IDRISI: The Kilimanjaro Edition

The 14th release of the IDRISI geoanalytic and image processing system extends the current analytical range while providing major enhancements in cartographic display.

Features of this upgrade include:


Display System Enhancements

Figure 1. A map composition created using the new cartographic display features, including layer blending and transparency, as well as the new image classification and advanced symbol selection tools.

We have implemented a complete overhaul of the display system - from the unglamorous low-level internal routines to the very evident control dialogs. Figure 1 illustrates many of these new features:

  • Enhanced Cartographic Symbolization - Now choose immediate classification of data into equal interval, quantile and standardized ranges. Further, the system now provides advanced symbol file selection. Through simple options for the data type (quantitative, qualitative or uniform) and varying options for the organizational character of the data (e.g., unipolar, bipolar, balance), this simple utility provides direct access to over 1300 symbol themes. (Figure 3)

  • Layer Blending - Visually merge layers using alpha blending. Figures 1 and 2, for example, illustrate the case of merging a hillshading layer with hypsometric tints. This was achieved by highlighting the DEM layer in Composer and then clicking on the blend button (the left-most of the new buttons indicated in Figure 4).
Figure 2. The stages in developing raster imagery used above. A hillshaded layer is displayed in gray; a DEM is added using the new default quantitative palette and 16 equal interval classes; Composer's blend function combines hillshading and DEM; a mask is added with 0's (displayed as black) in the area of interest; Composer's transparency button allows topography to be seen through the mask; a 60% blend limits the degree of transparency within the masked area.



 
Figure 3. IDRISI Kilimanjaro now includes enhanced cartographic symbolization.


  • Figure 4. The 24-bit composite was generated on the fly by assigning each of three layers the appropriate primary color using the new row of display buttons on Composer. From left to right, these buttons toggle blend, cyan, red, green and blue primary assignments, and transparency.

    Background Transparency - The backgrounds of raster layers can now be made transparent by clicking on the rightmost of the new Composer buttons (Figure 4). Figure 2 shows the effect of transparency in the second to last panel, and transparency and blend combined in the last panel.

  • Interactive RGB Compositing - Independently displayed layers can be designated as the red, green and blue layers of an RGB composite directly within Composer. To do so, simply select the layer in Composer and then click on the appropriate red, green or blue button (Figure 4).

  • Scale-dependent Visibility - Layers can be set to automatically become visible or invisible depending upon the scale.

  • Figure 5. Using the cyan and red layer buttons on Composer, it is easy to set up anaglyphic stereoscopic views from systems such as SPOT, ASTER or IKONOS (seen here).

     

    Figure 6. The new Fly Through module uses OpenGL to provide real-time interactive flight over a DEM.
     

    Anaglyphic 3-D - View stereo images from ASTER, SPOT or IKONOS in 3-D (Figure 5) using red/cyan, red/blue or red/green anaglyphic 3-D glasses (each copy of Kilimanjaro includes a free pair).

  • 3-D Interactive Fly Through - Using the power of OpenGL, Kilimanjaro brings truly interactive 3-D fly-through to the IDRISI system. It's quick, simple and dramatic. Specify a DEM and a drape image and you're ready to fly. The system provides complete control over altitude, orientation and movement. Competing software systems sell this facility alone for over three times the cost of a new license for the entire IDRISI system! (Figure 6)

Interface Enhancements
Another significant development is the introduction of persistent forms. In previous versions, clicking on the OK button of a dialog would cause the form to disappear. Now, it remains open with all its settings until you click the Close button - a great feature when you need to execute a series of similar operations. For those who prefer the older style, you can select either interface type in User Preferences. We have also consolidated a number of modules that were similar in character but separate because of their sequence of release. Thus the six raster/vector conversion modules (POINTRAS, LINERAS, POLYRAS, POINTVEC, LINEVEC, POLYVEC) are now replaced by a single integrated module. Similarly, the many generic import options of PARE, BILIDRIS and BIPIDRIS have been collected together. We will continue to streamline in releases to come.

Figure 7. IDRISI's new database support based on Microsoft's ADO includes special features for streamlining the import and export of raster and vector layer data.

