Week 2: A Lesson on History

Dev K -

Welcome to the second-week blog entry for my Senior Research Project!

This week, I’ve explored the evolution of weather prediction, from early statistical methods to modern machine learning. I looked into how traditional statistical models like Geographically Weighted Regression (GWR) and Spatio-Temporal Kriging (ST-Kriging) have worked and been used in weather prediction. Related to that, I also looked into the growing use of Machine Learning modeling has been implemented in forecasting systems.

Next week, I’ll be focusing on a comparative analysis of ML-based and traditional physics-based weather models, evaluating their respective strengths, weaknesses, and the critical role of high-performance computing in improving forecasting accuracy and efficiency.

Until next time,

Dev KC

More Posts

Comments:

All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.

    dea_k
    This is interesting, Dev. Are the different models better at predicting certain things or more suited for certain regions than others?
    February 22, 2025 at 2:13 pm - Reply
    deanna_b
    Wow, I never knew there was so much that goes into weather forecasting! Could you explain a bit more about GWR and ST-Kriging, and how they're used?
    February 22, 2025 at 10:31 pm - Reply
    czarina_p_s
    Wow, Dev! This is so intriguing! Do you have a preference for which model you've use?
    February 24, 2025 at 9:11 am - Reply
    adampongratz
    Dev, why is machine learning becoming more popular in a weather prediction application?
    February 24, 2025 at 5:10 pm - Reply
    dev_k
    Hey everyone, sorry for the late response! I really appreciate your questions and enthusiasm. Adam and Czarina's questions are rather similar so I will describe them together. For Adam, the reason why machine learning is becoming popular in weather forecasting is because it can process vast amounts of data quickly and identify complex patterns that traditional methods might miss. In addition, ML models can learn directly from data, making them particularly effective in regions with sparse observations or for predicting short-term weather events with high accuracy. Now despite this, I would still prefer to utilize more traditional means of forecasting, as ML models do have a tendency of not being able to predict long-term results well along with the fact that the exact interpretation of said results is hard to do due to the randomized nature of it's application. Great question Deanna! Geographically Weighted Regression (GWR) is a statistical technique that aids in predicting the weather by using spatial variability in relationships between various weather variables. In weather forecasting, it helps account for local variations, such as how temperature patterns shift across different terrains. Spatio-Temporal Kriging (ST-Kriging), on the other hand, is an interpolation method that predicts unknown values based on both space and time correlations. It's useful in filling gaps in weather data, such as estimating rainfall in areas without direct measurements. As for Dea, this is absolutely true. Different models excel at different tasks. Traditional physics-based models are still the best at large-scale, long-term forecasts because they rely on well-established atmospheric physics. However, machine learning models tend to outperform them in short-term predictions.
    February 26, 2025 at 3:50 pm - Reply

Leave a Reply

Your email address will not be published. Required fields are marked *