How to Access an API with RStudio
Autumn Biggie 10/3/2021
In this document, we’ll walk through how to connect to an API, using an example. The API we’ll be connecting to is the OpenWeather API. Specifically, we’ll be looking at current and forecast weather data from the One Call API, one of the many APIs that openweathermap.org offers.
Preliminary Steps
-
Many APIs require you to access the data using a unique API key. To acquire your free API key, register here.
-
The following packages will be necessary in order to connect with the API:
httrjsonlite
-
The following packages will be necessary in order to do some analyses after we load the data:
tidyverseanytimeggplot2chron
Function to Access API
Here, I’ve written a function weather.api to be able to easily access
the target API. The arguments are as follows:
latitude(required): latitude of the geographic location desiredlongitude(required): longitude of the geographic location desiredapi.id(required): input your unique API key hereexclude(optional): list any parts of the weather data you want to exclude from the results. This should be a comma-delimited list, with or without spaces. Case is unimportant. The options are:currentminutelyhourlydaily
units(optional): The unit of measurement in which values are returned. Options are:standardimperialmetric
Note: All parameters should be in character format i.e. latitude =
"-46.05".
weather.api <- function(latitude, longitude, api.id, exclude = NULL, units = "metric") {
#assign the individual pieces of the required url to their respective objects
base <- "https://api.openweathermap.org/data/2.5/onecall"
lat1 <- "lat="
lat2 <- latitude
lon1 <- "lon="
lon2 <- longitude
exc1 <- "exclude="
exc2 <- tolower(sub(" ", "", exclude)) #remove spaces and convert to lowercase
apid1 <- "appid="
apid2 <- api.id
units1 <- "units="
units2 <- units
#paste pieces together
lat <- paste(lat1, lat2, sep = "")
lon <- paste(lon1, lon2, sep = "")
ifelse(!is.null(exc2), exc <- paste(exc1, exc2, sep = ""), exc <- "nullexclude")
apid <- paste(apid1, apid2, sep = "")
units <- paste(units1, units2, sep = "")
#paste base and latitude pieces together
first <- paste(base, lat, sep = "?")
#create the entire url by pasting all pieces together using delimiter "&", conditioning on if any information is excluded
ifelse(!is.null(exc2), my.url <- paste(first, lon, exc, apid, units, sep = "&"), my.url <- paste(first, lon, apid, units, sep = "&"))
#access the Weather API using the URL assembled above
weather.info <- GET(my.url)
#convert the content of the accessed data from raw to character, then present in readable data frame format
final <- weather.info$content %>% rawToChar() %>% fromJSON()
return(final)
}
Exploratory Data Analysis
Now that we can easily access the API, let’s do some exploratory data analysis.
I want to explore the weather data of five locations that lie roughly along the same line of longitude.
Locations:
- 44.19’N, 69.47’W (Augusta, Maine, US)
- 21.28’N, 71.08’W (Cockburn Town, Turks & Caicos Islands, UK)
- 7.54’N, 72.3’W (Cucata, Colombia)
- 51.38’S, 69.13’W (Rio Gallegos, Argentina)
- 55.05’S, 67.05’W (Puerto Williams, Chile)
Maine <- weather.api(latitude = "44.19", longitude = "-69.47", api.id = "input_your_key", exclude = "current, minutely, hourly")
Maine.day <- Maine$daily
Maine.day
## dt sunrise sunset moonrise moonset
## 1 1633622400 1633603373 1633644485 1633607760 1633647660
## 2 1633708800 1633689846 1633730778 1633698900 1633735860
## 3 1633795200 1633776319 1633817071 1633790160 1633824300
## 4 1633881600 1633862792 1633903365 1633881360 1633913280
## 5 1633968000 1633949266 1633989660 1633972140 1634002920
## 6 1634054400 1634035740 1634075955 1634062320 1634093100
## 7 1634140800 1634122215 1634162251 1634151780 1634183700
## 8 1634227200 1634208690 1634248548 1634240580 0
## moon_phase temp.day temp.min temp.max temp.night temp.eve
## 1 0.05 19.31 9.91 21.15 11.54 15.25
## 2 0.08 20.61 10.44 20.79 11.58 13.63
## 