[Return to the Software Reviews Index]
IDRISI Kilimanjaro
Timothy Warner, Associate Professor, Department of Geology and Geography, P O
Box 6300, West Virginia University, Morgantown WV 26506-6300, 304-293-5603
x 4328 tim.warner@mail.wvu.edu
David Campagna, Adjunct Professor, Department of Geology and Geography, P O Box 6300, West Virginia University, Morgantown WV 26506-6300
(reprinted from PE&RS June 2004)
Product Information IDRISI Kilimanjaro Vendor Price Educational multi-user license Distribution Medium Program Minimum Requirements
|
Executive Summary IDRISI Kilimanjaro is one of the most widely used integrated remote sensing and raster GIS packages. It is highly affordable, yet provides a wide range of program modules that extend from the basic to highly sophisticated. IDRISI is also a very fast program, comparing favorably to competing commercial packages. IDRISI provides multiple methods for interacting with the individual modules, including a menu of modules, icons for the most common modules, a pick list of programs, a graphical macro modeling interface, and the option to write text-based macros. The Kilimanjaro version of IDRISI adds new display enhancements, including fly through and anaglyphic stereo viewing, as well as new classification, modeling and import/export routines. In addition, several programs have been combined to provide more integrated or interactive functionality. The new RESAMPLE program allows interactive selection of ground control points (GCPs) for image rectification. IDRISI includes excellent softcopy documentation, including a manual, a tutorial, and outstanding on-line help for each module. The tutorial includes images from a wide range of sensors and geographic regions. Technical support is available, and an unmoderated discussion list provides access to many in the large IDRISI user-community. The modular structure of IDRISI programs and data files is both an advantage and disadvantage to the program. It is an excellent teaching environment; users are forced to understand the sequence of steps involved in analysis routines. However, the numerous steps in undertaking an analysis, especially with multiband imagery, can be daunting. Clark Labs, the developer of IDRISI, is addressing this issue, and is beginning to rewrite the separate routines as integrated modules. Furthermore, options for specifying groups of files, such as the individual image bands or classification signatures, are provided. Features The IDRISI Kilimanjaro edition is the 14th major release, and represents more than 18 years of development of this popular rasterbased GIS and remote sensing package. Clark Labs, the developer of IDRISI, also sells a companion GIS product, CartaLinx, which has functionality focused on digitizing, database development and topological editing. IDRISI has nearly 200 modules, which are organized in menus in seven major groups, with most of the analytical functionality concentrated in the GIS Analysis and Image Processing Menus (Table 1, page 669). The IDRISI modules include a wide variety of basic analytical tools, as well as many advanced procedures. In addition, because IDRISI tends to generate many individual files, programs for file management are also included. A particularly effective program is the graphical Macro Modeler, which allows users to develop and link a sequence of image analysis routines. This program is useful for designing an image analysis in a conceptual manner, for speeding up implementation of a complex sequence of tasks, for building a macro for repeating an analysis sequence on different data sets, and for documenting analysis procedures. The module for image display is at once sophisticated and a little awkward. The awkwardness is perhaps a legacy of IDRISI’s GIS focus, and the IDRISI modular file structure. Thus display of a false color composite requires prior generation of the image with the program COMPOSITE. Alternatively a false color composite can be created on the fly, but the three image files have to be opened separately, and sequentially assigned to the red, green, or blue image plane. On the other hand, the controls for autoscaling, palette files (for map display colors), blending between multiple layers, and the ability to modify and save the display as a map file, are all impressive. Furthermore, this edition includes the new program, Fly Through, which allows the user an interactive fly over of an image draped over a digital elevation model (Figure 1). In addition, 3-D viewing of stereopairs is now possible, using anaglyphic glasses. Both the interactive fly over and anaglyphic viewing programs are intuitive, and easy to use. The Kilimanjaro edition includes a number of other new programs, such as new classification routines: a back propagation neural network (Figure 2), canonical components analysis and a Mahalanobis distance soft classifier. New modeling enhancements include the Revised Universal Soil Loss Equation, the GEOMOD predictive land change simulation model, and an enhanced RUNOFF module, which now allows the incorporation of variations in precipitation and permeability. The image import functions, a subject of complaint in a review of the first 32-bit version of IDRISI (Green and King, 2000), has been strengthened to include a wider range of data formats, and now includes, for example, 16-bit TIFF, HDF, Imagine .img, ENVI and ERMAPPER formats. Performance To test the performance of IDRISI, we compared IDRISI to two industry standard software packages, Leica Geosystems ERDAS Imagine 8.7, and Research Systems Inc. ENVI 4.0. The test machine was a 2.4 GHz Pentium 4 Dell, with 1 GB of RAM. The test data was an EROS Data Center NLAPS format Landsat 7 Enhanced Thematic Mapper Plus (ETM+) image (6,522 rows by 6289 columns, 6 bands, 240 MB). Five tests were carried out (Table 2), designed to test the range of spectral and spatial functions that a user might draw on: import, export, spectral transformation, classification, and spatial filtering. The benchmarks reported in Table 2 should be interpreted with caution. Firstly, the results potentially vary greatly with file size, format of output selected, and the hardware configuration. In addition, some of the differences tend to cancel out when the larger context is considered. For example, IDRISI’s native file format is very similar to that of the NLAPS format, with each band in a separate file. On the other hand, Imagine tiles the data, to facilitate rapid display of data, and also calculates the image statistics. Thus, although Imagine took over seven minutes compared to IDRISI’s 18 seconds to import the data, the extra time spent by Imagine data is potentially recovered later, in more rapid and interactive