Now in its 33rd year of development, Clark Labs continues to improve and extend the TerrSet suite of geospatial monitoring and modeling tools. Key developments include:
IDRISI – Remote Sensing
- SVM – a new classifier based on a Support Vector Machine logic. Both soft and hard classification is supported. In addition, an automated tool is offered for parameterization – a “grid search” procedure that tests varying values of parameters using cross-validation. Linear, radial basis function (RBF), polynomial and sigmoid kernels are supported. The user also has the ability to force SVM to used balanced samples (equal numbers for each class) and can evaluate the importance of variables.
- DECISIONFOREST – an implementation of the random forest algorithm for classification. In addition to the hard classification, probabilistic soft classification images can also be output. DECISIONFOREST models can also be saved and used to classify novel imagery. Thus models can be trained in one location, or time, and then applied in a different location or time so long as the same variables (bands) are used.
- CONSENSUS – a module that outputs the most common class assignment from a group of classifications. This is useful for the creation of ensemble classifications.
- A new PANSHARPEN module that offers both an Intensity-Hue-Saturation (HIS) model and a Hyperspherical Color Sharpening (HCS) model.
- SUBSAMPLE – a procedure for sub-sampling training data, including the creation of balanced samples where each class has an equal sized sample.
- AREAFILTER – a post-classification cleanup procedure that enforces a specified minimum mapping unit (MMU) by removing features (contiguous groups of pixels) smaller than the specified threshold.
- DIGITIZE – on-screen digitizing has been greatly streamlined to facilitate projects where a large number of training sites need to be developed. It also provides the option to monitor the areas of training sites. This is especially useful when trying to achieve balanced samples for machine learning.
- Post-classification image editing options have also been added to the DIGITIZE menu. These facilitate the manual cleanup of classification by direct digitization, or by using the intersection of a digitized feature and a mask or masked class.
- LANDSAT_UPSAMPLE – a module for the upsampling of LANDSAT 7 and 8 bands where pansharpening is not possible. The procedure is designed to honor the original values as closely as possible.
- SOILSALINITY – a module that creates indices to soil salinity based on reflectance in the blue, green, red and near-infrared bands. In total, 10 indices are supported.
- WATERINDEX – a module that creates indices that are sensitive to the presence of water based on reflectance data.
IDRISI – GIS
- LOGIC – a new module that provides the ability to perform logic operations between images – e.g., GT (>), GTE (>=), LT (<), LTE (<=), AND, OR, XOR, NOT, EQ (=), NEQ (<>)
- REGRESS adds options to output predicted and residual images
- KNNREGRESS – a non-linear, non-parametric regression procedure based on a K-Nearest Neighbor logic.
- INTERPOL (also called IDW) – inverse distance weighting interpolation modified to provide more control over the number of control points used in the interpolation.
- MCE – The Multi-Criteria-Evaluation procedure now adds automatic sensitivity analysis to gauge the sensitivity of factors to changes in the factor weights.
- POLY2LINE – converts polygons files to line files.
- PROJECT adds support for Cylindrical Equal Area, and has improved speed with some options.
- Database Workshop now supports tab delimited files as well as import of .xlsx files (in addition to .xls files). It also provides the ability to handle .csv and .txt files with a header row.
LCM
The Land Change Modeler has undergone a major revision:
- LCM includes three new empirical modeling tools – Support Vector Machine (SVM), Decision Forest (an implementation of Random Forest) and Weighted Normalized Likelihood (WNL). The latter is capable of modeling large numbers of transitions in a rapid time frame.
- We have also added the option to use a balanced sample with Logistic Regression, yielding transition potentials more in line with those produced by procedures such as LCM’s Multi-Layer Perceptron (MLP).
- Revised to allow the automatic modeling of multiple transitions for all procedures, with the exception of SimWeight.
- Adds the ability to quickly add all transitions to a single group, or to ungroup them and use automatic transition names.
- Greatly improved control of project flow.
Other Improvements
- COMPOSER – now adds PNG output of compositions with lossless compression. In addition, COMPOSER offers the ability to save compositions at resolutions higher than screen resolution.
- MACRO MODELER – added logic operations GT (>), GTE (>=), LT (<), LTE (<=), AND, OR, XOR, NOT, EQ (=), NEQ (<>).
- A wide variety of hotkeys have been added to facilitate user interaction. In addition, the user has the ability the customize the hotkey assignments.
- GDAL conversion utility – adds batch import, and has been updated to GDAL Version 2.4.1
- In addition to the above, many improvements have been made to the speed of existing modules through a combination of 64 bit versions (TerrSet automatically switches depending on your operating system), algorithm changes and multithreading.
- Code signing certificates have been added to every module to enhance system security.
TerrSet Explorer
Facilitates the display of color composites made from separate bands. Select any three bands, right click and choose the option to display as a color composite. Now supports deletion to the recycle bin (default) or permanent deletion (shift-delete)