3 0.12 14.25 6.20 14.53 6.52 8.87
## 4 0.16 15.19 6.52 16.03 10.83 14.10
## 5 0.20 17.71 11.30 21.26 15.31 17.69
## 6 0.25 20.91 14.21 22.56 15.25 18.92
## 7 0.27 19.30 16.09 21.40 16.09 20.18
## 8 0.30 20.70 14.49 23.48 15.45 20.08
## temp.morn feels_like.day feels_like.night feels_like.eve
## 1 9.91 19.05 11.20 15.00
## 2 10.44 20.16 10.36 13.06
## 3 6.32 13.11 6.52 7.87
## 4 6.74 14.41 10.53 13.60
## 5 11.30 17.49 15.46 17.71
## 6 14.21 20.88 15.39 18.93
## 7 16.83 19.77 16.29 20.60
## 8 14.49 20.71 15.37 20.18
## feels_like.morn pressure humidity dew_point wind_speed
## 1 9.05 1025 67 13.04 2.13
## 2 9.84 1023 55 11.03 5.02
## 3 5.00 1031 53 4.48 3.10
## 4 6.74 1029 63 7.99 3.41
## 5 11.04 1022 75 12.99 4.22
## 6 14.25 1019 70 15.11 3.54
## 7 17.15 1014 95 18.19 1.97
## 8 14.53 1015 72 15.29 2.12
## wind_deg wind_gust weather
## 1 195 3.03 800, Clear, clear sky, 01d
## 2 112 9.01 804, Clouds, overcast clouds, 04d
## 3 153 7.03 801, Clouds, few clouds, 02d
## 4 204 7.59 804, Clouds, overcast clouds, 04d
## 5 217 11.24 804, Clouds, overcast clouds, 04d
## 6 214 10.62 804, Clouds, overcast clouds, 04d
## 7 343 7.42 500, Rain, light rain, 10d
## 8 358 7.65 801, Clouds, few clouds, 02d
## clouds pop uvi rain
## 1 5 0.00 3.85 NA
## 2 93 0.00 3.83 NA
## 3 18 0.00 3.60 NA
## 4 100 0.00 3.30 NA
## 5 96 0.00 0.21 NA
## 6 100 0.00 1.00 NA
## 7 100 1.00 1.00 7.86
## 8 19 0.09 1.00 NA
Data Cleaning
Let’s convert all dates/times that are in unix form to a date/time stamp.
Maine.day$dt <- anydate(Maine.day$dt)
Maine.day$sunrise <- anytime(Maine.day$sunrise)
Maine.day$sunset <- anytime(Maine.day$sunset)
Maine.day$moonrise <- anytime(Maine.day$moonrise)
Maine.day$moonset <- anytime(Maine.day$moonset)
Convert the dt variable into three variables, Year, Month, Day,
and then change sunrise, sunset, moonrise, and moonset to only
include timestamps.
Maine.day <- Maine.day %>% separate(dt, c("Year", "Month", "Day"), sep = "-", convert = TRUE, remove = TRUE)
#change sunrise, sunset, moonrise, and moonset to only include timestamps
Maine.day <- Maine.day %>% separate(sunrise, c("Date", "Sunrise"), sep = " ", remove = TRUE) %>% subset(select = -Date)
Maine.day <- Maine.day %>% separate(sunset, c("Date", "Sunset"), sep = " ", remove = TRUE) %>% subset(select = -Date)
Maine.day <- Maine.day %>% separate(moonrise, c("Date", "Moonrise"), sep = " ", remove = TRUE) %>% subset(select = -Date)
Maine.day <- Maine.day %>% separate(moonset, c("Date", "Moonset"), sep = " ", remove = TRUE) %>% subset(select = -Date)
Lastly, combine the nested temp dataframe with Maine.day to create a
new dataframe called Maine.day1, removing unwanted variables.
Maine.day1 <- data.frame(Maine.day, Maine.day$temp) %>% select(Year:moon_phase, pressure:wind_gust, clouds, pop, min, max) %>% rename(mintemp = min, maxtemp = max)
Maine.day1
## Year Month Day Sunrise Sunset Moonrise Moonset
## 1 2021 10 7 06:42:53 18:08:05 07:56:00 19:01:00
## 2 2021 10 8 06:44:06 18:06:18 09:15:00 19:31:00
## 3 2021 10 9 06:45:19 18:04:31 10:36:00 20:05:00
## 4 2021 10 10 06:46:32 18:02:45 11:56:00 20:48:00
## 5 2021 10 11 06:47:46 18:01:00 13:09:00 21:42:00
## 6 2021 10 12 06:49:00 17:59:15 14:12:00 22:45:00
## 7 2021 10 13 06:50:15 17:57:31 15:03:00 23:55:00
## 8 2021 10 14 06:51:30 17:55:48 15:43:00 19:00:00
## moon_phase pressure humidity dew_point wind_speed wind_deg
## 1 0.05 1025 67 13.04 2.13 195
## 2 0.08 1023 55 11.03 5.02 112
## 3 0.12 1031 53 4.48 3.10 153
## 4 0.16 1029 63 7.99 3.41 204
## 5 0.20 1022 75 12.99 4.22 217
## 6 0.25 1019 70 15.11 3.54 214
## 7 0.27 1014 95 18.19 1.97 343
## 8 0.30 1015 72 15.29 2.12 358
## wind_gust clouds pop mintemp maxtemp
## 1 3.03 5 0.00 9.91 21.15
## 2 9.01 93 0.00 10.44 20.79
## 3 7.03 18 0.00 6.20 14.53
## 4 7.59 100 0.00 6.52 16.03
## 5 11.24 96 0.00 11.30 21.26
## 6 10.62 100 0.00 14.21 22.56
## 7 7.42 100 1.00 16.09 21.40
## 8 7.65 19 0.09 14.49 23.48
Awesome! Remember to clean the data returned from the other locations too.