New Database Support
Kilimanjaro also includes a completely revamped database management system based on Microsoft's ADO technology (Figure 7). Built from the ground up as a complete replacement for the previous system, IDRISI's database support is compatible with all versions of Microsoft Access and can easily import and export xBASE, CSV and Microsoft EXCEL formats. Database Workshop can now also connect to any Microsoft OLE database provider (e.g., SQL Server, Oracle, ODBC, OLAP) providing direct support for distributed databases. Multiple tables are now supported within one database with queries made across these relational tables using an Advanced SQL editor.

Direct links can be made between the database and the vector layer and between the vector layer and the database. Of particular significance is the ease with which vector and raster layers can be exported from and imported into the database - simply click the mouse in any column and then select the appropriate button - a highly efficient sequence!

Image Processing
IDRISI continues to develop its high-level support for image classification. Significant new developments include:

    Figure 9. NEURALNET undertakes the classification of remotely sensed imagery through an artificial neural network classifier using the back propagation.
  • NEURALNET - Implements a back propagation neural network classifier. (Figure 9) In cases where class reflectance distributions are not normal, neural network classifiers have been shown to produce superior results to parametric classifiers such as Maximum Likelihood. Although the user has complete control over all parameters such as the number of hidden layers, the learning rate, and the acceptable RMS, extensive work has been undertaken to develop a context-sensitive selection of control parameters. Thus the module is extremely easy to use. All it requires is a set of training sites (created with MAKESIG as usual). Note that the new CCA and PURIFY modules can be used to improve the quality of the training sites supplied to NEURALNET.

  • Figure 8. The new CCA module undertakes a canonical components analysis on a set of image bands specified from input signatures and produces a new set of optimized transformed images.

    CCA - Canonical Components Analysis (CCA) is related to Principal Components Analysis (PCA) in that it is a transformation of the original band achieved by a rotation of axes.(Figure 8) However, the intent here is to orient the axes such that they minimize the within-class variance and maximize the between-class variance for a set of signatures of interest. Thus the resulting images enhance the differences between the classes described by the signature set. CCA has significant value in the context of visual analysis, but can also be an important aid to the process of classification using a neural network.

  • PURIFY - As the name suggests, PURIFY filters training site pixels to remove unrepresentative cases. Two options are provided - parametric and non-parametric. In the case of the former, pixels are removed on the basis of their typicality - a metric derived by integrating the tail probabilities of the Chi Square distribution based on the Mahalanobis distance of the pixel from the mean of its class. Thus, for example, specifying a threshold of 0.01 will remove all pixels that have less than a 0.01 chance of belonging to the training site class. With the non-parametric option, PURIFY performs a cluster analysis within training site classes and removes all clusters that are smaller than a user-defined threshold. The parametric option is designed for use with MAXLIKE while the non-parametric option is designed for use with NEURALNET.

  • Figure 10. The revised CLUSTER modules uses raw imagery for unsupervised classification.

    MAHALCLASS - MAHALCLASS is a new soft classifier that expresses the typicalities of pixels for each class. As with all classifiers in the soft classifier group, MAHALCLASS produces a separate image for each training site class. In this instance, the images express the probability of finding a pixel with a Mahalanobis distance greater than or equal to that of the pixel being evaluated. The Mahalanobis distance is the multivariate equivalent of a z-score. Thus the measure expresses how typical (or atypical) the pixel is of the class in question. The results are thus effective in evaluating the quality of training sites and the presence of unknown classes. Additionally, a hard classification can be produced by submitting the images to the HARDEN module.

  • CLUSTER - The CLUSTER module has undergone a significant revision. (Figure 10) Previously CLUSTER performed a histogram peak cluster analysis based on the information in a three-band color composite. In the new version, you can work with up to seven raw bands. In addition, you have control over the parameters that control the histogram peak procedure.

  • ISOCLUST - The ISOCLUST module has also been revamped. In addition to using the new CLUSTER procedure for seeding clusters, the Iterative Self-Organizing Procedure also incorporates a threshold for cluster weeding.

