ArcSDM3 in ArcMap Overview
In implementing ArcSDM3 to run in ArcMap 8.3 it was our intent to translate the Arcview3 Avenue code to VB to operate with ArcMAP. We intended to fix known bugs in the Arcview3 version and to follow the look and feel of the Arcview3 version as closely as possible. In the implementation of the raster processing extension, Spatial Analyst, ESRI has made decisions that significantly change how ArcSDM3 is implemented. The most significant change is that attributes cannot be added to the VAT of a raster file. The solution we have adopted is to create tables that must be joined to the VAT of the evidential raster. Thus, when an evidential raster is generalized a new table, a GEN table, is created. The use of a joined table with the VAT requires the user to be familiar with the Join tools, which have been significantly enhanced in ArcMap. Associated with the use of joined tables with rasters is that ESRI does not provide the tools in version 8.3 to symbolize a raster in code. So the user has to symbolize after the tables are programmatically joined.
The Identify tool in ArcMap8 did not include the joined attributes for rasters. Similar problems continue with ArcMap 9.0
Because of these differences between rasters in Arcview3 and ArcGIS, projects created for ArcSDM in Arcview3 cannot be directly used in ArcSDM3 for ArcMap. It is best to recreate all the rasters to be used because the added attributes in the VAT of Arcview3 rasters cause conflicts when used in ArcSDM3 for ArcMap.
All of these issues are discussed below, and workarounds for these problems are described.
Software Requirements: ArcSDM3 requires ArcMap 9.0 and Spatial Analyst. Also, when installing ArcSDM3 do not update any files when requested during the install, always select to Keep existing files. Such updating is a feature that only a System Administrator should indulge as it can mess up system files. You may have to reinstall/update both Windows, ArcMap, and ArcSDM3. Or try the installations on a clean computer.
For ArcMap9 use ArcSDM3. There are essentially no differences between ArcSDM3 for ArcMap8.3 and ArcSDM3 for ArcMap9, except that ESRI reorganized the ArcObjects DLL library so modified VB code has to be appropriate for the ArcMap version.
It is critical that ArcGIS Service Packs should be installed. This software was developed with Service Pack 3 for ArcGIS 9.0 and will not work properly without this Service Pack.
The operating system should have current service packs installed. It is recommended that ArcSDM3 be used with XP
Next | Section Contents | Home |
Migration from Arcview3 to ArcMap
Rasters from Arcview3 with additional attributes cannot be used in ArcMap while using ArcSDM3. A raster in ArcMap made from a character attributed shapefile or coverage, will have a value, count, and a third attribute, which is named the same as the source attribute. These three attributes are all that is allowed in a raster derived from a character source. A raster made from a integer or floating valued shapefile or coverage will have only a value and count attribute. These two attributes are all that is allowed in ArcMap while using ArcSDM. Users will note, ESRI allows extra attributes in the output from the raster command COMBINE. These extra attributes are the values from the input rasters to the COMBINE command. So ESRI can add extra attributes, but the users cannot in ArcMap 8.3.
It is not advised to use a raster made in ArcMap in
Arcview3 with ArcSDM. Errors in calculation have been observed in Arcview3
when training site files made in ArcMap with the associated metadata are used in
Arcview3. So do not attempt to go back to Arcview3 to do an ArcSDM analysis with
things created in ArcMap.
Next | Section Contents | Home |
New and Changed Feature in ArcSDM3
The design concept in ArcSDM3 for ArcMap has been to translate the Arcview 3 implementation. Thus, we have attempted to implement the Arcview 3 approach as closely as possible. This may not be the best way to implement our spatial modeling methods in ArcMap.
The ArcSDM3 menu has been reorganized in submenus. The submenus are Preprocessing, WofE – LR, Fuzzy Logic, Neural Networks, and Postprocessing. The Preprocessing submenu includes Set Analysis Parameters, Add Bearing, and Extract Contact. The design concept is still that you should start modeling with the top selection and progress to the bottom. Thus modeling starts with the Preprocessing submenu in which the modeling method and the analysis parameters are selected. Then in the selected modeling method submenu, the process is to proceed from the first menu option through the last, sequentially.