Now, let’s combine all datasets into one called weather, creating a
new variable called location.
Maine.day1 <- Maine.day1 %>% mutate(location = "Maine, US")
Turks.day1 <- Turks.day1 %>% mutate(location = "Turks & Caicos")
Colombia.day1 <- Colombia.day1 %>% mutate(location = "Colombia")
Chile.day1 <- Chile.day1 %>% mutate(location = "Chile")
Argentina.day1 <- Argentina.day1 %>% mutate(location = "Argentina")
weather <- rbind(Maine.day1, Turks.day1, Colombia.day1, Argentina.day1, Chile.day1) %>% relocate(location, .before = Year)
Create New Variables
I’m interested in converting humidity and clouds into categorical
variables with different levels.
I’ll begin with humidity. Lets say that if humidity is less than or equal to 60, there is low humidity, if 60 < humidity is less than or equal to 80, there is medium humidity, and if humidity > 80, there is high humidity.
weather <- weather %>% mutate(humidity.status = as.factor(ifelse(humidity > 80, "High", ifelse(humidity >60, "Medium", "Low"))))
weather$humidity.status <- ordered(weather$humidity.status, levels = c("Low", "Medium", "High"))
Great! Now let’s look at clouds. If clouds is less than or equal to
25, then cloud coverage is low. If clouds > 75 then cloud coverage is
high, and anything in between is medium.
weather <- weather %>% mutate(cloud.coverage = as.factor(ifelse(clouds > 75, "High", ifelse(clouds > 25, "Medium", "Low"))))
weather$cloud.coverage <- ordered(weather$cloud.coverage, levels = c("Low", "Medium", "High"))
Here’s our cleaned dataset:
weather
## location Year Month Day Sunrise Sunset Moonrise
## 1 Maine, US 2021 10 7 06:42:53 18:08:05 07:56:00
## 2 Maine, US 2021 10 8 06:44:06 18:06:18 09:15:00
## 3 Maine, US 2021 10 9 06:45:19 18:04:31 10:36:00
## 4 Maine, US 2021 10 10 06:46:32 18:02:45 11:56:00
## 5 Maine, US 2021 10 11 06:47:46 18:01:00 13:09:00
## 6 Maine, US 2021 10 12 06:49:00 17:59:15 14:12:00
## 7 Maine, US 2021 10 13 06:50:15 17:57:31 15:03:00
## 8 Maine, US 2021 10 14 06:51:30 17:55:48 15:43:00
## 9 Turks & Caicos 2021 10 7 06:37:11 18:26:40 07:42:00
## 10 Turks & Caicos 2021 10 8 06:37:29 18:25:47 08:45:00
## 11 Turks & Caicos 2021 10 9 06:37:48 18:24:55 09:51:00
## 12 Turks & Caicos 2021 10 10 06:38:08 18:24:02 10:58:00
## 13 Turks & Caicos 2021 10 11 06:38:27 18:23:11 12:03:00
## 14 Turks & Caicos 2021 10 12 06:38:48 18:22:20 13:06:00
## 15 Turks & Caicos 2021 10 13 06:39:08 18:21:30 14:02:00
## 16 Turks & Caicos 2021 10 14 06:39:30 18:20:41 14:52:00
## 17 Colombia 2021 10 7 06:36:26 18:37:10 07:38:00
## 18 Colombia 2021 10 8 06:36:21 18:36:41 08:34:00
## 19 Colombia 2021 10 9 06:36:16 18:36:12 09:34:00
## 20 Colombia 2021 10 10 06:36:12 18:35:43 10:35:00
## 21 Colombia 2021 10 11 06:36:08 18:35:16 11:39:00
## 22 Colombia 2021 10 12 06:36:05 18:34:48 12:40:00
## 23 Colombia 2021 10 13 06:36:02 18:34:22 13:39:00
## 24 Colombia 2021 10 14 06:36:00 18:33:56 14:33:00
## 25 Argentina 2021 10 7 05:50:17 18:57:58 06:40:00
## 26 Argentina 2021 10 8 05:48:02 18:59:38 06:59:00
## 27 Argentina 2021 10 9 05:45:48 19:01:19 07:22:00
## 28 Argentina 2021 10 10 05:43:34 19:03:00 07:54:00
## 29 Argentina 2021 10 11 05:41:21 19:04:41 08:37:00
## 30 Argentina 2021 10 12 05:39:09 19:06:23 09:34:00
## 31 Argentina 2021 10 13 05:36:57 19:08:05 10:44:00
## 32 Argentina 2021 10 14 05:34:47 19:09:48 12:01:00
## 33 Chile 2021 10 7 05:37:23 18:54:14 06:25:00
## 34 Chile 2021 10 8 05:34:51 18:56:11 06:40:00
## 35 Chile 2021 10 9 05:32:20 18:58:09 06:58:00
## 36 Chile 2021 10 10 05:29:49 19:00:07 07:24:00
## 37 Chile 2021 10 11 05:27:18 19:02:06 08:02:00
## 38 Chile 2021 10 12 05:24:49 19:04:05 08:57:00
## 39 Chile 2021 10 13 05:22:19 19:06:05 10:09:00
## 40 Chile 2021 10 14 05:19:51 19:08:05 11:31:00
## Moonset moon_phase pressure humidity dew_point
## 1 19:01:00 0.05 1025 67 13.04
## 2 19:31:00 0.08 1023 55 11.03
## 