This new thread is just intended to add some info about "prospectivity analysis" and "weights of evidence" - mentioned in passing in my wide-ranging comments in the thread about databases. I was asked to amplify on a couple of things by the originator of the thread.

I'm by no means an expert.

What follows is very sketchy - I must recommend that a couple of the PDF files referred to be downloaded and read, but better - that those who are intrested might use Google (or your favourite search engine) to find out more. I give some search parameters below.

The ESRI (and MapInfo) bias is because I don't know much more about other GIS software that uses the technique. The geological bias is because that's my original training / career.

Weights of Evidence

--------------------------

Weights of evidence for mineral prosperity

* The WofE methodology was developed as a medical diagnostic tool

* Look for spatial correlations between themes - prior probabilities

* Generate aggregated themes using weights from post probabilities

Suggested search terms

-------------------------------

Google search - bonham-carter "weights of evidence" filetype:pdf (51)

Google search - geology "weights-of-evidence" (329)

Google search - ESRI "weights-of-evidence" (84)

Google search - geology "weights of evidence" filetype:pdf (101)

ESRI - WofE.pdf (and others)

------------------------------------

http://www.esri.com/news/arcuser/0400/files/wofe.pdf

Predictive Probabilistic Modeling Using ArcView GIS

By Gary L. Raines, Graeme F. BonhamûCarter, and Laura Kemp

ArcUser, AprilùJune 2000 (www.esri.com)

This is a part of the introductory material in this short article.

---

Weights-of-evidence methodology combines spatial data from diverse sources to describe and analyze interactions, provide support for decision makers, and make predictive models. The method was originally developed for a nonspatial application in medical diagnosis. In this application, the evidence consisted of a set of symptoms, and the hypothesis was "this patient has disease x." For each symptom, a pair of weights was calculated, one for presence of the symptom and one for absence of the symptom. The magnitude of the weights depended on the measured association between the symptom and the occurrence of disease in a large group of patients. The weights could then be used to estimate the probability that a new patient would get the disease, based on the presence or absence of symptoms.

The weights-of-evidence methodology was adapted for mineral potential mapping with GIS. The Arc-WofE extension uses the statistical association between a training points theme such as a theme showing mineral sites and evidential themes showing rock types, geochemistry, or geophysical measurements to determine the weights.

The weights-of-evidence method is based on the application of BayesÆ Rule of probability, with an assumption of conditional independence. The model is stated in loglinear form so the weights from the evidential themes can be added. The Arc-WofE extension uses the weights-of-evidence methodology to produce a response theme. The response theme is an output map that combines the weights of predictor variables from the evidential themes to express the probability that a unit cell (small unit of area) will contain a training point. Evidential themes may have categorical values (e.g., the classes on geology or soil maps) or ordered values (e.g., geochemical concentrations or distance to linear and other spatial objects).

Weight values are easy to interpret. A positive weight for a particular evidential-theme value indicates that more training points occur on that theme than would occur due to chance, whereas the converse is true for negative weights.

A weight of zero indicates that the training points are spatially uncorrelated to the theme.

The range-in-weight values for a particular evidential theme, known as the contrast, gives an overall measure of how important the theme is in the model. Uncertainties due to variances of weights and missing data allow the relative uncertainty in posterior probability to be estimated and mapped.

Because conditional independence is never satisfied completely, the posterior probabilities are usually overestimated in absolute terms. However, the relative variations in posterior probability (as seen in spatial patterns on the response map) are usually not much affected by violations of this assumption.

---

Gary L. Raines is a research geologist for the U.S. Geological Survey in Reno, Nevada. His research focuses on the integration of geoscience information for predictive modeling in mineral resource and environmental applications.

Graeme F. BonhamûCarter is a research geologist working in the Mineral Deposits Division of the Geological Survey of Canada, Ottawa. He is interested in applications of GIS to mineral exploration and environmental problems. He is editor-in-chief of Computers & Geosciences, a journal devoted to all aspects of computing in the geosciences.

Laura Kemp is an Ottawa, Ontario, based GIS technologist specializing in ArcView GIS application development.

