiple
stations if they

```{r}
include <- daily_bouts |>
  arrange(tagDeployID, dateBeginLocal) |>
  mutate(same_stn1 = lead(stn_group, default = "X") == stn_group & 
           difftime(lead(dateBeginLocal, default = ymd("2099-01-01")), dateBeginLocal, units = "days") < 3,
         same_stn2 = lag(stn_group, default = "X") == stn_group &
           difftime(dateBeginLocal, lag(dateBeginLocal, default = ymd("1900-01-01")), units = "days") < 3,
         .by = "tagDeployID") |>
  filter(!(!same_stn1 & !same_stn2)) |>
  mutate(start = same_stn1 & !same_stn2,
         end = !same_stn1 & same_stn2) |>
  mutate(group = cumsum(start), .by = "tagDeployID") |>
  mutate(n = n(),  min = min(dateBeginLocal), max = max(dateBeginLocal), 
         .by = c("tagDeployID", "group")) |>
  filter(n > 20, difftime(max, min, units = "days") > 30) |>
  select(tagDeployID, stn_group, group, n, min, max) |>
  distinct()

circadian <- bouts_split |>
  mutate(year = year(dateBeginLocal)) |>
  semi_join(include, by = join_by("tagDeployID", between(dateBeginLocal, min, max))) |>
  mutate(start = hour(timeBeginLocal) + minute(timeBeginLocal)/60,
         end = hour(timeEndLocal) + minute(timeEndLocal)/60,
         end = if_else(end == 0, 24, end), # Account for days where the bird was detected over midnight
         stn_nice = paste0(
           if_else(recvDeployLon < -94.96, "West", "East"), 
           ": Station (group) ", stn_group,
           " (", round(recvDeployLat, 4), ", ", round(recvDeployLon, 4), ")")) |>
  arrange(speciesID, recvDeployLon)

write_csv(select(circadian, -"stn_nice"), "Data/03_Final/summary_circadian_bouts.csv")
```

### Broad exploration

Now we'll create circadian plots for each of these individuals for the dates at which they
were consistently at a single station. 

```{r}
#| fig-height: 10
#| fig-asp: 1
#| fig-dpi: 75
#| message: false
#| code-fold: true
#| results: asis
walk(unique(circadian$tagDeployID), \(x) {
  cat("\n\n####", x, "\n\n")
  g <- ggplot(filter(circadian, tagDeployID == x), 
         aes(x = start, xend = end, y = dateBeginLocal, colour = stn_nice)) +
    theme(legend.position = "bottom", legend.title = element_blank()) +
  geom_segment(linewidth = 1.5) +
  facet_wrap(~ year, scales = "free") +
  labs(title = paste0(circadian$english[circadian$tagDeployID == x], ": ", x), 
       x = "Hour (Local timezone)")
  print(g)
})
```


## Future Ideas

- Assess how many local station along the path were **not** used (but were active). 
- Look at arrivals with respect to time of day to assess when birds arriving
  - Use timing as indicator of passing through or stopping over? (i.e early morning?)
  - Amie et al. use hit patterns and timing to determine if stopping in. 
- Explore timing of activity during migration stop-overs 
  - Are birds leaving specific areas at particular times, then returning?
  - *see Looking at circadian patterns* for a brief exploration


## Reproducibility
{{< include _reproducibility.qmd >}}