ArcSDM3 no longer allows evidence layers to be coverages and shapefiles. All evidence layers must be integer rasters. Evidence that is to be used for proximity to something, such as proximity to faults, can be made into a raster by buffering the selected features with the Spatial Analyst Distance tool, which creates a raster with values of distances around the selected features. Such a raster has floating or real values so it must be converted to an integer raster before it can be used by ArcSDM3. This conversion can be done with the Spatial Analyst Reclassify or Raster Calculator using the integer function.
ArcSDM3 requires that the training points be a shapefile. Points in a coverage or other format must be converted to a shapefile.
Training points can be given a fuzzy membership attribute that is used by the neural networks. The fuzzy membership can be thought of as a ranking from [0,1] indicating the quality of the training site. For example, large deposits might be given a larger membership than small deposits in "Deposits" training sites and some "Not Deposit" training sites might be somewhat similar to a "Deposit" training point.
Starting a new modeling project
See the instructions for starting a new modeling project in the Installation instructions.
After loading the study area and training points, the map document must be saved. ArcSDM3 will use the location of the map document (the MDX file) as the place to store all files created by ArcSDM3. All ArcSDM save operations will be directed to this workspace independent of all other system explorer operations. To change the workspace for the ArcSDM model, it is necessary to save a new map document with Save As.
There can be problems with long paths. It is recommended that the path to the workspace is short.
You can control where ArcMap Spatial Analyst writes temporary files in the Spatial Analyst Options menu.
The selections in the Preprocessing submenu work the same
as they did previously. Buffering of features and reclassification are now done
with the ArcMap tools. The Add Bearing and Extract Contacts works as before.
Before starting a model, it is assumed you will have a
training set, a study area raster that is an integer raster, and some evidence
layers that are integer rasters for all methods except fuzzy logic. All evidence
layers, except for the fuzzy logic method, need to be preprocessed to integer rasters. Shape files and coverages are no longer recognized as viable inputs.
Previously shapefiles and coverages were converted to rasters during the weights
of evidence processing. This significantly increased the processing time and
complexity of the programming.
The proper way to start a new modeling project is out lined below
Remember that any rasters used in ArcSDM must be permanent rasters.
The Add Azimuth function has been improved so the reported azimuth is the northern hemisphere azimuth. Optionally the azimuth may be reported as a bearing between -90° and +90°.
The study area raster defines the extent, mask, and cell size for the area to be analyzed and should define the projection of the Data Frame. This raster must an integer raster with a cell size equal to or less than the cell size of the evidential raster with the smallest cell size. This is because the study area raster is also used to find the area of missing data.
When making the study area raster, make a raster with only positive integer values greater than 0, that is unsigned pixel type. The study area raster can be multi-valued positive integers.
The simplest way to create a studyarea raster is to create a
polygon shapefile outlining the study area. Add a text or character attribute to
the shapefile. Then convert the shapefile to a raster using the text attribute
to create an integer-valued raster. If a numerical attribute of a polygon
shapefile is used to create the raster; this will produce a floating point
raster. ArcSDM requires an integer raster. The floating-point raster can be
converted to an integer raster using the Spatial Analyst Raster Calculator with
the integer function. The equation for a raster file called temp, would be int([temp]).
If you create a studyarea raster from another raster, such as a 16-bit raster, make sure that the attribute table opens. There is a known bug in ArcMap 8.3 that does not allow ArcMap to see a attribute table with only 1 record. This problem seems to only occur if the source of the study area raster is a 16-bit raster. It does not matter how many records are in the study area raster attribute table. So verify that the Study area raster exists by opening it.
This modeling submenu includes the data-driven and expert weights-of-evidence method and logistic regression. The major change in this submenu is first encountered in the Generalize Evidential Theme submenu selection. The generalization is now stored in a generalization table. The user should be familiar with the method to join tables and to manage table joins. The Conditional Independence Ratio and chi-squared test for conditional independence have been replaced by the new Agterberg-Cheng conditional independence test.
ArcSDM uses simple default settings to create graphs or
charts. Because of the many possible selections by the user and the potential
wide range of values, the graphs as initially defined are somewhat crude. If you
want to refine the graph, it is necessary to edit the properties of the graph.
The graph property window can be opened by right clicking on the title bar at
the top of the graph and selecting properties.
You can modify colors with the following steps.
You can modify tic marks with the following steps.