---

http://gis.esri.com/library/userconf/proc00/professional/abstracts/a414.htm

Using Weights of Evidence (WofE) in a GIS for the Prediction of Tuberculosis Cases: Application of a New ArcView Extension

http://www.esri.com/news/arcuser/0400/files/wofewpm.pdf

Pergamon, Oxford, 1990. Agterberg, FP, GF BonhamûCarter. Logistic Regression and Weights of Evidence Modelling in Mineral Exploration

http://www.nijpcs.org/CMRC/CMRC2002/Final_Agenda.pdf

... Fraser Moffatt Ottawa Police Service Determination of Input Variables and Their

Parameters for the Use of the Weights-of-Evidence Model for the Spatial ...

http://www.nijpcs.org/CMRC/CMRC2000/abs2000.pdf

... This Access database has an ESRI MapObjects component that geocodes the case ... Using Weights-of-Evidence Modeling as a Tool for the Spatial Prediction of Crime ...

Bonham-Carter

-------------------

http://geofocus.rediris.es/docPDF/Recurso3_2003.pdf

Agterberg, FP Bonham-Carter, GF (1990): "Deriving weights of evidence from geoscience contour maps for the prediction of discrete events", Geological ...

(in Spanish - references in English)

USGS - of01-501.pdf (and others)

------------------------------------------

http://geopubs.wr.usgs.gov/open-file/of01-291/

Mineral Potential Modelling of Gold and Silver Mineralization in the Nevada Great Basin -- A GIS-Based Analysis Using Weights of Evidence

http://geopubs.wr.usgs.gov/open-file/of01-501/of01-501.pdf

Assessment Method for Epithermal Gold Deposits in Northeast Washington State using Weights-of-Evidence GIS Modeling

Some other articles (available for download as PDF) give the mathematical background and the necessary algorithms for implementation.

Implementation in Manifold is of course quite possible.

:-)

Ian,

Thanks for providing so many references to look at. Reading through your references will expend a considerable amount of intellectual capital. So, before I do, I was curious: how different is this from logistic regression? It appears to be the same thing. Might you explain alittle more?

If it is similar to logistic regression, then I wonder if the discussions people are having regarding the integration of 'R' with Manifold might be valuable. Therefore, we could allow a real statistical tool to perform the statistical work, and just use Manifold as a front end tool to feed 'R' the data.

Thanks,

Art

artlembo - 2004-07-06 12:27 PM

...how different is this from logistic regression? It appears to be the same thing. Might you explain alittle more?

If it is similar to logistic regression, then I wonder if the discussions people are having regarding the integration of 'R' with Manifold might be valuable. Therefore, we could allow a real statistical tool to perform the statistical work, and just use Manifold as a front end tool to feed 'R' the data.

It's just one of several knowledge-driven and / or data-driven analysis methods to arrive at similar results (it's not anything like an exact science - the best analogy is that of overlaying a series of coloured, transparent map sheets and eyeballing the coincidence).

The best paper amongst the ones I referenced is that by Hongmei Wang + Gouray Cai + Qiuming Cheng (PennState and York U, Canada) "Data integration using weights of evidence model: Applications in mapping mineral resource potentials" - a paper from a symposium in Ottawa in 2002.

For the benefit of browsers who don't want to download and read the PDF, I quote this small section -

"Data-driven models include Logistic Regression, Weights of Evidence, Neural Networks, and so on, and the weights in those models are calculated from training models. While, the weights are estimated based on experts' opinions in knowledge-driven models. The knowledge-driven models include Fuzzy Logic, Demster-Schafer Belief Theory and their weights are given with expert opinion."

There's a lot more there that is quite readable and general, but later in the paper (just 6 pages long) they explain and present the probability framework that was used by Bonham-Carter based on set theory - ie, for the Weights of Evidnece (WofE) method.

A series of equations is presented, and the application to mineral prospecting follows.

My advice would be to skim over the parts that are a bit unfamiliar, but in particular read some of the discussion and conclusion. Why? Because there are several different approaches with WofE itself, and then there are the "other" methods in data- and knowledge-driven approaches.

So, certainly the use of 'external tools' like R is to be encouraged (no sense reinventing or implementing internally within manifold, if a 'tool' is controllable from a script or program that is 'driving' Manifold and the external tools).