data viewing. Nevertheless, it is very clear from Table 2 that IDRISI is generally very fast, and compares favorably with both of the other packages to which it was compared. Perhaps most impressive was the rapid speed of the IDRISI principal components analysis, just under three minutes, compared to over 11 and 12 minutes for ENVI and Imagine, respectively. The only slow IDRISI program in the list of benchmarks is the export to a 3-band jpeg format file, for which IDRISI took over 10 minutes, compared to less than half a minute for the other two software packages. For IDRISI, this procedure required two steps, first creating a 3-band false color composite image (which was a very rapid 22 seconds), and then the jpeg export step. This latter step was unfortunately somewhat confusing, because the program spent less then a minute speeding to a reported 99.98% completion, and then spent 10 minutes making no visible progress, before suddenly finishing. The first time we ran this program we ended up rebooting the computer after not seeing any progress for five minutes, and thinking the program had crashed. Documentation The IDRISI documentation is excellent, though entirely softcopy in format. The 328-page IDRISI Kilimanjaro Guide to GIS and Image Processing, in Adobe PDF format, provides both a thorough overview of GIS and image analysis topics, and an explanation of which Kilimanjaro programs relate to those topics. For example, the section on classification of remotely sensed data starts with a two and a half page description of classification. The basic ideas of class signatures, supervised, and unsupervised classification, are presented, followed by a discussion of fuzzy classification and a comparison between multispectral and hyperspectral methods. This in turn is followed by detailed descriptions of the sequence of Kilimanjaro programs that would be followed for supervised, unsupervised and hyperspectral classification. These descriptions of the recommended program sequences are important to users because, unlike other more integrated software, IDRISI tends to require many separate steps for a single task, such as classification. Given the many possible ways the steps could be put together, it is perhaps not surprising that the manual doesn’t include every program. Nevertheless, it is a pity because the real challenge for IDRISI users is to understand the sequence required to put the programs together. For example, the classification signature program PURIFY does not appear to be discussed in this particular manual, and therefore the user is not given an explicit guide on its overall position in a classification procedure. On the other hand, its sequence is implied from the menu structure, where it is grouped with other Signature Development programs. The second major manual is the 270-page IDRISI Kilimanjaro Tutorial, consisting of six major exercises, each comprising numerous subsections. Thus, for example, the Advanced Image Processing Exercise section includes subsections on standard parametric classification, and two soft (fuzzy) classifiers, as well as a section on image transformations such as PCA, and vegetation indices. The tutorial includes questions, with answers at the end of each section. A large amount of sample data (nearly 460 MB) is provided with the tutorial. In fact, the sample data makes up most of the almost 600 MB of the IDRISI installation. The sample data is drawn from a wide range of geographic areas and image types, which adds considerably to the value of the tutorial. A third manual, the 38-page IDRISI Applications Programming Interface (API) User’s Guide, provides information on how to use high level languages, such as Visual C++ or Visual Basic, as a macro language to control operations of IDRISI. In addition to the pdf documentation, IDRISI has an excellent html format help system, with useful search and index functions not available with pdf documents. The help button in the dialogue box of each program takes the user immediately to the specific help for that program (Figure 3), which includes:
Clark Labs also offers for sale six United Nations Institute for Training and Research (UNITAR) workbooks that cover GIS and remote sensing applications for specific topics such as change and time series analysis, or applications in hazard assessment and risk management. Four volumes of global change data, including monthly vegetation index data, elevation, and sea surface temperature, are also available for sale. Technical Support Clark labs offer a wide range of technical support options (see box at the start of this review). On the purchase of any Clark Labs product, with the exception of the Student Starter product, 30 days of free support is provided. Following that initial period, support can be purchased on an annual basis. Contact with technical support is through a web-based form or email. Support is offered to solve problems associated with software installation, tutorial exercise assistance, software operation, import and export of supported data formats, and identification of appropriate modules for a particular analysis. Support is thus particularly useful for IDRISI users relatively new to remote sensing who may need initial help with understanding the best approach for particular problems. Because of the large user base for IDRISI, the unmoderated IDRISI-L discussion list provides a good additional source of information. The IDRISI-L messages are also archived on the Clark Labs web site (follow the links for Resources, then User Forum), and represent yet another source of potential help. Ease of Installation IDRISI Kilimanjaro comes with the now standard Windows automated installation software. Installation is straightforward and relatively quick, considering the nearly 600 MB of disk space for a full installation. It is recommended that previous versions of IDRISI be uninstalled prior to installing the new version, which may not be convenient for users who prefer to transition slowly to this new version of the program. In addition, the default installation is not in the Program Files directory, where most software packages install. After initial installation of IDRISI Kilimanjaro, the program will run in trial mode for seven days. A licensing wizard is used to generate a unique code from hardware on the installed computer, and the user has to email, fax, or telephone this information to Clark Labs to obtain the permanent license key code. A patch for 14.02 was recently released, and a link to the download web page was available on the Clark Labs homepage at the time of writing this article. Ease of Use The IDRISI program interface allows a wide range of methods for running the programs.