This submenu selection works the same as in the previous
version except in one important aspect. It is no longer possible to add
additional attributes to a VAT; so the generalization process now creates a
generalization file, known as a GEN file. The naming convention for the GEN
file is Gen-evidential raster weight table.dbf. For example if the kbgeol
evidential raster, which uses a weights table kbgeol-CT.dbf, then the
generalization file created would be named GEN-kbgeol-CT.dbf. The characters
through the evidential raster name, “GEN-kbgeol-“, should not be modified by
the user because this is what the software uses to associate GEN files with
evidence layers. The user is free to modify the name following the second dash
as they see fit.
When the Gen table is created, it is joined to the evidential raster. ESRI has not allowed for programmatic symbolization on an attribute in a joined table. Therefore it is necessary to interactively symbolize by one of the generalization attributes.
This submenu selection works the same as in the
previous version except it is necessary to select a GEN file instead of a
attribute in the VAT to define which generalization to use.
The conditional independence test now utilizes the Agterberg-Chang conditional independence test in addition to the conditional independence ratio, CI ratio.
When the calculations are complete, the area (e.g. SDMUC001-Tbl.dbf) and wofe tables (e.g. wofe001.dbf) will be joined to the unique conditions raster. ESRI has not allowed for programmatic symbolization on an attribute in a joined table, therefore it is necessary to interactively symbolize by one of the posterior probability attributes.
The response theme will be given a name, such as SDMUC1, but in the Properties/Source this raster will be seen to point at a raster with a name like RASTER9, and this raster will be a temporary raster until it is made permanent or the project is saved. Therefore, when the user decides that a response theme should be kept and before the project is saved, it is a good practice to make the raster permanent and give it the name such as SDMUC1. It is best to keep this SDMUC name because all the associated tables will have this prefix in their name. Then if the project MXD file is lost or corrupted the response-theme rasters will still have recognizable names.
ArcGIS9 does not automatically make new rasters created by ArcSDM3 permanent. It is necessary to make the results (SDMUC rasters) permanent by right clicking on the raster in the Table of Contents and selecting the Make Permanent option. This will provide an opportunity to move this SDMUC-results raster to the working folder. It is recommended that you keep the default names to keep the association with tables and so ArcSDM3 can recognize file types.
Symbolization has to be done interactively. For WofE and LR models, symbolization by natural breaks seems to work well.
The fuzzy logic submenus still include a Fuzzy Membership and a Fuzzy Logic submenus. The Fuzzy Membership submenu is used to create a new raster representing the fuzzy membership values for an evidence layer. The naming convention for an evidential raster named kbgeol, is Fmkbgeol. Fmkbgeol is a floating point raster with values ranging from [0,1]. The Fuzzy Logic submenu is a fuzzy operator calculator. A fuzzy model DBF file is still created to store the parameters of a fuzzy logic model. We have not implemented the tools to draw a fuzzy-logic inference net nor to run an existing fuzzy logic model. The Fuzzy Membership values are now created by functions for ordered data and by a table method for categorical data. A table is created from the table method to record the fuzzy membership values assigned to the categorical data. These fuzzification functions will not deal with negative values in an evidence layer; so all evidence layers must be positive integers.
The neural network functions work the same except that it is necessary to create a generalization file before creating the unique-conditions raster, that is the Response Theme. The menu selection Generalize Evidential Theme has been added to the Neural Network submenu. This selection is the same as in the WofE menu. The Map Classes method of generalization has been modified to include a Unique button. Left clicking the Unique button will activate the OK button so a generalization file can be created that reproduces the original classes. There is also a Unique method on the main Generalization window. If for some reason another type of generalization is desired, for example one with fewer classes maybe to speed the neural network calculation, then other methods of generalization could be used.
A new feature to be added to the neural networks is
the ability to assign fuzzy membership values to the deposit and ‘not
deposit’ training sites. Previously, these were assigned fuzzy memberships of
1 or 0, but Carl Looney has advised us that the neural RBFLN will work better
if the training sites have fuzzy membership values other than just 0 or 1.
Fuzzy membership values might be defined on deposit size, for example.
ArcGIS9 does not automatically make new rasters created by ArcSDM3 permanent. It is necessary to make the results (SDMUC rasters) permanent by right clicking on the raster in the Table of Contents and selecting the Make Permanent option. This will provide an opportunity to move this SDMUC-results raster to the working folder. It is recommended that you keep the default names to keep the association with tables and so ArcSDM3 can recognize file types.