  • TASSCAP - The TASSCAP module now includes options for working with atmospherically-corrected reflectances and at-satellite reflectances for LANDSAT data as well as the direct Dn transformation previously supported.

GeoAnalysis
Although our emphasis with this release was on the interface and display system, IDRISI has continued to strengthen its base as a geoanalytic system. Our focus to date has been on developing foundation tools (which we pledge to continue), but with this release we have begun to direct our focus towards high level models of significant importance to resource managers and researchers. This release includes the first two of a series of well-established models we intend to implement:

  • Figure 11. The Revised Universal Soil Loss equation (RUSLE) module models soil erosion.

    RUSLE - Soil erosion is one of the most significant environmental problems we face today. Without question, the development of the Universal Soil Loss Equation (USLE) was a milestone in the modeling of soil loss. More recently, the Revised Universal Soil Loss Equation (RUSLE) has been developed by the US Department of Agriculture as the basis for computing annual average soil loss due to sheet and rill erosion. Its widespread utilization reflects the equation's minimal data demands with its success in estimating average, long-term erosion on field units of relative homogeneity. For many parts of the world, the advent of improved and inexpensive DEM generation through softcopy photogrammetry and interference SAR, along with precise land cover mapping using remotely sensed imagery, offers significant opportunities for the spatial delineation of nonchannelized erosion. Where accurate data exist for the RUSLE variables, the module will permit greater accuracy and consistency than current field methods. (Figure 11) Several new modules were developed to implement RUSLE and exist as stand-alone modules in IDRISI. These include SLOPELENGTH which calculates the longest slope length within regions, SEGMENT which produces an image of homogenous slope segments (in gradient and orientation) and GENERALIZATION for raster image generalization routines.

  • Figure 12. GEOMOD is a new predictive land change simulation modeling tool.

    GEOMOD - Building on its development of the CA_MARKOV module for land cover change projection, IDRISI now includes the first Windows implementation of the GEOMOD predictive land change simulation model. (Figure 12) Developed at the State University of New York (SUNY) Syracuse, GEOMOD has been applied throughout the world in a variety of policy-relevant land-use change modeling studies. GEOMOD predicts the locations where land is likely to change from one category to another, for example from forest to non-forest. Specifically, GEOMOD has been the model of choice to analyze the effectiveness of forest conservation projects that have been implemented under international agreements on climate change. The combination of GEOMOD with IDRISI's statistical modules such as ROC and VALIDATE allows the user to measure the certainty of scenarios of future land change. GEOMOD is a significant tool for land cover change modeling, and this implementation represents a major achievement in inter-institutional cooperation.

    Figure 13. The revised VALIDATE module now supports multi-categorical landuse for change analysis.



  • VALIDATE - This version adds a major revision of the VALIDATE module. (Figure 13) VALIDATE is a map comparison tool designed in particular for purposes of model projection validation. In our last release, VALIDATE was introduced with its detailed breakdown of the nature of agreement according to the degree to which it is attributable to the specification of quantity, location or chance. With this release, we extend the analysis to consider scale. A model that is performing less than satisfactorily at the resolution of individual pixels may be doing quite well at another scale. This is important feedback on the scale of the processes operating and the adequacy of the model.

  • Figure 14. IDRISI now includes import/export for SPLUS.
    Figure 15. Enhancements to the RUNOFF module include the addition of a permeability factor for drainage analysis.

    SPLUS - IDRISI now adds SPLUS support as a supplement to its existing support for STATISTICA. (Figure 14) SPLUS is supported in two ways. First, IDRISI offers full import/export of image and values file data. However, we have also created an SPLUS library that works directly within SPLUS. Further we provide the instructions on how to control IDRISI from SPLUS using IDRISI's COM interface.


  • RUNOFF - Another module that has undergone significant revision is RUNOFF - the module that determines the flow pattern over an elevation model. (Figure 15) In the previous version, precipitation was assumed to be uniformly distributed and all surfaces were assumed to be impermeable. In the new version, you can now specify both a precipitation surface and a permeability surface.





