Toolkit for Extremes Climate Analysis (TECA)
Overview
TECA is a general purpose tool for detecting discrete events in climate model output. It leverages a map-reduce framework for efficient parallelization at large scales (order 10K+ cores). Currently, TECA contains detection algorithms for tropical cyclones, atmospheric rivers, and extratropical cyclones; and plans are underway to implement algorithms for mesoscale convective complexes, African Easterly waves, atmospheric blocks, and fronts.
Technical Description
TECA has two main interfaces aimed at two distinct use-cases: use of ‘canned’ algorithms (TC, AR, ETC, etc.), and development of user-defined algorithms. The TECA code base is written in modern C++ (i.e., using C++ 2011 standards), and contains a Python API.
The first interface is a command-line interface, with command-line programs for each implemented detection algorithm. The command-line interfaces take file paths as input and contains a number of arguments for customization (e.g., setting of detection thresholds, setting output files). A typical parallel run on NERSC systems looks like the following:
srun -n 29200 teca_tc_detect --input_regex ${data_dir}/${files_regex} \
--candidate_file candidates_1990s.bin \
--track_file tracks_1990s.bin
The second interface leverages the Python API to allow non-C++-proficient developers to easily prototype parallel performant algorithms using a widely adopted language in the scientific community. Currently, no canned algorithms are developed with the Python interface (though a few collaborators are using this facility), but efforts are currently underway to implement several different atmospheric river detection algorithms using the Python interface: both for scientific use and to highlight the flexibility of the TECA pipeline when integrated into Python. While the Python API method is clearly more complicated from a user’s point of view, it offers the option to design custom workflows. The example at the end shows a simple TC-detector implemented with the Python API.
Quick links: Repository, Installation, Tutorials, Documentation, CAM Results
Contact: William D. Collins (wdcollins@lbl.gov)