Symbolization of the results from GeoXplore have been a problem. It has been found that symbolization works best using the quantile method with a larger number of classes. Fifteen to 20 classes is often a good starting place.
In the previous version,
most of the post processing functions had to do with symbolization. ESRI has not
allowed for programmatic symbolization on an attribute in a joined table.
Therefore, these symbolization selections cannot be implemented.
The following methods have been implemented and are the same as before: Compare Results, Associate Response with Point Theme, and Areas vs. Probability. In addition a new function, Create Raster has been added to Post Processing.
There is a known bug in
Compare Results when used to compare responses within a SDMUC raster, for
example to compare a logistic regression response with the associated
weights-of-evidence response in the same SDMUC file. The Spearman's correlation
will be reported as 1. While this number is generally about right for logistic
regression and weights of evidence from the same evidence, the correlation
between a Fuzzy Neural Network and RBFLN responses are often highly
noncorrelated. So Compare Response only works properly when comparing between
difference SDMUC files. A work around is to run a model twice so the
weights-of-evidence response and the logistic-regression response or Fuzzy
Neuraal Network and RBFLN responses are in separate SDMUC rasters.
Create
Raster is used to create a floating-valued raster from a attribute in a table
joined to the raster, such as the calculated
posterior probability or some integer-valued raster from a generalization. One use of such rasters would be as inputs to a cross
tabulation (Tabulate Area in Arcview 3) for correlation analysis. Another use
would be make a raster to mask out areas with values less than or equal to the
prior probability or a confidence mask as in the previous version.
New ArcSDM Rules for Version 3.
Next | Section Contents | Home |
Buffering tool has been replaced.
The Buffer tool available in the ArcView3 version of ArcSDM has been eliminated and replaced with the ArcMap Spatial Analyst buffer tool, Straight Line Distance (which is also called Euclidean Distance). The Spatial Analyst buffer or straight line distance tool for creating rasters of distance from a selected vector or raster class can be found in the Spatial Analyst Distance submenu by selecting the Straight Line tool. Information on this tool is found in the ArcMap Help by searching on Euclidean Distance tool/command and select the help in the Spatial Analyst Toolbox.
The Straight Line tool creates a raster of cells with the value in floating map units of the distance to the nearest occurrence of the object buffered. In order to use these distance rasters for buffer or proximity evidence in a model, it is necessary to convert the floating valued raster to an integer raster. Typically the units of distance will be in meters, which might be a finer resolution than the data supports; so in this conversion to integers the units are often rescaled to kilometers. The ArcMap Raster Calculator tool is useful for this rescaling and conversion to integer using the integer function, INT. Do not forget to set a mask in the Spatial Analyst Option submenu to limit the calculations to the study area.
Next | Section Contents | Home |
Reclassify tool has been replaced.
The Reclassify tool available in the Arcview3 version of ArcSDM has been eliminated and replaced with the ArcMap Spatial Analyst Reclassify tool. The Spatial Analyst reclassify tool for reclassification of rasters can be found in Spatial Analyst menu by selecting the Reclassify tool. For guidance on this tool see the ArcMap Help and search for Reclassify tool/command and select the help in the Spatial Analyst Toolbox. This tool provides many of the standard reclassification methods of ArcMap and produces an integer raster that can be used as evidence. In ArcMap8.3, the load and save buttons do not function, but this problem is fixed in the ArcMap9 version. These buttons are useful for saving a record of the reclassification and using it again later. It is necessary to make this raster permanent, rename it with a 9-character or less name, and store it in the ArcSDM3 work area folder. Do not forget to set a mask in the Spatial Analyst Option submenu to limit the calculations to the study area.
Next | Section Contents | Home |
Note on terminology
differences between old and new versions
Arcview 3 |
ArcMap |
Comment |
‘old’ term |
‘new’ term |
|
theme |
layer |
May consist of shapefiles or rasters |
Grid |
raster |
Gridss now termed rasters * |
*Whereas ArcSDM Version 1 (for Arcview 3) allowed input themes (termed evidential themes) to be either shapefiles or rasters, ArcSDM Version 2 (for ArcMap) will only work with input layers that are rasters. Accordingly, documentation for ArcSDM version 2 will refer to what was termed ‘evidential themes’ in ArcSDM version 1 as ‘evidential rasters’. However, this change may be not be consistently applied, so for example ‘response theme’—the term for an output map, often a posterior probability map-- is equivalent to the term ‘response raster’, and both are used in the current documentation interchangeably..
