Predicting the Future: Temporal Flood Models, LAI Hurdles, and World Town Planning
Welcome back to another weekly update from my internship at the KMUTT Geospatial Engineering and Innovation Center (KGEO). This week (November 2nd to 6th) was a deep dive into the realities of machine learning research. It was a week filled with major structural pivots in my code, navigating the unpredictable nature of new dataset features, and stepping out of the lab to attend a major national conference.
Rethinking Time: The Flood Risk Pivot
Earlier in the week, I had a breakthrough realization regarding the Nakhon Pathom and Sukhothai flood risk project. I completely restructured the workflow of the predictive model. Instead of just looking for static relationships between different environmental features, I shifted the model's focus to temporal pattern recognition.
I engineered the system to analyze historical data from the past 1 day, 3 days, and 7 days. By teaching the model to recognize these rolling historical patterns, it can now successfully predict flood risks 1, 3, or 7 days into the future. By Wednesday, this new temporal workflow was finalized and working beautifully.
The LAI Rollercoaster in Biomass Estimation
While the flood project went smoothly, my Above Ground Biomass (AGB) estimation project reminded me that research is never a straight line. I started the week executing my plan to integrate the Leaf Area Index (LAI) as a new feature. I even ran a training sequence using only LAI to isolate and observe its specific impact on the model.
Initially, adding LAI seemed to boost the model's overall accuracy. However, as I finalized the code and dug deeper into the results by Thursday, I realized things were not going perfectly. Introducing this complex new variable created some unexpected hurdles in the data behavior. It was a great lesson in trial and error: a new feature doesn't always mean a magical fix. My next step is to deeply analyze these anomalies and draft a comprehensive report on the LAI integration.
Progress Updates and Real-World Impact
Mid-week, I had a major progress meeting with Ajarn Peri, Ajarn Yota, and the rest of the KGEO staff. I presented the finalized temporal flood risk model and the updated AGB estimation workflow.
The presentation was a success, and I received fantastic feedback and impressions from the entire team.The most exciting takeaway from this meeting was our next action plan: we are going to integrate these machine learning models directly into a live website. Taking this research out of the terminal and putting it onto a web platform means our work will have a tangible, real-world impact for the people who need it most.
World Town Planning 2026 in Suphan Buri
Friday was an absolute highlight. I traveled with Ajarn Peri, a couple of statistics students, and two high school students to Suphan Buri to attend the World Town Planning 2026 event.
The conference was incredibly insightful, focusing on how smart town planning can build better, more resilient cities, especially in the face of seasonal disasters. Ajarn Peri even took the stage to speak about the critical importance of GIS applications in disaster management.
Afterward, we went on a fantastic road trip around Suphan Buri, eventually passing through Nakhon Pathom. It was an exciting, inspiring way to close out a heavy week of coding and research!









Comments
Post a Comment