- THE QUESTION WAS:

We are studying deforestation in the State of Campeche (South east of

Mexico). Overlaying two forest maps of different dates, we generated

a digital map which represents the evolution of forest areas. Classes

of this digital map are non forested areas, unchanged forest areas

and deforested areas. As a following step, we tried to find out the

relationship between deforestation and environmental and

socioeconomic factors such as distance from settlements and

characteristics of the population.

Socioeconomic data we used are organized by localities (data are

given for each settlement), the problem is how to deal with both

point data and surface data within a GIS. In order to represent the

influence of a settlement on its environment we developed a simple

model which is based on two simplifying assumptions: the influence of

a settlement a) depends on its characteristics and b) decreases with

the distance from it.

In order to develop the model, localization of each settlements was

captured in a point cover, then the settlements point cover was

converted to a grid using. The value assigned to

the grid cells was the value of a selected property (for example

number of people of the settlement) according to the point of its

overlay. If no points fell within the cell, a zero value was

assigned. Then a low pass filter was applied to the grid. The size of

the filter and the coefficients used to weight the average towards

the pixels were derived from the function that described the

relationship between deforestation and distance from settlements

pointed out in a previous step. Several layers were generated

developing the model with different socioeconomic features. Values of

the model layers were then grouped into ten classes representing

similar area and rate of deforestation was calculated for each of

them. Coefficient of correlation between the values of the model and

the rates of deforestation were calculated. Analysis of the

coefficients of correlation allows to determinate the relative

influence of the different socioeconomic characteristics we

integrated in the model on deforestation rates.

I have some doubts about the model and the interpretation of the

results: a) linear correlation is not always the more appropriate to

estimate the relationship between the variables,is there a better way

to estimate relation between two variables ? b) the studied

socioeconomic variables presented correlation between them, there is a

way to separate them ? c) the variable the model deals with is a

combination of a sociological variable and the distance (because of

the filtering), can be the variable "distance" be eliminated to

analyze the pure socioeconomic variable ? Any suggestions or

references about how to deal with socioeconomic point data and modeling

influence of populated areas on its environment are welcome.

*************************************************************

THANKS YOU VERY MUCH TO:

Vic Rudis, Alessandro Gimona, Vincent Simonneaux and G Nelson for

their very interesting comments:

*************************************************************

A BRIEF SUMMARY OF THE RESPONSES:

a) Logistic regression is a good way to estimate relationship

between deforestation and exploratory variables. Stepwise log.

regression allows to estimate relationship with each of the

exploratory variables. it has been widely applied in species

distribution studies.

b) New decorrelated variables can be produced using an PCA but the

problem is that new variables are linear combination of the

socioeconomic variables. Therefore it can be less easy to interpret

them.

c) A way could be to subdivide a variable into several variable: e.g.

subdivide the cities into several groups (large size, small size) and

study separately the variable "distance from one attribute" (cities of a given size

for example).

*********************************************************

I INCLUDE THE RESPONSES

Date sent: Thu, 23 Jan 1997 12:19:56 -0600

To: jeanfmas@...

From: Gerald Nelson <g-nelson@...>

Subject: socioecon factors

Saw your note on the the geostats list and have a couple of

references, two on Mexico and one on Belize, that might be of

interest. I'd also like to stay in touch because what you are doing

sounds like it is of great relevance for work I'm doing. Anyway, the

references:

Gerald Nelson and Daniel Hellerstein, "Do Roads Cause Deforestation?

Using Satellite Images in Econometric Analysis of Land Use",

forthcoming, American Journal of Agricultural Economics (with Daniel

Hellerstein).

Gerald Nelson and Daniel Hellerstein, "Deforestation in Central

Mexico: Satellite Images as Data in Economic Models of Land Use ",

forthcoming, Pecora 13 Conference Proceedings.

You can get these from the web as pdf docs at

http://w3.aces.uiuc.edu/ACE/faculty/GNelson/papers/paperindex.htm

Chomitz, K. M., & Gray, D. A. (1995). Roads, Land, Markets and

Deforestation: A Spatial Model of Land Use in Belize working paper,

World Bank Policy Research Department, Environment Infrastructure, and

Agriculture Division. Also published in World Bank Research Observer

in summer 1996.

A working paper by Klaus Deininger, also from the World Bank that

incorporates population, wage rates, into a model that is similar to

that used by the papers above.

Let me know if you have any trouble getting these.

By the way, I bounced into your web site. It looks interesting,

although the English version wasn't working and my Spanish is a bit

weak.

Regards, Jerry Nelson

Gerald C. Nelson

Assoc. Professor, Department of Agricultural and Consumer Economics

University of Illinois

g-nelson@...

217-333-6465

Date sent: Tue, 21 Jan 1997 15:09:08 -0600 (CST)

From: Vic Rudis <vrudis@...>

To: jeanfmas@...