Next | Section Contents | Home |
Join Management
ArcSMD2 creates GEN tables to store the generalization of rasters (evidence layers). These GEN tables are automatically joined to the raster to be generalized so the user can view the generalization. When the user performs a Calculate Response in the weights-of-evidence tools or Generate Input Files in the neural network tools, the generalization to be used is selected by the user. Then ArcSDM3 automatically unjoins previously joined tables and joins the selected GEN table. If for some reason the user has several tables joined to the raster, all of the previously joined tables will be unjoined. Consequently, while exploring data and deciding which generalizations to use, the user will have to make use of the Join tool.
Join management turns out to involve some interesting quirks. The way ArcMap deals with joins is when a second join is added to a raster, which already has a join, the two joined tables are put into a Rel-Query table. Then if a third table is joined to the same raster, this third table is actually joined to the Rel-Query table. Consequently, if the user wants to unjoin the second joined table, all of the tables will be unjoined. So joins are managed by ArcMap, by creating a hierarchy of joins. Only the last table joined can be individually unjoined. Any unjoining of tables joined in the middle of the sequence will delete at least all of the tables at lower levels in the hierarchy.
More information about table joins can be found in the ArcMap help file under the Contents tab, then open ArcMap, Working with tables, Joining attribute tables. This section will provide an overview of the join tools, but it give no insights into the limitations and differences when joining tables to rasters.
Next | Section Contents | Home |
Limitations of Identify tool:
ArcMap8.3 will not identify attributes in tables joined to rasters. Thus, it is impossible to see any attribute other than the value attribute. To see the unique-conditions attributes, it is necessary to use the identify to get the value attribute and then to manually inspect the unique-conditions raster attribute table.
This limitation is gone in ArcMap9.
Next | Section Contents | Home |
ArcMap9 will not allow for symbolization of a raster based on an attribute in a joined table programmatically with VB code. Therefore, when a model is completed the user is informed the raster is ready for symbolization. This will typically be the top raster in the tables of contents and it will have as name like SDMUC001. A convenient way to symbolize this raster is to create a color ramp with the colors you typically like to use. For example, flipping the Full Spectrum color ramp that come with the ESRI style is a common color ramp to apply to a posterior probability attribute. This color ramp can be save to your personnel style or saved in a modeling style; so it is always available. Search ArcMap's Help file on Styles and Creating and Style Manager to find guidance on this process.
There is a known bug in ArcGIS9.0 that causes improper symbolization of attributes in a table joined to a raster by the Classified methods. Thus, the posterior probability will symbolize improperly if symbolized by Natural Breaks. Typically the error is seen as the highest value is symbolized in the lowest color class. The work around for this bug is to symbolize by Unique values for the joined attributes. We are told this known bug will be fixed in ArcGIS9.1.
Next | Section Contents | Home |
In order to modify tables in ArcMap, one way to to change text and values in a table simply by opening the table, selecting the rows and single column to be modified, and then right-clicking on that row to select Calculate Values. Then you can calculate new values or text for the selected rows in that selected column. The calculator can be used to calculate a new value based on other values in the table or simply to add a number by putting that number in the calculator. To calculate a character value, enclose the characters in quotes ("xyz").
If you want to add an attribute field to an existing table, it is quite simple; however there are some confusing properties of the numerical attributes for which the documentation is difficult to find. First, it is not possible to add a field to an attribute table of a raster. You can only add fields to DBF tables and attribute tables of shapefiles. To add a field, open the table, and select Add Field... in the Options menu. If you are adding a integer field, the form asks for the precisions. If you adding a floating or double numerical field, the form asks for precision and scale. The definitions of precision and scale are found in "Building a Geodatabase" by Andrew MacDonald on pages 57-58. This book is on the Digital Books CD-ROM that comes with the installation disks for ArcMap.
The precision and scale of a field describe the maximum size and precision of data that can be stored in it. The precision describes the number of digits that can be stored in the field, while the scale describes the number of decimal places for float and double fields. When creating a new field in a geodatabase feature class or table, you can specify the field’s type, precision, and scale. When the field is actually created in the database, the field type may be changed based on the precision and scale values you specify.
Use
the following guidelines for choosing the correct field type for a given
precision and scale:
Next | Section Contents | Home |