Bug Fixing, FIBO Collaborations, and Deploying Robust Flood Models

Welcome back to another weekly update from my internship at the KMUTT Geospatial Engineering and Innovation Center (KGEO). This past week (November 9th to 13th) was a perfect blend of polishing web applications, handing over technical infrastructure, refining machine learning models, and opening doors to exciting new collaborations!

WefGIS Polish and New FIBO Collaboration Horizons


The week kicked off on Monday with a bit of multi-device testing. When I opened the WefGIS platform on my tablet, I noticed a few pesky UI bugs on both the landing page and the core app. I immediately rolled up my sleeves, fixed the issues, and reported the updates to the Head of the TRPL study program back at Undiksha.

Following that, my mentor and lecturer from Undiksha brought some thrilling news: a new collaborative opportunity is on the horizon with FIBO (Institute of Field Robotics) at KMUTT. This is a massive step for both KGEO and Undiksha, opening up possibilities to expand our projects and create even more impactful cross-institutional programs. We spent time brainstorming what kind of joint projects we can bring to life.

Finalizing Mangrove AGB Code and Server Documentation


On Tuesday, I reached a major milestone on the environmental side: I officially wrapped up the mangrove biomass project with the Leaf Area Index (LAI) integration. I compiled the final GeoTIFF results alongside the full Jupyter Notebook source code and submitted everything to Pi Boy for his thorough review.

Mid-week, I shifted gears back to IT infrastructure. To ensure the 5TB KGEO cloud storage server I built earlier remains sustainable, I spent Wednesday writing comprehensive technical documentation. It covers everything from initial architecture and system design to a step-by-step troubleshooting guide, ensuring KGEO staff can easily manage the server if any issues arise in the future.

Organizing Directories and River Body Analysis

Thursday was all about housekeeping and looking ahead.I spent time neatly organizing all project directories for my work at KGEO to keep everything clean and reproducible.

With the data organized, I started exploring a brand-new research angle: river body analysis and its direct correlation with flood risk. I also revisited my flood risk dataset to test the performance differences when including versus excluding station data, ensuring our training features are as optimized as possible.

Squashing Deployment Anomalies for T+3 and T+7 Predictions

Friday brought a classic engineering puzzle. While testing the deployed flood risk model, I noticed a logic anomaly: because I hadn't configured separate models for different time horizons yet, the system was predicting T+3 and T+7 using the exact same output logic as T+1.

Since short-term and medium-term weather patterns behave differently, this was a crucial fix. I set up dedicated model structures for T+3 and T+7 predictions. The binary flood risk classification (yes or no) is now working remarkably well. With the technical pipeline stabilized, our next major step is rolling out user testing to gather real feedback and make the flood risk platform as intuitive and impactful as possible.

Stay tuned as we get ready to put these models into the hands of real users!

Comments