Pros IDRISI’s academic origins are evident in the many sophisticated and innovative algorithms it includes. To take just one example, the range of classification modules in IDRISI is amongst the best of any remote sensing package: 8 hard classifiers, including a neural network classifier, 5 soft classifiers and related hardeners, and a suite of hard and soft hyperspectral classifiers, including an algorithm based on the USGS Tetracorder program. IDRISI tends to be highly modular in design. Depending
on your outlook this can be a strength or weakness.
It is a superb teaching tool, because the user is forced
to understand every step in the procedure. Thus for
example, the sequence for a maximum likelihood classification
starts with the user displaying the image, and digitizing
training sites on the screen, using only numbers and
not names to represent each class. The digitized polygons
are saved, and then subsequently run through a separate
program (MAKESIG) to generate the class statistics,
or signatures. The signatures can be compared, graphed,
or even refined, all through additional separate programs,
such as SCATTER, SIGCOMP, and PURIFY, before running
them through the independent MAXLIKE program, or one
of the many Another of IDRISI’s strengths is the excellent documentation and help. Remote sensing software often provides rather limited help; IDRISI sets a high standard that its more expensive competition should emulate. Cons The single biggest drawback with IDRISI is that the highly modular structure of the software can make for rather tedious and repetitious work. As discussed in the previous section, this modular structure may be a benefit for teaching, but for operational use, it can be daunting. The difficulties of dealing with modular programs are exacerbated by the fact that a modular structure is also used for data files. Each separate type of information is for the most part stored in its own separate file. Thus the IDRISI File Explorer (an interface for managing and viewing IDRISI data files) lists 29 different basic categories of files, some of which comprise two separate types of files, because the metadata are stored separately from the data. For example, each band of an image is stored in a separate file, with a separate metadata file. Likewise, each signature for a classification is stored in two separate files, one with the class statistics and another with the original pixel values for the signature. The difficulties associated with the modular approach to IDRISI have been addressed by a number of recent innovations in the program. The first is the very impressive Macro Modeler program, discussed earlier. Secondly, some file types, including notably raster layers and signatures, can be linked through so-called group files. Unfortunately, though, the display routines do not allow one to use the group files as a way of speeding up the selection of bands for a false color composite. Thirdly, new modules are being developed to address some of the least interactive programs. For example, a new version of RESAMPLE, distributed in patch 14.02, provides an interactive, integrated interface for image rectification. Prior to this patch, the user manually entered the ground control points (pixel coordinates and equivalent ground locations) in a text file, prior to running the module that generates the geometric transformation. Licensing is the bane of all software companies, because of the difficulty in providing a secure, but simple, method for licensing. Clark Labs has tried to be creative in establishing a licensing policy that keeps the price of IDRISI very low by using licensing software that is keyed to the hardware and operating system of the licensed machine. Unfortunately, such an approach is inconvenient for those wishing to move the program between computers, such as a workstation and a laptop. This problem could be overcome were Clark Labs to offer a USB or parallel port hardware key, which could be moved easily. However, this would add to the cost of the program (James Toledano, Clark Labs, personal communication, 2004). Recommendations It is not surprising that there are, according to Clark Labs, more than 35,000 installations of IDRISI in 170 countries. IDRISI has a most impressive combination of low price and high power. As a consequence, the users are highly diverse, including both advanced researchers, such as research scientists interested in the geostatistical analysis utilities, and new users, drawn by the low price. As new programs are developed to increase the interactive nature of the more tedious utilities, IDRISI will gain even more supporters. A particularly important step will occur when IDRISI offers a more direct method for displaying a three-band false color composite, so that the image need not be prepared in advance, or created through a cumbersome series of file opening steps. References Green, D. R. and S. D. King, 2000. IDRISI32 Software Review. Photogrammetric Engineering and Remote Sensing 66 (11): 1295-1299. Reviewed by: David Campagna
|