Jan 1, 2026Web Application

Tropical Cyclone Trajectory Forecasting System (LSTM & ONNX)

LSTM-based tropical cyclone trajectory prediction and visualization system evaluating 12 experimental scenarios (mean Haversine error of 14.2 km and R² of 0.999). The optimal model is converted to ONNX format for direct in-browser execution via ONNX Runtime on an interactive map built with Astro Island Architecture.

Role & Responsibilities

Frontend & Research Developer

Release Date

Jan 1, 2026

Discipline Category

Web Application

Verification Status
Production deployed & verified
Evaluasi Model (12 Skenario)Haversine 14,2 km | R² 0,999
Frontend ArchitectureAstro Island Architecture
Browser InferenceONNX Runtime Web
Context & Requirements

System Specifications & Context

Deep learning models for meteorological trajectory forecasting are predominantly restricted to offline computing environments, creating a barrier to interactive spatial visualization and accessible browser-based simulation.

Engineering Decisions

Architecture & Implementation

Evaluated 12 empirical LSTM scenarios (achieving a 14.2 km mean Haversine error and 0.999 R²) and converted the optimal model to ONNX format, enabling direct client-side inference in the browser via ONNX Runtime within an Astro Island Architecture geospatial interface.

Technology Stack & Toolchain

AstroTypeScriptONNX RuntimeLSTMGeospatial UIIsland ArchitectureServerlessTailwind CSS

Additional Engineering Projects

Explore further frontend and web application case studies.

View all