3. Initial EDA (this week)

We have data! Now is the time to explore it and to make some initial models.

Instructions

Set-up

You can choose whether to use your RStudio cloud account or your own computers for this task (For RStudioCloud it seems that only one person can use a project effectively, so you’re welcome to use your shared space, but set up individual projects).

You can always copy/paste your final results together.

EVERYONE (individually)

Set up a project and get the data

1. Use Tutorial 2 to create a new project (or use a sensible existing project). If you are using the shared cloud space, work in individual projects. Keep following the tutorial to create a new .RmD file.

2. At the very top, either create or edit a code chunk that looks like this. This will stop the annoying library packages loading when you knit. Here’s the text to copy/paste:

   • {r setup, include=FALSE}

   • knitr::opts_chunk$set(echo = TRUE, message=FALSE,warning=FALSE)

3. Below that, create a new code chunk to load these libraries. You might need to install some using Tutorial 2.2.

library(readxl)  # read in data
library(sf)      # spatial
library(sp)      # spatial
library(raster)  # spatial
library(terra)   # spatial
library(tidyverse) # ALL
library(tmap)    # mapping

4. Go to visual mode. Create a new header 1 called Exploratory Analysis
   
5. Click this link to download the data: CLICK HERE. and put it in your project file. Use these commands to read it into R. The first one will create a data.frame (table) of your data. The second will create a “spatial version” that is easy to map. Press knit and check all is well.

DataIn <- read_excel("STAT462_TrainingData.xlsx")
DataIn <- na.omit(DataIn)
DataIn.sf <- st_as_sf(DataIn,coords=c( "X","Y"),crs="EPSG:32617")

names(DataIn)

 [1] "Serial"                  "X"                      
 [3] "Y"                       "Elevation_1m"           
 [5] "Slope_1m"                "Roughness_1m"           
 [7] "Aspect_1m"               "LandCover_Code"         
 [9] "NCLD_Percent_Tree_Cover" "NCLD_Percent_Concrete"  
[11] "Date"                    "Hour"                   
[13] "Temp_F"                  "Light_Lux"              
[15] "Notes"                  

6. Open the data and take a look at it. (click on its name or put View(DataIn) IN THE CONSOLE
   
   You should see the following columns

   • Serial: the serial number of the item

   • X and Y. The location of the item in X and Y metre coordinates (Long/Lat coordinates are poor over small areas, so I’m using the WGS84 17N / Pennsylvania North map projection (https://epsg.io/32617))

   • Elevation_1m : The elevation of the sensor to the closest metre from the USGS 1 meter Digital Elevation Model

   • Slope_1m : The slope of the location where the sensor was placed, from the same USGS DEM

   • Aspect_1m : The aspect of the location where the sensor was placed., from the same USGS DEM

   • Roughness_1m : The roughness of the location where the sensor was placed, from the same USGS DEM

   • LandCover_Code: The NCLD landcover code for that location

   • NCLD_Percent_Tree_Cover: The percent tree cover for that location

   • NCLD_Percent_Concrete : The percent concrete at that location

   • Date: Date of the measurement

   • Hour: Hour of the measurement

   • Temp_F: The temperature measurement

   • Light_Lux: The Light measurement

   • Notes: Placement Notes

GROUP challenges

Now, there are several challenges that you can split as a group depending on your strengths.

• Describing the data and the meta data. This means

  • Explaining the unit of analysis, the population and the variable

  • Researching and writing up why they have been “transformed” into UTM (hint: CLICK HERE)

  • Researching and writing about each of the predictor variables (e.g. a few sentences about what it is etc, google and the links above should help, or ask me questions on Teams)
    

• Making summary statistics and plots of the DataIn table. For example

  • How many sensors are there (hint, table command)

  • Any missing data?

  • Filtering the data to just one sensor, or one time of day (hint, this will really help: https://crd150.github.io/lab2.html#Data_Wrangling) and making a plot of the temperature over time, or say a histogram of the temps on a single day.

  • Filtering the data in other ways and making exploratory plots and histograms.
    

• Quality controlling the data. For example,

  • Can you trust all the readings at all times of day? Can you see examples of where there are issues.

  • What about over all dates? Anything weird happening?

  • If you find things, decide whether you should keep or remove them AND WRITE DOWN WHAT YOU DID/CHANGED.
    

• Using DataIn.sf Make some initial maps, especially of static variables that don’t change over time.

• You could use QTM from the previous lab, for example

tmap_mode("view")

qtm(DataIn.sf,
            dots.col="NCLD_Percent_Tree_Cover")+
           tm_layout(legend.outside = TRUE)

• Or better still.. follow these instructions (https://psu-spatial.github.io/Geog364-2021/364Data_TutorialWranglePoint.html) or https://r-tmap.github.io/tmap-book/nutshell.html to make prettier ones.
  
• Explore the data.
  
• Finally, choose one different predictor variable each - and make a scatterplot to look at the impact on temperatures and a basic linear model.
  
• Critique your model so far, especially seeing if it visually fits the LINE assumptions.
  How can we do better? Feel free to try other options and Lab 3 will help..

Combine your work into an exploratory data analysis part of your report.