3 20:05:00 0.12 1031 53 4.48
## 4 20:48:00 0.16 1029 63 7.99
## 5 21:42:00 0.20 1022 75 12.99
## 6 22:45:00 0.25 1019 70 15.11
## 7 23:55:00 0.27 1014 95 18.19
## 8 19:00:00 0.30 1015 72 15.29
## 9 19:35:00 0.05 1017 71 23.09
## 10 20:19:00 0.08 1014 76 23.56
## 11 21:09:00 0.12 1014 75 23.64
## 12 22:03:00 0.16 1014 77 24.17
## 13 23:02:00 0.20 1015 76 23.87
## 14 19:00:00 0.25 1015 78 24.19
## 15 00:04:00 0.27 1013 82 21.66
## 16 01:06:00 0.30 1010 80 23.62
## 17 19:52:00 0.05 1015 54 13.41
## 18 20:43:00 0.08 1015 74 15.43
## 19 21:38:00 0.12 1014 64 14.83
## 20 22:36:00 0.16 1016 64 13.61
## 21 23:38:00 0.20 1015 73 15.74
## 22 19:00:00 0.25 1016 73 15.69
## 23 00:38:00 0.27 1015 73 15.59
## 24 01:38:00 0.30 1014 72 16.14
## 25 20:46:00 0.04 1015 53 0.22
## 26 22:15:00 0.08 1005 56 0.80
## 27 19:00:00 0.12 1001 44 3.13
## 28 23:43:00 0.16 1008 52 -1.07
## 29 01:07:00 0.19 1005 37 -4.39
## 30 02:18:00 0.23 1001 40 -1.60
## 31 03:13:00 0.25 987 45 -1.19
## 32 03:53:00 0.30 988 36 -6.64
## 33 20:47:00 0.04 1008 54 -1.26
## 34 22:21:00 0.08 992 71 -0.86
## 35 19:00:00 0.12 994 66 1.87
## 36 23:56:00 0.16 1000 54 -1.56
## 37 01:23:00 0.19 998 74 2.30
## 38 02:37:00 0.23 1000 50 0.82
## 39 03:30:00 0.25 981 65 1.20
## 40 04:06:00 0.30 982 72 -1.37
## wind_speed wind_deg wind_gust clouds pop mintemp maxtemp
## 1 2.13 195 3.03 5 0.00 9.91 21.15
## 2 5.02 112 9.01 93 0.00 10.44 20.79
## 3 3.10 153 7.03 18 0.00 6.20 14.53
## 4 3.41 204 7.59 100 0.00 6.52 16.03
## 5 4.22 217 11.24 96 0.00 11.30 21.26
## 6 3.54 214 10.62 100 0.00 14.21 22.56
## 7 1.97 343 7.42 100 1.00 16.09 21.40
## 8 2.12 358 7.65 19 0.09 14.49 23.48
## 9 9.85 73 10.56 32 0.94 27.86 29.01
## 10 8.02 81 8.33 4 0.78 27.69 28.28
## 11 7.95 107 8.29 12 0.76 27.85 28.54
## 12 9.35 113 10.22 74 0.57 27.97 28.65
## 13 9.18 92 10.04 21 0.22 28.12 28.59
## 14 9.29 88 10.56 100 1.00 27.28 28.35
## 15 15.50 85 17.47 100 1.00 24.90 27.33
## 16 10.23 117 10.94 97 0.96 26.87 27.45
## 17 3.22 110 4.69 79 1.00 12.00 23.57
## 18 2.07 108 3.15 76 1.00 12.70 19.31
## 19 2.13 81 2.65 69 1.00 13.33 20.96
## 20 1.73 99 2.52 84 1.00 13.52 20.05
## 21 2.24 103 3.17 81 1.00 13.09 20.05
## 22 2.38 100 3.16 95 1.00 13.25 20.16
## 23 1.86 89 2.43 95 0.99 13.91 19.55
## 24 1.57 59 2.32 99 1.00 13.30 20.20
## 25 12.48 238 17.91 100 0.00 2.77 9.70
## 26 15.76 237 20.20 81 0.00 4.77 12.19
## 27 12.55 316 19.46 100 0.00 3.92 16.86
## 28 11.95 246 18.01 91 0.23 4.57 12.31
## 29 6.26 354 8.18 83 0.00 4.87 12.95
## 30 8.90 311 14.76 48 0.00 6.11 14.42
## 31 12.04 340 19.46 25 0.00 7.47 11.88
## 32 13.97 255 17.83 3 0.28 3.48 7.50
## 33 8.29 227 15.59 100 0.63 1.91 8.52
## 34 13.01 262 24.04 84 0.80 4.22 9.04
## 35 11.15 342 23.77 100 0.56 2.37 9.52
## 36 8.61 325 15.78 1 0.67 3.33 9.11
## 37 8.93 328 16.99 100 0.21 4.85 9.80
## 38 7.10 346 13.10 55 0.00 4.14 11.75
## 39 10.27 339 18.34 100 0.25 5.37 10.91
## 40 6.27 334 12.95 100 0.94 -0.21 5.28
## humidity.status cloud.coverage
## 1 Medium Low
## 2 Low High
## 3 Low Low
## 4 Medium High
## 5 Medium High
## 6 Medium High
## 7 High High
## 8 Medium Low
## 9 Medium Medium
## 10 Medium Low
## 11 Medium Low
## 12 Medium Medium
## 13 Medium Low
## 14 Medium High
## 15 High High
## 16 Medium High
## 17 Low High
## 18 Medium High
## 19 Medium Medium
## 20 Medium High
## 21 Medium High
## 22 Medium High
## 23 Medium High
## 24 Medium High
## 25 Low High
## 26 Low High
## 27 Low High
## 28 Low High
## 29 Low High
## 30 Low Medium
## 31 Low Low
## 32 Low Low
## 33 Low High
## 34 Medium High
## 35 Medium High
## 36 Low Low
## 37 Medium High
## 38 Low Medium
## 39 Medium High
## 40 Medium High
Contingency Tables
With the new categorical variables I’ve created, let’s create some
contingency tables. tabz1 will show the counts of observations within
each level combination of cloud.coverage and humidity.status.
