root/MGET/Branches/Jason/PythonPackage/dist/TracOnlineDocumentation/Documentation/ArcGISReference/GAM.PredictFromArcGISRasters.html @ 842

Revision 842, 19.1 KB (checked in by jjr8, 19 months ago)

Rebuilt installation packages again. This will be merged with the Trunk and released as MGET 0.8a28.
Line 
1<?xml version="1.0" encoding="utf-8"?>
2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
3<html xmlns="http://www.w3.org/1999/xhtml"><head><link rel="stylesheet" type="text/css" href="81help.css?format=raw" /><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>Predict GAM From Rasters</title></head><body><table style="margin-top:-1em; margin-bottom:0; padding:0; margin-left:-1em"><tr><td style="background:white"><img width="875" height="70" alt="ArcToolbox banner" src="AHBanner_ArcToolbox.gif?format=raw" /></td></tr></table><h1>Predict GAM From Rasters</h1><p></p><p>Using a fitted generalized additive model (GAM), this tool creates a raster representing the response variable predicted from ArcGIS rasters representing the predictor variables.</p><br /><p><h2><img width="11" height="11" border="0" src="sm_arrow_down.gif?format=raw" /> Command line syntax</h2></p><div Class="expand" id="id103139">GAMPredictFromArcGISRasters_GeoEco &lt;inputModelFile&gt; &lt;inputPredictorRasters;inputPredictorRasters...&gt; &lt;variableNames;variableNames...&gt; &lt;outputResponseRaster&gt; {templateRaster} {resamplingTechniques;resamplingTechniques...} {ignoreOutOfRangeValues} {outputErrorRaster} {outputBinaryResponseRaster} {cutoff} {buildPyramids} <br /><br /><b>Parameters</b><br /><table width="100%" border="0" cellpadding="5"><tbody><tr><th width="40%"><b>Expression</b></th><th width="60%"><b>Explanation</b></th></tr><tr><td class="info">&lt;inputModelFile&gt;</td><td class="info" align="left"><p>File that contains the fitted model, generated by the Fit GAM
4tool.</p></td></tr><tr><td class="info">&lt;inputPredictorRasters;inputPredictorRasters...&gt;</td><td class="info" align="left"><p>Rasters that represent the predictor variables used in the model.</p><p>This list must include one raster for each predictor variable that
5appears in the model's formula. The list provided for the Model
6Variable Names for Predictor Rasters parameter specifies the predictor
7variable that each raster represents.</p><p>For example, if your model used the formula:</p><dl><dt></dt><dd><pre>Presence ~ SST + Chl + Depth</pre></dd></dl><p>You must specify three rasters, one for each predictor variable. If
8you want to predict Presence for the month of January 1999, you might
9use a static bathymetry raster and SST and chlorophyll rasters for
10that month:</p><dl><dt></dt><dd><pre>C:\Data\etopo2v2
11C:\Data\SST\sst199901
12C:\Data\Chlorophyll\chl199901</pre></dd></dl><p>Then, for the Model Variable Names for Predictor Rasters parameter,
13you would specify the predictor variable that each raster represents:</p><dl><dt></dt><dd><pre>Depth
14SST
15Chl</pre></dd></dl><p>You can specify extra predictors that do not appear in the formula.
16They will be ignored. In the example above, you might include the
17variable SSH (sea surface height) and the raster
18C:\Data\SSH\ssh199901. Because this variable does not appear
19in the model formula, it will be ignored.</p><p>All of the predictor rasters must have a coordinate system defined.