I hope that's what you were after.

thanks to gxdata for this huge amount of information and the latest reply to my database question. Probabaly it prevented me from following the wrong path.

Initially I was attracted by the weighting of layers in the prospectivity analysis area.

My intended use is more in the land-use suitability field.

In both area weigthed judgements are necessary. So I had a first look on the SDM tool you mentioned and got the impression that rasterization of polygon-layer information is an important part of this approach (beside buffering etc.). This would be what I expected in my latest database question.

Is this a field, where Manifold can be compared with other tools at present ?

gutenberg - 2004-07-07 2:29 AM

thanks to gxdata for this huge amount of information and the latest reply to my database question. Probabaly it prevented me from following the wrong path.

Initially I was attracted by the weighting of layers in the prospectivity analysis area.

My intended use is more in the land-use suitability field.

In both area weigthed judgements are necessary. So I had a first look on the SDM tool you mentioned and got the impression that rasterization of polygon-layer information is an important part of this approach (beside buffering etc.). This would be what I expected in my latest database question.

Is this a field, where Manifold can be compared with other tools at present ?

I'm sorry, I'm not sure what you mean by -

Probabaly it prevented me from following the wrong path

and also by -

Is this a field, where Manifold can be compared with other tools at present ?

Can you explain a bit more, please? Someone else here may be able to give a clearer answer than I have - perhaps I have given too much non-essential information.

gxdata - 2004-07-06 10:27 PM

The best paper amongst the ones I referenced is that by Hongmei Wang + Gouray Cai + Qiuming Cheng (PennState and York U, Canada) "Data integration using weights of evidence model: Applications in mapping mineral resource potentials" - a paper from a symposium in Ottawa in 2002.

Actually, I didn't refer to the Wang-Cai-Cheng paper in my original post.

Here it is (PDF download, 330K)

http://www.isprs.org/commission4/proceedings/pdfpapers/268.pdf

Also, Bonham-Carter's book -

Graeme Bonham-Carter, 1994: Geographic information systems for geoscientists: modelling with GIS. 398pp. Publ. Pergamon (Elsevier). 1st edition 1994, reprinted 1996.

This is an excellent introduction to WofE and puts the technique in context with other similar analytical techniques. The worked examples with very accessible presentation of the statistical techniques and the easily-implemented map algebra pseudo-language should serve to de-mystify this topic, and take it out of the realms of 'black box' technology.

There are many journal papers in which Weights of Evidence is explained in simple terms, or detailed case histories are presented, or very specific maths and algorithms are presented.

I'm sorry, I'm not sure what you mean by -

Probabaly it prevented me from following the wrong path

and also by -

Is this a field, where Manifold can be compared with other tools at present ?

Can you explain a bit more, please? Someone else here may be able to give a clearer answer than I have - perhaps I have given too much non-essential information.

Of course I can explain my intention an bit more.

Perhaps I will start with some general things. The background of my questions is, that I'm looking for a suitable GIS tool for doing some land suitability stuff. After having used MI, a little bit Manifold (until v4.5), a little bit AV several years (I had a look on several extensions to MI and AV) I dropped GIS. Now I'm returning to GIS and the question is on what software to rely

in respect to land suitablity.

I'd like to do this with manifold but I'm not sure if I take into account the 5% problem

(that means that 95% of the tasks could be performed in Manifold, but perhaps the last but sometimes important 5% cannot and I would have to switch to another software).

I regard the database capabilities as an important GIS especially as the analytical capabilites are closely related to this (in my understanding).

After one of your replies I expected, that weighting layers is connected with MapAlgebra

(that means the application of simple mathematical expressions on several layers). In this case it would be necessary to use a GIS that have some kind of MapAlgebra included. You said MapAlgebra isn't included - so it prevented me form a wrong path !

Another assumption was that I would have to rely to some extent on raster grid processing capabilities (in addition to vector analysis) in GIS. I took your hint to MI-SDM as a prove of this assumption because I found a lot of data preparation regarding conversion from polygons to raster layers in the MI-SDM tool.

That's why I asked for a comparison regarding raster processing capabilities in Manifold.

Of the same importance is the modeling capability itself, a field that isn't covered by Manifold at present (correct me if I'm wrong).

