To begin the infamous Dashboard Week here at the DS, our first task was to create a dashboard which analyses the changes in ocean temperature in the Gulf of Mexico.

My Approach

I began with data prep in Alteryx. Thankfully, this wasn't too complicated today. It merely involved a combining a date column with a time column and then a simple DateTime conversion.

I also added the spatial tool 'create points' in for the latitude and longitude columns in order to leverage Tableau's spatial/mapping ability, which became useful later in the day. My Alteryx workflow can be seen below:

Upon bringing the new data into Tableau, it became clear that the data set was incredibly large and difficult for Tableau (and Alteryx for that matter) to process. As a result, I decided to add a data source filter to only include values from the years 1994 - 2014. This gave me a large enough sample size to draw insights from, but without Tableau grinding to a halt after every query.

The next step I took was to group the several different organisations that were collecting the data into smaller jurisdiction subsets. These can be seen in the sparkline charts at the top of the dashboard, which aim to compare the ocean temperatures over time against the overall average of the (filtered) data set.

I then created a map from the spatial points created in Alteryx, and a jitter plot to highlight any clustering of higher temperatures in specific locations and also across different years.

I felt as if the data set provided didn't quite provide enough context to draw any significant analysis from, so I added some supplementary data regarding storms in the gulf of Mexico.

Whilst I am fairly pleased with the appearance of the dashboard, there was definitely some room from improvement...

Where could I have improved?

• I should have been more wary of outliers in the data set and used median as the default aggregation of temperature instead of average. This would have avoided the data becoming skewed by extreme outliers.
• I also should have noticed that 2014 data was only half complete, and as a result, only included winter temperature readings. Again, this skewed my findings and as a consequence, I was not able to draw the valuable insights that I was hoping to.

The Dashboard

A link to Tableau Public can be found here...