Computer Methods in Meteorological Forecasting and Analysis

Spring Semester 2015

Description. 

This course explores the computationally intensive statistical methods used in the development of automated weather analysis and forecasting systems. The focus of the course is on learning to develop and use artificially intelligent automated systems to perform data quality control, quantitative analysis of large meteorological data sets, and weather forecasting.

Instructor:

Dr. Jon Nese, 518 Walker, 863-4076, Twitter: @jmnese

Office Hours:

Mon 4-5 PM, Thu 1-2 PM, Fri 8-9 AM, and by appointment

Course meets:

Wed & Fri  1:15 – 3:20 PM, 126 Walker

Prerequisites:

STAT 301 or STAT 401 or EBF 472

Teaching Assistant:

Eric Wendoloski, Office:  623 Walker, Office Hours:  Tue 1-2PM, Wed 11-12PM 

Course Objectives: 

• Learn computer methods that can be used for statistical analysis and forecasting of weather, climate and other environmental phenomena.
• Learn to apply those methods to develop accurate and robust forecast systems.

Course Outcomes: 

By the time this class if over, you should be able to:

• conduct statistical verification of a weather or climate forecast

• demonstrate knowledge of methods for statistical prediction of atmospheric phenomena

• select, implement and evaluation appropriate statistical forecasting methods

• alter forecast metrics to account for various measures of forecast performance

Motivation: 

Advances in statistical forecasting techniques are making automated prediction systems ever more competitive with the human forecaster, enabled by statistical methods of analysis that are increasingly accessible. These statistical methods, including artificial intelligence systems, can evaluate situations where multiple variables contribute via multiple possible mechanisms. To succeed in this changing environment, the operational meteorologist must become a builder of forecast systems who uses meteorological insight and statistical knowledge to create ever more accurate forecast systems. Meteorology 474 provides hands-on experience building and testing such statistical forecast systems.

Materials

I often place lecture materials on ANGEL prior to class. These lecture notes and information gleaned from the assignments will be the main resources for this course.  Some course materials will be taken from the text Data Mining: Practical Machine Learning Tools and Techniques, which is on reserve in the EMS Library.  In addition, a software package called WEKA, available on the department’s computing cluster, will be used.  Here are some Web Resources for WEKA

• Overview - http://www.cs.waikato.ac.nz/~ml/
• Download - http://www.cs.waikato.ac.nz/~ml/weka/index.html
• Tutorials - http://www.cs.waikato.ac.nz/~ml/weka/index.html (under documentation)
• File Formats - http://www.cs.waikato.ac.nz/~ml/weka/arff.html

Enrollment policy. 

Students who do not meet the prerequisites may be dis-enrolled during the first 10-day free add-drop period after being informed in writing by the instructor (see:  http://www.psu.edu/dept/oue/aappm/C-5.html).  If you have not completed the listed prerequisites, then consult with the instructor.  Students who re-enroll after being dis-enrolled according to this policy are in violation of the Student Code of Conduct.

Assessment Tools and Grading.  Assignments will focus on projects that illustrate the creation and testing of forecast systems using machine-learning algorithms.  Assignments must be submitted ON TIME – there will be a 25% penalty for any assignment handed in late, and no credit will be given for assignments handed in more than 24 hours after the deadline.  There will be a midterm (in class, date to be announced) and final exam (scheduled by the university) as well as very short quizzes every other Friday, starting on January 23 (so there will likely be a total of 7 quizzes – there will be no make-up quizzes, but I’ll drop your lowest quiz grade).  In addition, participation and attendance will factor into your overall course grade.  The weighting of the components of your course grade is as follows:

• Midterm: 20%
• Final Exam: 25%
• Quizzes: 10%
• Assignments: 40%
• Participation / Attendance: 5%

A standard grading scale will apply to the course, shown below:

• A 90-100
• B 80-90           
• C 70-80             
• D 60-70           
• F <60   

However, I will use the ‘-‘ and ‘+’ system as well.  For example, a grade just below 90 may receive a B+.  Also, I may curve the grades, so it is possible that the thresholds will go down (for example, the boundary between and A and a B may be less than 90).

General Course Outline (modifications possible, as needed):

Introduction

• Go Over Course Syllabus
• Statistical Review
• Components of an Automated Analysis/Forecasting System

Forecast Verification

• Purpose of Verification
  
• Statistical Basis of Verification
  
• Verification Techniques
  
• Forecast System Robustness
  

Decision Trees

• Introduction and Applications
  
• CART Algorithm for Growing a Decision Tree
  
• Pruning a Decision Tree to Prevent Overfitting
  
• Data Quality Control
  

Neural Nets

• Introduction and Applications
• Weighting, Objective Functions, and Iterative Improvement
• Training via Back Propagation
• Learning Rate versus Convergence
• Overfitting
• Model Development – Varying the Design
• Verification – Impact of Design Effects

Value of Forecasts

• Valuation of Forecasts
  
• Value of Probability Forecasts
  

Academic Integrity
This course adopts the Academic Integrity Policy of the EMS College. Briefly, students are expected to do their own problem sets and to work the exams on their own. Class members may work on the problem sets in groups, but then each student must write up his or her answers separately. Students may not copy problem set or exam answers from another person's paper and present them as their own even if you worked together to figure out how to solve the problem. Students who present other people's work as their own, as well as the students providing the answers, will receive at least a 0 on the assignment and may well receive an F in the course.

Every so often, we see that on the homework one or more students have copied the files of another student. This is easy to spot, and deal with.

Please refer to the Academic Integrity and Research Ethics page on the EMS website for details of University and College policies.

Students with Learning Disabilities
Penn State welcomes students with disabilities into the University's educational programs. Every Penn State campus has an office for students with disabilities. The Office for Disability Services (ODS) provides contact information for every Penn State campus. For further information, please visit the Office for Disability Services website.

In order to receive consideration for reasonable accommodations, you must contact the appropriate disability services office at the campus where you are officially enrolled, participate in an intake interview, and provide documentation based on the Educational Equity Documentation Guidelines. The documentation supports your request for reasonable accommodations, your campus’s disability services office will provide you with an accommodation letter. Please share this letter with your instructors and discuss the accommodations with them as early in your courses as possible. You must follow this process for every semester that you request accommodations.

Attendance. 

This course abides by the Penn State Class Attendance policy. Informally, I assume you’re in class unless you have a really good excuse to miss, and you’re responsible for all that you miss.

Cancellations and delays. 

Campus emergencies, including weather delays, are announced on Penn State Live and communicated to cellphones, email, the Penn State Facebook page, and Twitter via PSUTXT (to sign up, please see http://live.psu.edu/psutxt).