Lab 03 - Exploratory Data Analysis
Learning Goals
- Read in and get familiar with the meteorology dataset
- Step through the EDA “checklist” presented in the class slides
- Practice making exploratory graphs
As you do this, think about what questions you would like to ask regarding this data. What would you ask a collaborator who was more familiar with it?
Lab Description
We will work with the meteorological data presented in lecture. Recall the dataset consists of weather station readings in the continental US.
The objectives of the lab are to find the weather station with the highest elevation and look at patterns in the time series of its wind speed and temperature.
1. Read in the data
First download and then read in with data.table::fread(). This is slightly faster than some of the more common functions, such as read.table, but it produces a different type of object, which is why we need to convert it into a data.frame after reading it in.
download.file(
"https://raw.githubusercontent.com/USCbiostats/data-science-data/master/02_met/met_all.gz",
destfile = file.path("~", "Downloads", "met_all.gz"),
method = "libcurl",
timeout = 60
)
met <- data.table::fread(file.path("~", "Downloads", "met_all.gz"))
met <- as.data.frame(met)
2. Check the dimensions, headers, footers.
How many columns, rows are there? Some useful functions for this are dim, head, and tail.
3. Take a look at the variables.
Show the type (class) of each variable (hint: try the str function).
What are the names of the key variables related to our question of interest?
4. Take a closer look at the key variables.
table(met$year)
##
## 2019
## 2377343
table(met$day)
##
## 1 2 3 4 5 6 7 8 9 10 11 12 13
## 75975 75923 76915 76594 76332 76734 77677 77766 75366 75450 76187 75052 76906
## 14 15 16 17 18 19 20 21 22 23 24 25 26
## 77852 76217 78015 78219 79191 76709 75527 75786 78312 77413 76965 76806 79114
## 27 28 29 30 31
## 79789 77059 71712 74931 74849
table(met$hour)
##
## 0 1 2 3 4 5 6 7 8 9 10
## 99434 93482 93770 96703 110504 112128 106235 101985 100310 102915 101880
## 11 12 13 14 15 16 17 18 19 20 21
## 100470 103605 97004 96507 97635 94942 94184 100179 94604 94928 96070
## 22 23
## 94046 93823
summary(met$temp)
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -40.00 19.60 23.50 23.59 27.80 56.00 60089
summary(met$elev)
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -13.0 101.0 252.0 415.8 400.0 9999.0
summary(met$wind.sp)
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## 0.00 0.00 2.10 2.46 3.60 36.00 79693
It looks like the elevation variable has observations with 9999.0, which is probably an indicator for missing. We should take a deeper look at the data dictionary to confirm. The wind speed variable is OK but there is a lot of missing data.
After checking the data we should make the appropriate modifications. Replace elevations with 9999 as NA.
met[met$elev==9999.0, ] <- NA
summary(met$elev)
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -13 101 252 413 400 4113 710
At what elevation is the highest weather station?
We also have the issue of the minimum temperature being -40C, which seems implausible, so we should remove those observations.
met <- met[met$temp > -40, ]
head(met[order(met$temp), ])
## USAFID WBAN year month day hour min lat lon elev wind.dir
## 1203053 722817 3068 2019 8 1 0 56 38.767 -104.3 1838 190
## 1203055 722817 3068 2019 8 1 1 56 38.767 -104.3 1838 180
## 1203128 722817 3068 2019 8 3 11 56 38.767 -104.3 1838 NA
## 1203129 722817 3068 2019 8 3 12 56 38.767 -104.3 1838 NA
## 1203222 722817 3068 2019 8 6 21 56 38.767 -104.3 1838 280
## 1203225 722817 3068 2019 8 6 22 56 38.767 -104.3 1838 240
## wind.dir.qc wind.type.code wind.sp wind.sp.qc ceiling.ht ceiling.ht.qc
## 1203053 5 N 7.2 5 NA 9
## 1203055 5 N 7.7 5 NA 9
## 1203128 9 C 0.0 5 NA 9
## 1203129 9 C 0.0 5 NA 9
## 1203222 5 N 2.6 5 NA 9
## 1203225 5 N 7.7 5 NA 9
## ceiling.ht.method sky.cond vis.dist vis.dist.qc vis.var vis.var.qc
## 1203053 9 N NA 9 N 5
## 1203055 9 N NA 9 N 5
## 1203128 9 N NA 9 N 5
## 1203129 9 N NA 9 N 5
## 1203222 9 N NA 9 N 5
## 1203225 9 N NA 9 N 5
## temp temp.qc dew.point dew.point.qc atm.press atm.press.qc rh
## 1203053 -17.2 5 NA 9 NA 9 NA
## 1203055 -17.2 5 NA 9 NA 9 NA
## 1203128 -17.2 5 NA 9 NA 9 NA
## 1203129 -17.2 5 NA 9 NA 9 NA
## 1203222 -17.2 5 NA 9 NA 9 NA
## 1203225 -17.2 5 NA 9 NA 9 NA
There are still some suspiciously low values for temperature (-17.2C), but we will deal with those later.