Subject: Re: GEOSTATS: assessing socioeconomic factors

influence in deforestation

> Date: Tue, 21 Jan 1997 10:20:50 +1100

You did not say what type of correlations were done.

> From: jeanfmas@...

> To: ai-geostats@...

> Subject: GEOSTATS: assessing socioeconomic factors influence in

> deforestation

>

> I have some doubts about the model and the interpretation of the

> results: a) linear correlation is not always the more appropriate to

> estimate the relationship between the variables,is there a better

> way to estimate relation between two variables ? b) the studied

> socioeconomic variables presented correlation between them, there is

> a way to separate them ? c) the variable the model deals with is a

> combination of a sociological variable and the distance (because of

> the filtering), can be the variable "distance" be eliminated to

> analyze the pure socioeconomic variable ? Jean-Francois Mas

> Laboratorio de Percepcion remota y SIG Programa EPOMEX Universidad

> Autonoma de Campeche

1a. What you describe can be handled by logistic regression and

point data. What the relationship shows is the "probability" of an

association, rather than a linear relationship, between two variables.

Equations predict point probabilities: 0o0c, 0o1c, 1o0c, and 1o1c,

where 0=nonforested and 1=forested, and o=last time, and c=this time.

1b. I used stepwise logistic regression and standardized coefficients,

but there are other approaches. Standardized coefficients provide one

with the "relative" importance of variables in the equations. See my

paper in Landscape Ecology 10(5): 291-307, 1995, as one example of

this approach. The article's model indicated those variables that best

predicted forest fragment size class for south central U.S. bottomland

hardwood forests. Proximity to agricultural land was the primary

correlate, and that variable dominated the equations. Degree of nearby

road development was a secondary predictor among small fragment

classes and proximity to urban land was secondary among large fragment

classes.

1c. I don't know. One variable such as city population size could be

subdivided into several variables: e.g., cities of large size, cities

of small size. Then, distance from one attribute (cities of a given

size class) to selected forest or nonforest points serve as potential

variables for entry into a stepwise logistic regression equation.

_ _

/s/ Vic Rudis 1o,o1

|Personal Page "http://www2.msstate.edu/~vrudis/index.html"

| |U.S. forest ecoinventory

"http://www.msstate.edu/Dept/Forestry/ecofia.html"| |U.S. forest

survey (FIA) "http://www.srsfia.usfs.msstate.edu/" |

|Other regional surveys

"http://www2.msstate.edu/~vrudis/fiaaffinity.html"|

\------------ Opinions expressed are mine and not my

employer's.----------/>

From: geo293@...

Subject: Re: GEOSTATS: assessing socioeconomic factors

influence in deforestation To: jeanfmas@...

Date sent: Tue, 21 Jan 1997 19:11:57 +0000 (GMT)

>

(I am not entirely clear as to all the steps especially

> We are studying deforestation in the State of Campeche (South east

> of

> Mexico). Overlaying two forest maps of different dates, we

> generated

> a digital map which represents the evolution of forest areas.

> Classes of this digital map are non forested areas, unchanged

> forest areas and deforested areas. As a following step, we tried to

> find out the relationship between deforestation and environmental

> and socioeconomic factors such as distance from settlements and

> characteristics of the population.

>

> Socioeconomic data we used are organized by localities (data are

> given for each settlement), the problem is how to deal with both

> point data and surface data within a GIS. In order to represent the

> influence of a settlement on its environment we developed a simple

> model which is based on two simplifying assumptions: the influence

> of a settlement a) depends on its characteristics and b) decreases

> with the distance from it.

>

> In order to develop the model, localization of each settlements was

> captured in a point cover, then the settlements point cover was

> converted to a grid using. The value assigned to the grid cells was

> the value of a selected property (for example number of people of

> the settlement) according to the point of its overlay. If no points

> fell within the cell, a zero value was assigned. Then a low pass

> filter was applied to the grid. The size of the filter and the

> coefficients used to weight the average towards the pixels were

> derived from the function that described the relationship between

> deforestation and distance from settlements pointed out in a

> previous step. Several layers were generated developing the model

> with different socioeconomic features. Values of the model layers

> were then grouped into ten classes representing similar area and

> rate of deforestation was calculated for each of them. Coefficient

> of correlation between the values of the model and the rates of

> deforestation were calculated. Analysis of the coefficients of

> correlation allows to determinate the relative influence of the

> different socioeconomic characteristics we integrated in the model

> on deforestation rates.

>

> I have some doubts about the model and the interpretation of the

> results: a) linear correlation is not always the more appropriate to

> estimate the relationship between the variables,is there a better

> way to estimate relation between two variables ?

the filtering one, so apologies if this does not

make sense).

you could try ordinal logistic regression

b) the studied> socioeconomic variables presented correlation between them, there is

You could use a stepwise technique and/or run

> a way to separate them ?