Now I'm on the way to analyze the current state-of-the-art regarding vector analysis, as I found that there are obviously some special overlay techniques (simple limitation factor etc.)

Hope that these explanations are better to understand.

gutenberg - 2004-07-08 2:42 AM

Perhaps I will start with some general things. The background of my questions is, that I'm looking for a suitable GIS tool for doing some land suitability stuff. After having used MI, a little bit Manifold (until v4.5), a little bit AV several years (I had a look on several extensions to MI and AV) I dropped GIS. Now I'm returning to GIS and the question is on what software to rely in respect to land suitablity.

I'd like to do this with manifold but I'm not sure if I take into account the 5% problem

(that means that 95% of the tasks could be performed in Manifold, but perhaps the last but sometimes important 5% cannot and I would have to switch to another software).

I think from your remarks, it's the 5% of 'other' things that many GIS cannot do that has led you to use the widely-available statistical and analytical tools (that are appropriate to your land suitability problem), and to drop the spatial analysis (ie, the GIS).

Of course it is possible to combine multivariate statistics (or other more sophisticated statistical techniques that you would have been applying to your data), and your GIS (ie, spatial analysis).

I believe that some of the forum members actually do that. But (someone will correct me if I am wrong) it does require some effort and perhaps some reasonably developed ability to script, in order to do that with Manifold. I think simple statistics are straightforward though.

gutenberg - 2004-07-08 2:42 AM

I regard the database capabilities as an important GIS especially as the analytical capabilites are closely related to this (in my understanding).

Having your data including the non-spatial information conveniently packaged and separated in a database is (in my opinion) always a good idea, and Manifold 'connects' quite well to databases. Also, its ability to query )especially 'spatially query') information linked or imported from the external database is qood.

One should not down-play the importance of spatial analysis and queries - it is often very hard to get a similar understanding of distribution patterns without using GIS; and a lot of misleading conclusions may be reached by just considering the non-spatial data. I'm sure you know that.

gutenberg - 2004-07-08 2:42 AM

After one of your replies I expected, that weighting layers is connected with MapAlgebra

(that means the application of simple mathematical expressions on several layers). In this case it would be necessary to use a GIS that have some kind of MapAlgebra included. You said MapAlgebra isn't included - so it prevented me form a wrong path !

Approximately correct. I was generalizing and simplifying. There is still a lot of effort to doing a data-driven or knowledge-driven analysis using WofE, Logistic Regression, Fuzzy Logic, Neural Networks, "etc". The 'hard' bit is not the simple 'map algebra' or the simple statistics, it's the number of steps to prepare data, prepare maps in the right form, assess the suitability or success, then produce appropriate output. (Again, I am generalizing and simplifying).

gutenberg - 2004-07-08 2:42 AM

Another assumption was that I would have to rely to some extent on raster grid processing capabilities (in addition to vector analysis) in GIS. I took your hint to MI-SDM as a prove of this assumption because I found a lot of data preparation regarding conversion from polygons to raster layers in the MI-SDM tool.

That's why I asked for a comparison regarding raster processing capabilities in Manifold.

No, that's not correct. Although pixel-by-pixel comparison of raster maps may be done, most WofE that I have seen or that I have been involved with uses vector source maps. The Bayesian approach used in WofE depends on a training set of 'knowns', and also depends on simply classifying the mapped polygons on a simple scale (eg, 0 or 1 = binary, or 0 to 3 or 0 to 5, etc). In order to understand this properly, I would need to show you some simple diagrams and tables of numbers - it is best to have a look at Chapter 9 of Bonham-Carter's book and follow through the method. It is quite easy to understand (I'm not kidding you).

gutenberg - 2004-07-08 2:42 AM

Of the same importance is the modeling capability itself, a field that isn't covered by Manifold at present (correct me if I'm wrong).

Yes, you are correct.

gutenberg - 2004-07-08 2:42 AM

Now I'm on the way to analyze the current state-of-the-art regarding vector analysis, as I found that there are obviously some special overlay techniques (simple limitation factor etc.)

As I have said above, 'vector analysis' is what the "Weights of Evidence ETCETERA" is all about - rasters can be treated, but are often an inconvenience.