tabz1 <- table(weather$humidity.status, weather$cloud.coverage, deparse.level = 2)
tabz1
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 4 2 8
## Medium 5 3 16
## High 0 0 2
It looks like there were 8 forecast observations that predicted high cloud coverage and low humidity, and 16 observations that predicted high cloud coverage and medium humidity. There weren’t many observations that predicted high humidity, but it appears that when high humidity was predicted, it was most often coupled with high cloud coverage.
tabz2 will show the counts of days within each level combination of
cloud.coverage and humidity.status, separated by location.
tabz2 <- table(weather$humidity.status, weather$cloud.coverage, weather$location, deparse.level = 2)
tabz2
## , , weather$location = Argentina
##
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 2 1 5
## Medium 0 0 0
## High 0 0 0
##
## , , weather$location = Chile
##
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 1 1 1
## Medium 0 0 5
## High 0 0 0
##
## , , weather$location = Colombia
##
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 0 0 1
## Medium 0 1 6
## High 0 0 0
##
## , , weather$location = Maine, US
##
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 1 0 1
## Medium 2 0 3
## High 0 0 1
##
## , , weather$location = Turks & Caicos
##
## weather$cloud.coverage
## weather$humidity.status Low Medium High
## Low 0 0 0
## Medium 3 2 2
## High 0 0 1
This table gives us an idea of how the relationship between humidity status and cloud coverage can change based on location. An observation to note is that the locations in Chile and Argentina, which are close to the south pole, most often have medium humidity and high cloud coverage or low humidity and high cloud coverage, respectively. I’d be interested to find out why this is the case, or if the data only appears this way because we’re not working with many observations. It makes sense that Puerto Williams, Chile would have higher humidity than the Argentina location because it is a costal city.
Numerical Summaries
Now that we’ve explored our categorical variables, let’s take a look at some numeric summaries.
The table below lists the average maximum temperature, average minimum temperature, their respective standard deviations, and the interquartile range for each location.
weather$location <- ordered(weather$location, levels = c("Chile", "Argentina", "Colombia", "Turks & Caicos", "Maine, US")) #convert location into an ordered variable
weather %>% group_by(location) %>% summarise(avghigh = mean(maxtemp), avglow = mean(mintemp), sdhigh = sd(maxtemp), sdlow = sd(mintemp), IQR = IQR(maxtemp))
## # A tibble: 5 x 6
## location avghigh avglow sdhigh sdlow IQR
## <ord> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 Chile 9.24 3.25 1.92 1.83 1.17
## 2 Argentina 12.2 4.74 2.82 1.49 1.98
## 3 Colombia 20.5 13.1 1.34 0.574 0.465
## 4 Turks & Caicos 28.3 27.3 0.589 1.06 0.532
## 5 Maine, US 20.2 11.1 3.15 3.64 2.09
It’s clear that the average high and average low peak near the middle of the globe, with greater variation near the poles. These numbers could not only be affected by latitude, but also by elevation.
Now we’ll look at the average humidity forecasted for each location, as well as their standard deviations.
weather %>% group_by(location) %>% summarise(avg_humidity = mean(humidity), sd.humidity = sd(humidity))
## # A tibble: 5 x 3
## location avg_humidity sd.humidity
## <ord> <dbl> <dbl>
## 1 Chile 63.2 9.33
## 2 Argentina 45.4 7.60
## 3 Colombia 68.4 7.11
## 4 Turks & Caicos 76.9 3.31
## 5 Maine, US 68.8 13.2
Here we see a clear relationship between location (organized by latitude) and humidity. The locations nearest to the poles generally have the lowest humidity, but the Turks and Caicos Islands as well as the Chile location have the highest humidity, most likely because they are close to or in the ocean.