20They must all share the same datum. If you also specify the Template
21Raster For Outputs parameter, they must have the datum of the template
22raster. If you do not specify that parameter, the first predictor
23raster will be used as the template raster.</p><p>The predictor rasters need not have the same coordinate system,
24extent, or cell size as the template raster. If these characteristics
25differ from the template raster, this tool will automatically project
26and clip the predictor rasters to conform to the template raster. The
27predictor rasters must be able to be projected to the template
28raster's coordinate system without requiring the specification of a
29geographic transformation. An error will be reported if a geographic
30transformation must be specified. In this case, you must project the
31predictor rasters manually before providing them to this tool.</p><p>By default, floating point predictor rasters will be projected using
32bilinear interpolation and integer predictor rasters will be projected
33using nearest neighbor assignment. Bilinear interpolation is
34appropriate for continuous variables such as sea surface temperature,
35while nearest neighbor is appropriate for categorical variables such
36as bottom substrate type. If these defaults are not appropriate for
37your predictors, specify different algorithms using the Resampling
38Techniques for Predictor Rasters parameter.</p></td></tr><tr><td class="info">&lt;variableNames;variableNames...&gt;</td><td class="info" align="left"><p>Model variable names for the input predictor rasters. The variable
39names are case sensitive. Please see the documentation for the Input
40Predictor Rasters parameter for more information.</p></td></tr><tr><td class="info">&lt;outputResponseRaster&gt;</td><td class="info" align="left"><p>Output raster representing the response predicted by the model
41from the input predictor rasters.</p><p>If any pixel is NoData in any predictor raster (after it has been
42projected and clipped as needed), the pixel will be NoData in the
43output raster as well. This is because the response variable cannot be
44predicted by the model if any of the predictor variables are
45missing.</p></td></tr><tr><td class="info">{templateRaster}</td><td class="info" align="left"><p>Template raster that defines the coordinate system, extent, and
46cell size of the output rasters produced by this tool.</p><p>If you do not specify a template raster, the first predictor raster
47will be used instead.</p><p>All predictor rasters must share the same datum as the template
48raster. The predictor rasters need not have the same coordinate
49system, extent, or cell size as the template raster. If these
50characteristics differ from the template raster, this tool will
51automatically project and clip the predictor rasters to conform to the
52template raster. Please see the documentation for the Input Predictor
53Rasters parameter for more information.</p></td></tr><tr><td class="info">{resamplingTechniques;resamplingTechniques...}</td><td class="info" align="left"><p>Resampling technique to be used for each input predictor raster if
54that raster needs to be automatically projected, as described in the
55documentation for the Input Predictor Rasters parameter.</p><p>The available resampling techniques are:</p><ul><li><p>NEAREST - Nearest neighbor assignment</p></li></ul><ul><li><p>BILINEAR - Bilinear interpolation</p></li></ul><ul><li><p>CUBIC - Cubic convolution</p></li></ul><p>If you do not specify a resampling technique for a given input
56predictor raster, BILINEAR will be used.</p><p>The ArcGIS 9.2 documentation provides the following information about the
57resampling techniques:</p><ul><li><p>The NEAREST option, which performs a nearest neighbor assignment, is
58the fastest of the interpolation methods. It is primarily used for
59categorical data, such as a land use classification, because it will
60not change the cell values. Do not use NEAREST for continuous data,
61such as elevation surfaces.</p></li></ul><ul><li><p>The BILINEAR option, bilinear interpolation, determines the new
62value of a cell based on a weighted distance average of surrounding
63cells. The CUBIC option, cubic convolution, determines the new cell
64value by fitting a smooth curve through the surrounding points.
65These are most appropriate for continuous data and may cause some
66smoothing; also, cubic convolution may result in the output raster
67containing values outside the range of the input raster. It is not
68recommended that BILINEAR or CUBIC be used with categorical data
69because the cell values may be altered.</p></li></ul></td></tr><tr><td class="info">{ignoreOutOfRangeValues}</td><td class="info" align="left"><p>If True, predictions will not be made where the values of
70predictor rasters fall outside of the range of values used to fit the
71model. These cells will appear as NoData in the output rasters.</p><p>If False, predictions will be attempted regardless of what the
72predictor values are.</p><p>This parameter is set to True by default because many believe that it
73is a bad practice to extrapolate a model beyond the range of values
74used to fit it. But if your model provides a very strong fit, or you
75have some other reason to believe it is very robust, you can set this
76parameter to False to perform the extrapolation.</p></td></tr><tr><td class="info">{outputErrorRaster}</td><td class="info" align="left"><p>Output raster representing the standard errors of the predicted
77response.</p><p>If any pixel is NoData in any predictor raster (after it has been