 

Import/Export
Since the last full release, we have developed a series of major import/export procedures. Some of these were slipstreamed into a patch release last spring, such as the support for ERDAS Imagine IMG files and full support for HDF. The HDF reader is a major addition, allowing the full ingest of ASTER and MODIS data, as an example. (Figure 16) With this release, we have revised our GEOTIFF support to include the non-standard (but increasingly used) 16-bit and 32-bit formats. (Figure 17) As a result, IDRISI now offers support for all IKONOS and QUICKBIRD data formats. Other new import routines added since the last full release include support for ERMAPPER data and the Argentinean satellite SAC-C.

Figure 16. IDRISI supports HDF-EOS 4 format, useful for importing ASTER data.

Figure 17. The revised GEOTIFF/TIFF module now supports export from IDRISI to GEOTIFF as well as import/export for 8-bit, 16-bit, 24-bit, and 32-bit GEOTIFF formats, such as QUICKBIRD.


Network Compatibility and License Management
Network administrators will be pleased to learn that this version is fully compliant with the new security features of Windows 2000 and Windows XP. Thus registry changes are no longer needed to support users who are not part of the Power Users group. We are also introducing license management with this version of IDRISI. For network administrators, this will allow the setup and management of multiple concurrent clients from a single server - a major improvement in system management. For standalone users, license management will simply require that they register their copy within 7 days of installation.

Figure 18. Macro Modeler incorporates dynamic modeling features along with over 100 modeling components.

Missed Release Two?
If you haven't upgraded from Idrisi32 to Idrisi32 Release Two, the Kilimanjaro Edition is doubly significant. When you purchase Kilimanjaro, you also get all of the developments of Release Two, including:

  • the most extensive graphical modeling environment in the industry, including dynamic modeling. (Figure 18) IDRISI's new Macro Modeler provides a drag-and-drop graphical interface for the programming of analytical sequences, including the ability to create and link submodels that extend system functionality. Using feedback loops, a cellular automata tool and dynamic iteration structures, users have simple access to the full power of dynamic modeling. This of course is additional to IDRISI's fully COM compliant programming interface that allows direct control of IDRISI from programming languages and scripting tools such as Visual Basic, Visual C++, Delphi and Python.
     
  • major enhancements to the system's image processing capabilities, including Linear Spectral Unmixing, Linear Discriminant Analysis (Fisher Classifier), Automatic Mosaicking, Full Atmospheric Correction, AOI (flood polygon) training site delineation, and greatly extended hyperspectral capabilities. (Figures 19-21)
     
  • an extensive set of new change and time series analysis tools, including change vector analysis, temporal correlation, Markov Chain analysis, and a cellular-automata based change projection procedure.
   
Figure 19. Atmospheric Correction.   Figure 20. Automatic Mosaicking.    

 

Figure 21. Probability Guided Linear Spectral Unmixing for sub pixel classification.


Why Kilimanjaro?
Mount Kilimanjaro lies very close to the equator. Only by virtue of its extraordinary height does it carry permanent snow. However, the evidence is clear that the snows of Kilimanjaro have been receding. Is this the effect of global warming or a natural climatic cycle? The debate is passionate and highly political and its resolution is essential to our ability to carry out responsible and sustainable environmental planning. However, to do so, we require innovative and powerful analytical tools. It is towards the development of these tools that we are dedicated.

IDRISI is now in its 18th year of continuous development, and offers the most extensive analytical suite of any software system in the geoanalytical domain, particularly in the areas of decision support, uncertainty management, image processing and change and time series analysis. Built by researchers for researchers, IDRISI is a professional level tool that represents the outcome of one of the most extensive and sustained research and development efforts in the industry, firmly grounded within a non-profit philosophy.

Perhaps most importantly, IDRISI is the tangible outcome of a vigorous exchange of ideas. Intellectual contributions to its development come from a wide spectrum of users, as is clearly evident in our international network of Resource Centers. Please join us in our quest for an affordable, approachable, extensible and innovative platform for responsible environmental management.

 

 

 

reddog@geoafrica.co.za

 Ph: +27-(0)11-467-3371

Mobile: 082 89 2 9771

 26 Crestwood Drive, Lonehill