Above: 1.5-layer quasi-geostrophic model on a 193 × 193 grid, potential vorticity (left) and streamfunction (right), one of the benchmark models in PyTEDA-web.
PyTEDA-web is a FastAPI platform built on the TEDA code base. Instead of running scripts locally, you configure a data assimilation experiment in the browser, launch it, and watch the output arrive in real time while it runs on the server.
Every experiment is tracked and persisted, so results from different methods, ensemble sizes, observation networks or models can be compared side by side days later, without rerunning anything.
Besides the Duffing, Lorenz-63 and Lorenz-96 models inherited from TEDA, PyTEDA-web adds larger test cases such as the 1.5-layer quasi-geostrophic model shown above, which brings realistic multi-scale dynamics to the benchmarks at a size that still runs interactively.
All ensemble-based methods available in TEDA can be benchmarked: the stochastic EnKF and its Cholesky and modified Cholesky variants, shrinkage precision estimators, B-localization, EnSRF, ETKF, LETKF and LEnKF.
