Welcome back to my internship journey at the KMUTT Geospatial Engineering and Innovation Center! My second week here in Bangkok was packed with rigorous simulations, algorithmic comparisons, and some valuable lessons in both geospatial science and cross-cultural communication.
Here is a look at what I worked on this week.
September 7, 2026: GIS Class and Refining Communication
I kicked off the week by joining a GIS class with Ajarn Parivate. It was a great opportunity to present some of my recent work and share a progress update. Aside from the class, I spent time exploring biomass and mangrove datasets to prepare for upcoming code simulations. Fortunately, the day went smoothly without any major technical hiccups. My biggest takeaway today wasn't just technical. It was learning how to present my work more efficiently and communicate complex progress properly.
September 8, 2026: The Reality of Biomass Mapping & Language Barriers
Today involved deeper explorations into mangrove biomass mapping across several provinces in Thailand. I quickly realized that this is a highly complex research area. It requires contributions from multiple sectors and a massive amount of physical effort to measure trees in the field. To get accurate ground truth data and build reliable models, that on-the-ground effort is absolutely mandatory.
Later, I ran a full simulation of my workflow, from creating datasets to gathering satellite imagery to training the models. Structurally, the pipeline performed beautifully, but it really highlighted the need for high-quality, real-world data.
On a lighter note, I am still navigating the language barrier! Pi Boy’s English accent is getting easier for me to understand, but the local Thai accent can still be a bit challenging to catch sometimes. It is all part of the fun, though. Next up: balancing the biomass research with my ongoing flood mapping project.
September 9, 2026: End-to-End Biomass Mapping Simulation
Today was all about executing a fixed simulation for the biomass research. I built a comprehensive workflow testing environment:
Dataset Creation: I generated a dummy dataset to simulate ground truth mangrove data across our Area of Interest (AOI). I then integrated real satellite features, specifically extracting VV and VH backscatter indices from Sentinel-1 alongside Sentinel-2 optical data.
Model Training: I trained two Machine Learning models: Random Forest and XGBoost.
Because the ground truth data was randomly generated (dummy data), the models understandably didn't perform very well in terms of accuracy. However, today taught me a crucial engineering lesson: running simulations with dummy data is an absolute must to validate if a pipeline is structurally sound before deploying it with real, hard-earned datasets. My next step is to dive into more academic papers to refine the methodology.
September 10, 2026: The 5-Algorithm Showdown
I took the modeling a step further today by running a comparative analysis across five different Machine Learning algorithms. The goal was to find the absolute most optimal model for mapping and analyzing Above-Ground Biomass (AGB). It was a great exercise in optimization, and it reinforced the idea that aggressively comparing tools and algorithms is the best way to guarantee you are using the right mathematical approach for the job.
With the model architecture looking solid, I plan to shift some focus back to the flood project tomorrow.
September 11, 2026: Back to the Basics of Remote Sensing
To wrap up the week, I decided to take a step back and strengthen my foundational knowledge. I spent the day doing a deep dive into the core mechanics of Remote Sensing. I was exploring exactly how the physics of the sensors work and looking into broader real-world applications. Strengthening this baseline is going to be incredibly helpful for my future research.
Now, it is time to organize all these findings, prepare for my next progress update presentation, and get ready for the actual data integration in the biomass research next week!
Comments
Post a Comment