Welcome to Threshold Enhancer Web Page
Threshold enhancer (or ThEnhancer) is a program writen by Adrian Jacobo at IFISC (previously a part of IMEDEA) in colaboration with Emilio Hernández-García and Pere Colet, in the context of the THRESHOLDS Integrated Project. It is a tool to process time series which are thought to show an ecological regime shift (Scheffer et al., 2001; Muradian, 2001; Scheffer et al., 2003) to obtain a new one in which the possible jump is greatly enhanced and clarified. An estimation of the jump location is provided. ThEnhancer uses the data provided as the forcing term in a nonlinear diffusion equation, which is then integrated to enhance the contrast in the data set and clean the noise in it. In addition to time series, other kinds of univariate data sets can be equally analized.
The software is written in Pyhton and it can be downloaded as an standalone application for both Windows® and Mac®. If requested the Python scripts can be delivered to use the software under Linux.
The data can be opened from an Excel® file or from a text file with several columns separated by tabs, comas or spaces (the decimal separator should be the point). Once the data set is read, a number of operations can be done, and the user can control 3 different parameters:
- Smoothing: When high, fast variations in the data are smoothed out.
- Reference Level: Has to be set to a value between the mean values of the data before and after the jump. It can be scanned to look for different jumps in the data, between different values.
- Data Strength: It can be used to control how much of the detail of the input data is transfered to the final output. Its effect is not as obvious as the previous two parameters but it can be used to fine tune the filtering.
The program output, i.e. the processed time series, can be stored in two different ways: as a text file of two columns tab-separated or as a plot in a postscript file.
The program is an exploratory tool which does not perform any statistical testing. It suggest a location which is suggestive of the occurrence of a jump in the time series, but robustness of the finding against change in the program parameters, and proper statistical testing, are recommended.