We should also check the wind speed variable for any abnormalities.
How many missing values are there in the wind.sp variable?
5. Check the data against an external data source.
We should check the suspicious temperature value (where is it located?) and validate that the range of elevations make sense (-13m to 4113m).
Google is your friend here.
Fix any problems that arise in your checks.
Where was the location for the coldest temperature readings (-17.2C)? Do these seem reasonable in context?
Does the range of values for elevation make sense? Why or why not?
6. Calculate summary statistics
Remember to keep the initial question in mind. We want to pick out the weather station with maximum elevation and examine its wind speed and temperature.
Some ideas: select the weather station with maximum elevation; look at the correlation between temperature and wind speed; look at the correlation between temperature and wind speed with hour and day of the month.
elev <- met[which(met$elev == max(met$elev, na.rm = TRUE)), ]
summary(elev)
## USAFID WBAN year month day
## Min. :720385 Min. :419 Min. :2019 Min. :8 Min. : 1.0
## 1st Qu.:720385 1st Qu.:419 1st Qu.:2019 1st Qu.:8 1st Qu.: 8.0
## Median :720385 Median :419 Median :2019 Median :8 Median :16.0
## Mean :720385 Mean :419 Mean :2019 Mean :8 Mean :16.1
## 3rd Qu.:720385 3rd Qu.:419 3rd Qu.:2019 3rd Qu.:8 3rd Qu.:24.0
## Max. :720385 Max. :419 Max. :2019 Max. :8 Max. :31.0
##
## hour min lat lon elev
## Min. : 0.00 Min. : 6.00 Min. :39.8 Min. :-105.8 Min. :4113
## 1st Qu.: 6.00 1st Qu.:13.00 1st Qu.:39.8 1st Qu.:-105.8 1st Qu.:4113
## Median :12.00 Median :36.00 Median :39.8 Median :-105.8 Median :4113
## Mean :11.66 Mean :34.38 Mean :39.8 Mean :-105.8 Mean :4113
## 3rd Qu.:18.00 3rd Qu.:53.00 3rd Qu.:39.8 3rd Qu.:-105.8 3rd Qu.:4113
## Max. :23.00 Max. :59.00 Max. :39.8 Max. :-105.8 Max. :4113
##
## wind.dir wind.dir.qc wind.type.code wind.sp
## Min. : 10.0 Length:2117 Length:2117 Min. : 0.000
## 1st Qu.:250.0 Class :character Class :character 1st Qu.: 4.100
## Median :300.0 Mode :character Mode :character Median : 6.700
## Mean :261.5 Mean : 7.245
## 3rd Qu.:310.0 3rd Qu.: 9.800
## Max. :360.0 Max. :21.100
## NA's :237 NA's :168
## wind.sp.qc ceiling.ht ceiling.ht.qc ceiling.ht.method
## Length:2117 Min. : 30 Min. :5.000 Length:2117
## Class :character 1st Qu.: 2591 1st Qu.:5.000 Class :character
## Mode :character Median :22000 Median :5.000 Mode :character
## Mean :15145 Mean :5.008
## 3rd Qu.:22000 3rd Qu.:5.000
## Max. :22000 Max. :9.000
## NA's :4
## sky.cond vis.dist vis.dist.qc vis.var
## Length:2117 Min. : 0 Length:2117 Length:2117
## Class :character 1st Qu.:16093 Class :character Class :character
## Mode :character Median :16093 Mode :character Mode :character
## Mean :15913
## 3rd Qu.:16093
## Max. :16093
## NA's :683
## vis.var.qc temp temp.qc dew.point
## Length:2117 Min. : 1.00 Length:2117 Min. :-6.0000
## Class :character 1st Qu.: 6.00 Class :character 1st Qu.: 0.0000
## Mode :character Median : 8.00 Mode :character Median : 0.0000
## Mean : 8.13 Mean : 0.8729
## 3rd Qu.:10.00 3rd Qu.: 2.0000
## Max. :15.00 Max. : 7.0000
##
## dew.point.qc atm.press atm.press.qc rh
## Length:2117 Min. : NA Min. :9 Min. :53.63
## Class :character 1st Qu.: NA 1st Qu.:9 1st Qu.:58.10
## Mode :character Median : NA Median :9 Median :61.39
## Mean :NaN Mean :9 Mean :60.62
## 3rd Qu.: NA 3rd Qu.:9 3rd Qu.:61.85
## Max. : NA Max. :9 Max. :70.01
## NA's :2117
Note that to find the maximum elevation, we had to add na.rm = TRUE, because the elevation variable contains missing values. This is an example of how missing values can quickly propagate throughout an analysis (as the “maximum” of 1, 2, and NA is NA, because it cannot be defined).
