Algorithms & Frameworks

Production-ready methods that power inversions, data fusion, and interactive products.

Linear Algebra Sparse Ops Geostatistics Agent-Based Bayesian Inversions

O(n2.5) Kronecker Methods for Large Inverse Problems

GMD, 2013

Reduces dominant multiplications by exploiting Kronecker structure in state/observation operators, enabling materially larger state vectors and faster UQ on commodity hardware. Demonstrated on continental-scale CO₂ flux estimation with practical implementation guidance.

Kronecker productsUQScaling

Sparse×Sparse Covariance Multiplication for Inversions

GMD Discuss., 2016

Memory-aware, parallel sparse-sparse multiplication tailored to inversion covariances/Jacobians, improving throughput and cache behavior in production pipelines for trace-gas estimation. Preprint details hybrid parallelization and triangular/dense-target variants for HQHᵀ patterns.

OpenMPCovariance opsThroughput

Sign-Aware Jaccard Correlograms & PSD Kernels

Technical note

A bounded, scale-invariant similarity for mixed-sign signals that cleanly separates **phase (sign)** and **magnitude** disagreement. The induced negative-type distance yields valid PSD kernels for kriging/GPs and interpretable maps that distinguish phase vs. amplitude anisotropy.

CorrelogramsVariogramsPSD kernels

Census-Tract Agent-Based Decarbonization Model

Docs

Tract-level ABM simulating household adoption of EVs, efficiency upgrades, and solar under configurable policy instruments (income targeting, stacking, timing). Integrates ACS demographics and grid pathways, reporting equity, cost-effectiveness, and leverage metrics across parallel runs.

Policy designEquity metricsParallel runs

Bayesian / Geostatistical Inversions (Fortran Core)

Code

Fortran-based inversion core implementing covariance construction, H·Q·Hᵀ operators, and parallel drivers for large atmospheric problems. Includes batch/parallel scripts and modular source layout for operational deployment in production pipelines.

Fortran 90HQHᵀ opsParallel