Data Visualization
Now I’ll use some tools for visualizing the data we’ve collected from the API.
Let’s visualize the second contingency table we made above using a bar
graph with humidity.status and cloud.coverage, separating the
results by location.
sum.tab <- weather %>% group_by(location, humidity.status, cloud.coverage) %>% summarise(count = n())
g6 <- ggplot(sum.tab, aes(x = humidity.status, y = count))
g6 + geom_bar(aes(fill = cloud.coverage), stat = "identity", position = "dodge") + facet_wrap(~ location) + labs(title = "8 Day Forecast: Humidity Status vs. Cloud Coverage by Location", x = "Humidity Status") + scale_fill_discrete(name = "Cloud Coverage") + theme(axis.text.x = element_text(angle = 45, vjust = .8, hjust = 1))
Here we can see the observation we made earlier about Chile and Argentina having high cloud coverage, but most often medium or low humidity, respectively. In addition, it’s also easier to see that the forecast for Colombia most often includes medium humidity and high cloud coverage. A puzzling observation is that Turks and Caicos always has medium humidity, but either low or high cloud coverage.
Next, I’ll look at how location affects minimum and maximum daily temperature using boxplots.
max.means <- weather %>% group_by(location) %>% summarise(average = mean(maxtemp))
g <- ggplot(weather, aes(x = location, y = maxtemp))
g + geom_boxplot(fill = "grey") + geom_point(max.means, mapping = aes(x = location, y = average), color = "purple") + geom_line(max.means, mapping = aes(x = location, y = average, group = 1), color = "purple") + labs(title = "8 Day Forecast: Maximum Daily Temperature",x = "Location (highest to lowest latitude)", y = "Maximum Daily Temperature (C)") + coord_flip()
min.means <- weather %>% group_by(location) %>% summarise(average = mean(mintemp))
g1 <- ggplot(weather, aes(x = location, y = mintemp))
g1 + geom_boxplot(fill = "grey") + geom_point(min.means, mapping = aes(x = location, y = average), color = "green") + geom_line(min.means, mapping = aes(x = location, y = average, group = 1), color = "green") + labs(title = "8 Day Forecast: Minimum Daily Temperature", x = "Location (highest to lowest latitude)", y = "Minimum Daily Temperature (C)") + coord_flip()
Both boxplots are fairly consistent with each other, showing Turks and Caicos as having the highest minimum and maximum temperature, and locations at the highest and lowest latitudes as having the lowest minimum and maximum temperatures. Although Cucata, Colombia is closer to the equator than Turks and Caicos Islands, a possible reason as to why it may not be warmer is its elevation at 320 meters (1,050 ft).
Now we’ll turn to exploring humidity as a quantitative variable,
creating a histogram with density plots of humidity facetted by
location. We’ll also create a boxplot of the same information.
g2 <- ggplot(weather, aes(x = humidity))
g2 + geom_histogram(bins = 20, aes(y = ..density..)) + geom_density(color = "aquamarine", fill = "aquamarine") + facet_wrap(~location) + labs(title = "8 Day Forecast: Histogram of Humidity by Location", x = "Humidity (%)")
hum.means <- weather %>% group_by(location) %>% summarise(average = mean(humidity))
g3 <- ggplot(weather, aes(x = location, y = humidity))
g3 + geom_boxplot(fill = "grey") + geom_point(hum.means, mapping = aes(x = location, y = average), color = "red") + geom_line(hum.means, mapping = aes(x = location, y = average, group = 1), color = "red") + labs(title = "8 Day Forecast: Average Daily Humidity by Location", x = "Location (highest to lowest latitude)", y = "Average Daily Humidity (C)") + coord_flip()
This histogram and boxplot are very helpful in displaying how humidity
varies by location. Generally, it looks like humidity follows a similar
trend to temperature, varying with latitude. However, it’s clear that
the cities closest to the ocean (Turks & Caicos and Puerto Williams,
Chile) display higher humidity levels than expected if humidity.status
only depended on latitude.
To prepare for the next plot, I’ll convert all variables with timestamps to numeric variables of time since 00:00:00, calculated in minutes.
