Predicting Rainy Days

[#1]

Hi all! I'm wondering if JMP can help with spot a pattern in the days leading up to rain.

If I have years of historical daily (even hourly) environmental data which includes temperature, humidity, pollution, etc, would JMP be able to help spot a pattern leading up to rainy days? If so, what type of analysis would be best suited for a study like this? The "sequential" element is tricky to me, as I'd like to find a pattern in the sequence of events leading up to the rainy day.

For example, an ideal finding may be a relationship between a steady rise of temperature and decline of humidity in the 5 days leading up to rainy days, with a sudden temperature decline finally triggering the rain.

Any thoughts would be appreciated! Thank you.

[#2]

Hi,
Considering that it takes the National Weather Service and NOAA thousands of sensors, dozens of satellites, and a few supercomputers to produce the weather forecast everyday, I'm not sure that JMP is exactly up to the task. However, if your inquiry is more about learning about the analysis of time-series analysis, you may find the following post useful:

Pattern recognition in time series 

Time Series Analysis Predicting Future Values 
Let us know if any of this helps.
Best,
TS

[#3]

Thank you so much! This is a push in the right direction for me. And yes, I am just using this as a learning project, I’d love to get more comfortable with this type of analysis.

[#4]

In addition to the thoughtful contribution from @Thierry_S your scenario is tailor made for JMP Pro Functional Data Explorer/Modeling capabilities. Here is link to the JMP Pro documentation for FDE:JMP Pro Functional Data Explorer 

[#5]

Thank you for the recommendation! I'm excited to learn more about this.

[#6]

gudetama,

Just a note, as you have already been provided with some good direction by previous responders, you might start learning about these methods with a simpler situation.  Predicting the weather is very noisy and you might get discouraged when your models don't predict very well.  In any case, have fun in your pursuit.

[#7]

Thank you, I will start with the sample dataset included for the FDE capability in jmp and poke around first. I appreciate the suggestion!

[#8]

FDE is but one method...I can think of about a dozen others that could work too. Starting with good old fashioned ordinary least squares regression to non linear methods such as tree based techniques. I would start with just time series plots of all the variables.