An AI-Powered Platform for Quantifying Scientific Impact Across JPL Earth System Models and Missions

About this technology

Overview

An integrated platform combining four innovations: (1) Automated citation analytics using large language models (Gemini, Gemma3) to classify engagement levels (4 tiers: background citation foundational method usage), research domains (10 Earth science categories), and geographic reach across 26,741 peer-reviewed publications spanning 8 JPL Earth system models (ECCO, ISSM, RAPID, CMS-Flux, CARDAMOM, MOMO-CHEM, LES, EDMF) and 2 NASA observation missions (GRACE, SWOT). (2) Multi-agent data verification with five specialized agents (Team Paper Categorizer, Crossref Agent, Semantic Scholar Agent, Keyword Classifier, Deduplication Agent) achieving 91-96% keyword relevance match. (3) Three-phase Bayesian uncertainty quantification pipeline deterministic scoring, multi-temperature LLM ensemble sampling (3 passes at temperatures 0.1, 0.5, 1.0), and adversarial skeptic review providing calibrated confidence scores for every classification. (4) Cross-model network analysis identifying bridge papers, co-authorship networks, and domain overlap across all models. The platform is implemented as a React-based interactive web dashboard with D3.js/TopoJSON geographic visualizations and Recharts charting.