weather$Sunrise <- round(60 * 24 * as.numeric(times(weather$Sunrise)),digits = 2)
weather$Sunset <- round(60 * 24 * as.numeric(times(weather$Sunset)),digits = 2)
weather$Moonrise <- round(60 * 24 * as.numeric(times(weather$Moonrise)), digits = 2)
weather$Moonset <- round(60 * 24 * as.numeric(times(weather$Moonset)), digits = 2)
weather
## location Year Month Day Sunrise Sunset Moonrise
## 1 Maine, US 2021 10 7 402.88 1088.08 476
## 2 Maine, US 2021 10 8 404.10 1086.30 555
## 3 Maine, US 2021 10 9 405.32 1084.52 636
## 4 Maine, US 2021 10 10 406.53 1082.75 716
## 5 Maine, US 2021 10 11 407.77 1081.00 789
## 6 Maine, US 2021 10 12 409.00 1079.25 852
## 7 Maine, US 2021 10 13 410.25 1077.52 903
## 8 Maine, US 2021 10 14 411.50 1075.80 943
## 9 Turks & Caicos 2021 10 7 397.18 1106.67 462
## 10 Turks & Caicos 2021 10 8 397.48 1105.78 525
## 11 Turks & Caicos 2021 10 9 397.80 1104.92 591
## 12 Turks & Caicos 2021 10 10 398.13 1104.03 658
## 13 Turks & Caicos 2021 10 11 398.45 1103.18 723
## 14 Turks & Caicos 2021 10 12 398.80 1102.33 786
## 15 Turks & Caicos 2021 10 13 399.13 1101.50 842
## 16 Turks & Caicos 2021 10 14 399.50 1100.68 892
## 17 Colombia 2021 10 7 396.43 1117.17 458
## 18 Colombia 2021 10 8 396.35 1116.68 514
## 19 Colombia 2021 10 9 396.27 1116.20 574
## 20 Colombia 2021 10 10 396.20 1115.72 635
## 21 Colombia 2021 10 11 396.13 1115.27 699
## 22 Colombia 2021 10 12 396.08 1114.80 760
## 23 Colombia 2021 10 13 396.03 1114.37 819
## 24 Colombia 2021 10 14 396.00 1113.93 873
## 25 Argentina 2021 10 7 350.28 1137.97 400
## 26 Argentina 2021 10 8 348.03 1139.63 419
## 27 Argentina 2021 10 9 345.80 1141.32 442
## 28 Argentina 2021 10 10 343.57 1143.00 474
## 29 Argentina 2021 10 11 341.35 1144.68 517
## 30 Argentina 2021 10 12 339.15 1146.38 574
## 31 Argentina 2021 10 13 336.95 1148.08 644
## 32 Argentina 2021 10 14 334.78 1149.80 721
## 33 Chile 2021 10 7 337.38 1134.23 385
## 34 Chile 2021 10 8 334.85 1136.18 400
## 35 Chile 2021 10 9 332.33 1138.15 418
## 36 Chile 2021 10 10 329.82 1140.12 444
## 37 Chile 2021 10 11 327.30 1142.10 482
## 38 Chile 2021 10 12 324.82 1144.08 537
## 39 Chile 2021 10 13 322.32 1146.08 609
## 40 Chile 2021 10 14 319.85 1148.08 691
## Moonset moon_phase pressure humidity dew_point wind_speed
## 1 1141 0.05 1025 67 13.04 2.13
## 2 1171 0.08 1023 55 11.03 5.02
## 3 1205 0.12 1031 53 4.48 3.10
## 4 1248 0.16 1029 63 7.99 3.41
## 5 1302 0.20 1022 75 12.99 4.22
## 6 1365 0.25 1019 70 15.11 3.54
## 7 1435 0.27 1014 95 18.19 1.97
## 8 1140 0.30 1015 72 15.29 2.12
## 9 1175 0.05 1017 71 23.09 9.85
## 10 1219 0.08 1014 76 23.56 8.02
## 11 1269 0.12 1014 75 23.64 7.95
## 12 1323 0.16 1014 77 24.17 9.35
## 13 1382 0.20 1015 76 23.87 9.18
## 14 1140 0.25 1015 78 24.19 9.29
## 15 4 0.27 1013 82 21.66 15.50
## 16 66 0.30 1010 80 23.62 10.23
## 17 1192 0.05 1015 54 13.41 3.22
## 18 1243 0.08 1015 74 15.43 2.07
## 19 1298 0.12 1014 64 14.83 2.13
## 20 1356 0.16 1016 64 13.61 1.73
## 21 1418 0.20 1015 73 15.74 2.24
## 22 1140 0.25 1016 73 15.69 2.38
## 23 38 0.27 1015 73 15.59 1.86
## 24 98 0.30 1014 72 16.14 1.57
## 25 1246 0.04 1015 53 0.22 12.48
## 26 1335 0.08 1005 56 0.80 15.76
## 27 1140 0.12 1001 44 3.13 12.55
## 28 1423 0.16 1008 52 -1.07 11.95
## 29 67 0.19 1005 37 -4.39 6.26
## 30 138 0.23 1001 40 -1.60 8.90
## 31 193 0.25 987 45 -1.19 12.04
## 32 233 0.30 988 36 -6.64 13.97
## 33 1247 0.04 1008 54 -1.26 8.29
## 34 1341 0.08 992 71 -0.86 13.01
## 35 1140 0.12 994 66 1.87 11.15
## 36 1436 0.16 1000 54 -1.56 8.61
## 37 83 0.19 998 74 2.30 8.93
## 38 157 0.23 1000 50 0.82 7.10
## 39 210 0.25 981 65 1.20 10.27
## 40 246 0.30 982 72 -1.37 6.27
## wind_deg wind_gust clouds pop mintemp maxtemp
## 1 195 3.03 5 0.00 9.91 21.15
## 2 112 9.01 93 0.00 10.44 20.79
## 3 153 7.03 18 0.00 6.20 14.53
## 4 204 7.59 100 0.00 6.52 16.03
## 5 217 