Also note that we used the which function to tell us which elements of the logical comparison are TRUE. We did this because some of them were NA, which can lead to issues when subsetting by a logical variable.
cor(elev$temp, elev$wind.sp, use="complete")
## [1] -0.09373843
cor(elev$temp, elev$hour, use="complete")
## [1] 0.4397261
cor(elev$wind.sp, elev$day, use="complete")
## [1] 0.3643079
cor(elev$wind.sp, elev$hour, use="complete")
## [1] 0.08807315
cor(elev$temp, elev$day, use="complete")
## [1] -0.003857766
The use="complete" argument is another thing we added to avoid compounding NAs.
7. Exploratory graphs
We should look at the distributions of all of the key variables to make sure there are no remaining issues with the data.
Use the hist function to make histograms of the elevation, temperature, and wind speed variables for the whole dataset
One thing we should consider for later analyses is to log transform wind speed and elevation as they are very skewed.
Look at where the weather station with highest elevation is located.
leaflet(elev) %>%
addProviderTiles('OpenStreetMap') %>%
addCircles(lat=~lat,lng=~lon, opacity=1, fillOpacity=1, radius=100)
Look at the time series of temperature and wind speed at this location. For this we will need to create a date-time variable for the x-axis.
library(lubridate)
elev$date <- with(elev, ymd_h(paste(year, month, day, hour, sep= ' ')))
summary(elev$date)
## Min. 1st Qu.
## "2019-08-01 00:00:00.0000" "2019-08-08 11:00:00.0000"
## Median Mean
## "2019-08-16 22:00:00.0000" "2019-08-16 14:09:56.8823"
## 3rd Qu. Max.
## "2019-08-24 11:00:00.0000" "2019-08-31 22:00:00.0000"
elev <- elev[order(elev$date), ]
head(elev)
## USAFID WBAN year month day hour min lat lon elev wind.dir
## 221697 720385 419 2019 8 1 0 36 39.8 -105.766 4113 170
## 221698 720385 419 2019 8 1 0 54 39.8 -105.766 4113 100
## 221699 720385 419 2019 8 1 1 12 39.8 -105.766 4113 90
## 221700 720385 419 2019 8 1 1 35 39.8 -105.766 4113 110
## 221701 720385 419 2019 8 1 1 53 39.8 -105.766 4113 120
## 221702 720385 419 2019 8 1 2 12 39.8 -105.766 4113 120
## wind.dir.qc wind.type.code wind.sp wind.sp.qc ceiling.ht ceiling.ht.qc
## 221697 5 N 8.8 5 1372 5
## 221698 5 N 2.6 5 1372 5
## 221699 5 N 3.1 5 1981 5
## 221700 5 N 4.1 5 2134 5
## 221701 5 N 4.6 5 2134 5
## 221702 5 N 6.2 5 22000 5
## ceiling.ht.method sky.cond vis.dist vis.dist.qc vis.var vis.var.qc temp
## 221697 M N NA 9 N 5 9
## 221698 M N NA 9 N 5 9
## 221699 M N NA 9 N 5 9
## 221700 M N NA 9 N 5 9
## 221701 M N NA 9 N 5 9
## 221702 9 N NA 9 N 5 9
## temp.qc dew.point dew.point.qc atm.press atm.press.qc rh
## 221697 5 1 5 NA 9 57.61039
## 221698 5 1 5 NA 9 57.61039
## 221699 5 2 5 NA 9 61.85243
## 221700 5 2 5 NA 9 61.85243
## 221701 5 2 5 NA 9 61.85243
## 221702 5 2 5 NA 9 61.85243
## date
## 221697 2019-08-01 00:00:00
## 221698 2019-08-01 00:00:00
## 221699 2019-08-01 01:00:00
## 221700 2019-08-01 01:00:00
## 221701 2019-08-01 01:00:00
## 221702 2019-08-01 02:00:00
With the date-time variable we can plot the time series of temperature and wind speed.
Use the plot function to make line graphs of temperature vs. date and wind speed vs. date
Summarize any trends that you see in these time series plots.
8. Ask questions
By now, you might have some specific questions about how the data was gathered and what some of the different variables and values mean. Alternatively, maybe you have an idea for how some of the variable should be related and you want to explore that relationship. In a real-world analysis, these questions could potentially be answered by a collaborator, who may have been part of the team that collected the data.
What questions do you have about the data?
If you haven’t already, now would be a good time to look at the accompanying data dictionary for this dataset and see if it can answer any of your questions. If you have questions about the nature of the dataset and how it was gathered, this might be able to help.
For questions about variables in the dataset or relationships between them, try making some more exploratory plots. Do you see the patterns you would expect?
There are many different types of summaries and visualization strategies that we have not discussed, but which could provide interesting perspectives on the data.
Some other useful plotting functions include:
pairsfor making all pairwise scatter plots in a dataset with >2 dimensions.heatmapand/orcorrplot(from thecorrplotpackage) for visualizing matrices in general or correlation matrices in particular.imagea low-level matrix visualization functionbarplot, especially withtable, for visualizing frequencies of categorical variables.