Principal component analysis on the data array and use the PC scores

as new input in step a)

c) the variable the model deals with is a

> combination of a sociological variable and the distance (because of

Sorry, I don't know about this point.

> the filtering), can be the variable "distance" be eliminated to

> analyze the pure socioeconomic variable ? Any suggestions or

> references about how to deal with socioeconomic point data and

> modeling

> influence of populated areas on its environment are welcome.

>

A last note: if the data present spatial autocorrelation in the

dependent variable you might have to use a spatial simulation approach

to test for significance.

'HOPE IT HELPS

Alessandro gimona

MLURI

Aberdeen

Scotland

Date sent: Fri, 24 Jan 1997 18:39:19 +0100

From: simonne@... (Vincent

Simonneaux) To: jeanfmas@... Subject:

Re: GEOSTATS: assessing socioeconomic factors influence in

deforestation

Bonjour,

Je suppose que tu comprends le francais.

Le resume de cette question m'interessera car nous aurons bientot a

etudier egalement des Pb semblables Pour ce qui est de tes questions,

quelques reflexions

a) il existe effectivement d'autres methodes d'etude des correlations

qui sont non parametriques (ex: coeff de coor. de rang de spearman,

"statistique S", Khi2 entre deux distributions). (Le coeff de corr

classique fait l'hypo gaussienne) C'est explique dans tout manuel de

stats basique

b) Tu ne peux supprimer les correlations entre variables qu'en en

creant d'autres non correlees (ex: Analyse en composantes principales

pour les variables quantitatives, mais difficile pour les

qualitatives)

c) Je ne comprends pas bien ou est le Pb. La variable socio "pure"

n'existe pas il faut bien que tu la localises dans l'espace (?). (Et

dans ce cas, si la pression socio est liee a la distance et la

deforestation a la pression, alors tu ne pourras pas empecher la

deforestation d'etre liee a la distance.

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

------- Vincent Simonneaux ORSTOM (Institut FranTauais de Recherche

Scient. pour le Developpement en Cooperation) 32 Avenue Henri Varagnat

93143 BONDY Cedex Tel: (1) 48 02 55 77 / Fax: (1) 48 47 30 88 /

Email: simonneaux@...

______________________________________________________________________

_______

*************************************************

Jean-Francois Mas

Laboratorio de Percepcion Remota y SIG

Centro EPOMEX

Universidad Autonoma de Campeche

AP 520 CP 24030 CAMPECHE, CAMP, MEXICO

Tel (52) (981) 11600

Fax (52) (981) 65954

E-mail jeanfmas@...

WWW : epomex.uacam.mx

*************************************************

--

*To post a message to the list, send it to ai-geostats@....

*As a general service to list users, please remember to post a summary

of any useful responses to your questions.

*To unsubscribe, send email to majordomo@... with no subject and

"unsubscribe ai-geostats" in the message body.

DO NOT SEND Subscribe/Unsubscribe requests to the list! - a] ... for starters.. you can rank the 2, i.e. your model and actual

data, and check if the 2 ranks correspond... positively or negatively..

if there is a rank correlation then that suggets that there is some other

kind of correlation.. after which you can try linear, quadratic etc.

Another good thing wouold be to visually look at tye graphs.. that will

give you some idea about whether there is some sort of relationship

between the wto, or your points are merely randomly distributed.

b].....do a principal components analysis on your socioeconomic

variables.. and repeat the correlation studies on the 1st, 2nd etc

components. If you get significant results, then look at each component

and see what it is composed of.. i.e. how much each socioeconomic

variable contributes to this component.. of course, this approach has

problems which have already been pointed out..

c].....if you want to eliminate distance.. I guess one thing you can do is,

treat the variable as having equal influence over some fixed distance..

and after that no influence.. what this distance is is something you will

have to decide on based on your "feel" for the data..

Hope this helps..

Harini Nagendra

Centre for Ecological Sciences

Indian Institute of Science>

--

> I have some doubts about the model and the interpretation of the

> results: a) linear correlation is not always the more appropriate to

> estimate the relationship between the variables,is there a better way

> to estimate relation between two variables ? b) the studied

> socioeconomic variables presented correlation between them, there is a

> way to separate them ? c) the variable the model deals with is a

> combination of a sociological variable and the distance (because of

> the filtering), can be the variable "distance" be eliminated to

> analyze the pure socioeconomic variable ? Any suggestions or

> references about how to deal with socioeconomic point data and modeling

> influence of populated areas on its environment are welcome.

>

*To post a message to the list, send it to ai-geostats@....

*As a general service to list users, please remember to post a summary

of any useful responses to your questions.

*To unsubscribe, send email to majordomo@... with no subject and

"unsubscribe ai-geostats" in the message body.

DO NOT SEND Subscribe/Unsubscribe requests to the list!