11.24 96 0.00 11.30 21.26
## 6 214 10.62 100 0.00 14.21 22.56
## 7 343 7.42 100 1.00 16.09 21.40
## 8 358 7.65 19 0.09 14.49 23.48
## 9 73 10.56 32 0.94 27.86 29.01
## 10 81 8.33 4 0.78 27.69 28.28
## 11 107 8.29 12 0.76 27.85 28.54
## 12 113 10.22 74 0.57 27.97 28.65
## 13 92 10.04 21 0.22 28.12 28.59
## 14 88 10.56 100 1.00 27.28 28.35
## 15 85 17.47 100 1.00 24.90 27.33
## 16 117 10.94 97 0.96 26.87 27.45
## 17 110 4.69 79 1.00 12.00 23.57
## 18 108 3.15 76 1.00 12.70 19.31
## 19 81 2.65 69 1.00 13.33 20.96
## 20 99 2.52 84 1.00 13.52 20.05
## 21 103 3.17 81 1.00 13.09 20.05
## 22 100 3.16 95 1.00 13.25 20.16
## 23 89 2.43 95 0.99 13.91 19.55
## 24 59 2.32 99 1.00 13.30 20.20
## 25 238 17.91 100 0.00 2.77 9.70
## 26 237 20.20 81 0.00 4.77 12.19
## 27 316 19.46 100 0.00 3.92 16.86
## 28 246 18.01 91 0.23 4.57 12.31
## 29 354 8.18 83 0.00 4.87 12.95
## 30 311 14.76 48 0.00 6.11 14.42
## 31 340 19.46 25 0.00 7.47 11.88
## 32 255 17.83 3 0.28 3.48 7.50
## 33 227 15.59 100 0.63 1.91 8.52
## 34 262 24.04 84 0.80 4.22 9.04
## 35 342 23.77 100 0.56 2.37 9.52
## 36 325 15.78 1 0.67 3.33 9.11
## 37 328 16.99 100 0.21 4.85 9.80
## 38 346 13.10 55 0.00 4.14 11.75
## 39 339 18.34 100 0.25 5.37 10.91
## 40 334 12.95 100 0.94 -0.21 5.28
## humidity.status cloud.coverage
## 1 Medium Low
## 2 Low High
## 3 Low Low
## 4 Medium High
## 5 Medium High
## 6 Medium High
## 7 High High
## 8 Medium Low
## 9 Medium Medium
## 10 Medium Low
## 11 Medium Low
## 12 Medium Medium
## 13 Medium Low
## 14 Medium High
## 15 High High
## 16 Medium High
## 17 Low High
## 18 Medium High
## 19 Medium Medium
## 20 Medium High
## 21 Medium High
## 22 Medium High
## 23 Medium High
## 24 Medium High
## 25 Low High
## 26 Low High
## 27 Low High
## 28 Low High
## 29 Low High
## 30 Low Medium
## 31 Low Low
## 32 Low Low
## 33 Low High
## 34 Medium High
## 35 Medium High
## 36 Low Low
## 37 Medium High
## 38 Low Medium
## 39 Medium High
## 40 Medium High
The scatterplot below allows us to look at how sunrise time changes over the course of 8 days in each location.
g4 <- ggplot(weather, aes(x = Day, y = Sunrise))
g4 + geom_point(aes(color = location), size = 2) + scale_color_discrete(name = "Location") + labs(title = "8 Day Forecast for Sunrise Time by Location", x = "Day of the Month", y = "Sunrise Time (minutes after 12am)")
I think this plot is especially cool to look at, since one can clearly distinguish which locations are in the southern hemisphere, and which are in the northern hemisphere. While the sun is rising increasingly later in the northern hemisphere, it’s rising increasingly earlier in the southern hemisphere.
Lastly, we’ll explore the relationship between maximum daily temperature and daily pressure in the scatterplot below.
g5 <- ggplot(weather, aes(x = maxtemp, y = pressure))
g5 + geom_point(aes(color = location), size = 2) + geom_smooth(method = lm, formula = y~poly(x,2), color = "black") + scale_color_discrete(name = "Location") + labs(title = "8 Day Forecast: Maximum Daily Temperature vs. Daily Pressure", x = "Maximum Daily Temperature (C)", y = "Daily Pressure (millibars)") + scale_size_continuous()
There seems to be a medium-strength positive relationship between temperature and pressure. However, the data from the Turks and Caicos Islands appears to deviate slightly from the general trend, which is why I decided to fit a quadratic model. To fit a more accurate model, we may need to pull data from more locations from the API.
Final Thoughts
Extracting data from an API follows a generally simple process, although different APIs have varying syntax for how to assemble the URL. Hopefully, this vignette was helpful in exemplifying how to access a typical API, clean the data, and perform a few basic analyses.
To read my blog post about this project, visit http://atbiggie.github.io.