78projected and clipped as needed), it will be NoData in the output
79raster as well. This is because the response variable cannot be
80predicted by the model if any of the predictor variables are
81missing.</p></td></tr><tr><td class="info">{outputBinaryResponseRaster}</td><td class="info" align="left"><p>Output raster with the predicted response classified into two
82possible values, 0 or 1, according to the cutoff parameter (which you
83must also provide). Predicted response values less than the cutoff
84will be classified as 0; values greater than or equal to the cutoff
85will be classified as 1.</p><p>If any pixel is NoData in any predictor raster (after it has been
86projected and clipped as needed), it will be NoData in the output
87raster as well. This is because the response variable cannot be
88predicted by the model if any of the predictor variables are
89missing.</p><p>You can examine the performance of your model for different cutoff
90values using the Plot ROC of Binary Classification Model tool or the
91Plot Performance of Binary Classification Model tool.</p></td></tr><tr><td class="info">{cutoff}</td><td class="info" align="left"><p>Cutoff for classifying the continous predicted response into a
92binary response. See the documentation for the Output Binary Response
93Raster for more information.</p></td></tr><tr><td class="info">{buildPyramids}</td><td class="info" align="left"><p>If True, pyramids will be built for the output rasters, which will
94improve their display speed in the ArcGIS user interface.</p></td></tr></tbody></table></div><p><h2><img width="11" height="11" border="0" src="sm_arrow_down.gif?format=raw" /> Scripting syntax</h2></p><div Class="expand" id="TEST">GAMPredictFromArcGISRasters_GeoEco (inputModelFile, inputPredictorRasters, variableNames, outputResponseRaster, templateRaster, resamplingTechniques, ignoreOutOfRangeValues, outputErrorRaster, outputBinaryResponseRaster, cutoff, buildPyramids) <br /><br /><b>Parameters</b><br /><table width="100%" border="0" cellpadding="5"><tbody><tr><th width="40%"><b>Expression</b></th><th width="60%"><b>Explanation</b></th></tr><tr><td class="info">Input model file (Required) </td><td class="info" align="left"><p>File that contains the fitted model, generated by the Fit GAM
95tool.</p></td></tr><tr><td class="info">Input predictor rasters (Required) </td><td class="info" align="left"><p>Rasters that represent the predictor variables used in the model.</p><p>This list must include one raster for each predictor variable that
96appears in the model's formula. The list provided for the Model
97Variable Names for Predictor Rasters parameter specifies the predictor
98variable that each raster represents.</p><p>For example, if your model used the formula:</p><dl><dt></dt><dd><pre>Presence ~ SST + Chl + Depth</pre></dd></dl><p>You must specify three rasters, one for each predictor variable. If
99you want to predict Presence for the month of January 1999, you might
100use a static bathymetry raster and SST and chlorophyll rasters for
101that month:</p><dl><dt></dt><dd><pre>C:\Data\etopo2v2
102C:\Data\SST\sst199901
103C:\Data\Chlorophyll\chl199901</pre></dd></dl><p>Then, for the Model Variable Names for Predictor Rasters parameter,
104you would specify the predictor variable that each raster represents:</p><dl><dt></dt><dd><pre>Depth
105SST
106Chl</pre></dd></dl><p>You can specify extra predictors that do not appear in the formula.
107They will be ignored. In the example above, you might include the
108variable SSH (sea surface height) and the raster
109C:\Data\SSH\ssh199901. Because this variable does not appear
110in the model formula, it will be ignored.</p><p>All of the predictor rasters must have a coordinate system defined.
111They must all share the same datum. If you also specify the Template
112Raster For Outputs parameter, they must have the datum of the template
113raster. If you do not specify that parameter, the first predictor
114raster will be used as the template raster.</p><p>The predictor rasters need not have the same coordinate system,
115extent, or cell size as the template raster. If these characteristics
116differ from the template raster, this tool will automatically project
117and clip the predictor rasters to conform to the template raster. The
118predictor rasters must be able to be projected to the template
119raster's coordinate system without requiring the specification of a
120geographic transformation. An error will be reported if a geographic
121transformation must be specified. In this case, you must project the
122predictor rasters manually before providing them to this tool.</p><p>By default, floating point predictor rasters will be projected using
123bilinear interpolation and integer predictor rasters will be projected
124using nearest neighbor assignment. Bilinear interpolation is
125appropriate for continuous variables such as sea surface temperature,
126while nearest neighbor is appropriate for categorical variables such
127as bottom substrate type. If these defaults are not appropriate for
128your predictors, specify different algorithms using the Resampling
129Techniques for Predictor Rasters parameter.</p></td></tr><tr><td class="info">Model variable names for predictor rasters (Required) </td><td class="info" align="left"><p>Model variable names for the input predictor rasters. The variable
130names are case sensitive. Please see the documentation for the Input
131Predictor Rasters parameter for more information.</p></td></tr><tr><td class="info">Output response raster (Required) </td><td class="info" align="left"><p>Output raster representing the response predicted by the model
