Segmented Linear Integral Correlation Kernel Ensemble Reconstruction

Project Description

Segmented Linear Integral Correlation Kernel Ensemble Reconstruction (SLICKER) is a numerical method to reconstruct a (typically climate) time-series from multiple proxy data-series. SLICKER allows for

• Univariate or multivariate reconstructions
• Uneven and different data-spacing between allow proxies and the reconstruction target
• Different linear and non-linear relationships between the target and each proxy
• Ensembles to allow for robust estimates of the central tendency, uncertainty in the central tendency, and spread of the reconstruction
• Parallel execution is available via OpenMP

SLICKER is written in FORTRAN, with wrappers for MATLAB, Python and R. A windows executable is also supplied for users without access to a FORTRAN compiler.

Citation

If using SLICKER for scientific studies, please cite

Roberts et al (submitted), Segmented Linear Integral Correlation Kernel Ensemble Reconstruction: A new method for climate reconstructions with applications to Holocene era proxies from an East Antarctic ice core PLOS ONE

Building Source Code

For systems with access to a FORTRAN compiler, source code is included in the ../FORTRAN_source_code directory.

• Edit the "system variables" section of the Makefile to set system specific options, noting that defaults for ifort, gfortran and nvfortran are provided, for both serial and OpenMP parallel configurations.
• Compile the code by typing "make" at the command line, will generate the executable "reconstruct"
• copy "reconstruct" to the "wrappers_and_examples" directory by typing "make install" at the command line

Wrappers and examples

Wrappers and example usage is provided for MATLAB, Python and R. The particular text example for the wrappers is from Figure S2 c of the above PLOS ONE journal paper a test case with two input proxies with 20% of the data missing from both proxies. Note that the solver relies on random numbers, so the solution from subsequent runs of the same problem will be slightly different. Examples for both text and netCDF input data are provided. All other test examples from the PLOS ONE journal paper are included in the Journal_paper_test_cases directory.

Directory	code execution
MATLAB_netcdf_example	matlab -nodisplay test5.m
MATLAB_text_example	matlab -nodisplay test5.m
Python	python reconstruct.py
Python	jupyter lab reconstruct_example.ipynb
R	R --no-save < test5.R