Weather data: precipitation, cloud cover, humidity
We post a short extract from this student project to show the types of analyses you can do with weather data. This candidate regresses precipitation (rainfall and snowfall) on cloud cover and humidity.
The candidate forms a scatter-plot to see if a regression equation is reasonable. This is good practice; the graph tells you if the two variables are linearly related or if a transformation of one of the variables is appropriate.
Look at the scatter-plot of precipitation and humidity.
For humidity levels of 20% to 65%, precipitation is zero or close to zero.
For humidity levels of 80% to 95%, precipitation is high. All seven points with the highest precipitation levels fall in this humidity range.
The two variables are related, but the relation is not linear. For a linear regression:
Use a transformation of the variables to make the relation linear.
Divide the data points into two sets.
For your student project, you might try the following:
Divide the points into humidity less than 65% and more than or equal to 65%.
For humidity less than 65%, precipitation
. 0.003 inches, with no significant relation to humidity.For humidity more than or equal to 65%, precipitation is strongly related to humidity.
You might compare linear regression, polynomial regression, and logit regression. If H = humidity, the independent variable may be H, H2, or H/(1–H), or a combination of these. See which regression equation fits best.
Many days have no rainfall or snowfall. For the dependent variable, you might use the average precipitation for a given humidity percentage. Use a weighted regression, where the weights are the number of days with that humidity level.
These adjustments give higher R2 results and a better relation of precipitation to the other variables.
Many student projects can be done with weather data. Rainfall occurs when a cold front hits a warm front. A good indicator of rainfall is the change in the temperature. For a student project, regress precipitation of the absolute value of the change in the daily temperature from the previous day to the current day.