132from the input predictor rasters.</p><p>If any pixel is NoData in any predictor raster (after it has been
133projected and clipped as needed), the pixel will be NoData in the
134output raster as well. This is because the response variable cannot be
135predicted by the model if any of the predictor variables are
136missing.</p></td></tr><tr><td class="info">Template raster for outputs (Optional) </td><td class="info" align="left"><p>Template raster that defines the coordinate system, extent, and
137cell size of the output rasters produced by this tool.</p><p>If you do not specify a template raster, the first predictor raster
138will be used instead.</p><p>All predictor rasters must share the same datum as the template
139raster. The predictor rasters need not have the same coordinate
140system, extent, or cell size as the template raster. If these
141characteristics differ from the template raster, this tool will
142automatically project and clip the predictor rasters to conform to the
143template raster. Please see the documentation for the Input Predictor
144Rasters parameter for more information.</p></td></tr><tr><td class="info">Resampling techniques (Optional) </td><td class="info" align="left"><p>Resampling technique to be used for each input predictor raster if
145that raster needs to be automatically projected, as described in the
146documentation for the Input Predictor Rasters parameter.</p><p>The available resampling techniques are:</p><ul><li><p>NEAREST - Nearest neighbor assignment</p></li></ul><ul><li><p>BILINEAR - Bilinear interpolation</p></li></ul><ul><li><p>CUBIC - Cubic convolution</p></li></ul><p>If you do not specify a resampling technique for a given input
147predictor raster, BILINEAR will be used.</p><p>The ArcGIS 9.2 documentation provides the following information about the
148resampling techniques:</p><ul><li><p>The NEAREST option, which performs a nearest neighbor assignment, is
149the fastest of the interpolation methods. It is primarily used for
150categorical data, such as a land use classification, because it will
151not change the cell values. Do not use NEAREST for continuous data,
152such as elevation surfaces.</p></li></ul><ul><li><p>The BILINEAR option, bilinear interpolation, determines the new
153value of a cell based on a weighted distance average of surrounding
154cells. The CUBIC option, cubic convolution, determines the new cell
155value by fitting a smooth curve through the surrounding points.
156These are most appropriate for continuous data and may cause some
157smoothing; also, cubic convolution may result in the output raster
158containing values outside the range of the input raster. It is not
159recommended that BILINEAR or CUBIC be used with categorical data
160because the cell values may be altered.</p></li></ul></td></tr><tr><td class="info">Ignore raster values outside the modeled range (Optional) </td><td class="info" align="left"><p>If True, predictions will not be made where the values of
161predictor rasters fall outside of the range of values used to fit the
162model. These cells will appear as NoData in the output rasters.</p><p>If False, predictions will be attempted regardless of what the
163predictor values are.</p><p>This parameter is set to True by default because many believe that it
164is a bad practice to extrapolate a model beyond the range of values
165used to fit it. But if your model provides a very strong fit, or you
166have some other reason to believe it is very robust, you can set this
167parameter to False to perform the extrapolation.</p></td></tr><tr><td class="info">Output standard error raster (Optional) </td><td class="info" align="left"><p>Output raster representing the standard errors of the predicted
168response.</p><p>If any pixel is NoData in any predictor raster (after it has been
169projected and clipped as needed), it will be NoData in the output
170raster as well. This is because the response variable cannot be
171predicted by the model if any of the predictor variables are
172missing.</p></td></tr><tr><td class="info">Output binary response raster (Optional) </td><td class="info" align="left"><p>Output raster with the predicted response classified into two
173possible values, 0 or 1, according to the cutoff parameter (which you
174must also provide). Predicted response values less than the cutoff
175will be classified as 0; values greater than or equal to the cutoff
176will be classified as 1.</p><p>If any pixel is NoData in any predictor raster (after it has been
177projected and clipped as needed), it will be NoData in the output
178raster as well. This is because the response variable cannot be
179predicted by the model if any of the predictor variables are
180missing.</p><p>You can examine the performance of your model for different cutoff
181values using the Plot ROC of Binary Classification Model tool or the
182Plot Performance of Binary Classification Model tool.</p></td></tr><tr><td class="info">Cutoff (Optional) </td><td class="info" align="left"><p>Cutoff for classifying the continous predicted response into a
183binary response. See the documentation for the Output Binary Response
184Raster for more information.</p></td></tr><tr><td class="info">Build pyramids (Optional) </td><td class="info" align="left"><p>If True, pyramids will be built for the output rasters, which will
185improve their display speed in the ArcGIS user interface.</p></td></tr></tbody></table></div></body></html>
Note: See TracBrowser for help on using the browser.