GIS Methodologies for Developing Conservation Strategies Part 6 potx

The relative weights PRki,j are also normalized to 1, with theresult that, on average weighted by the areas covered by each threat, PRki,j is 1.Hence, multiplying PRki,j by PAk will prod

mentioned earlier The relative weights PRk(i,j) are also normalized to 1, with theresult that, on average (weighted by the areas covered by each threat), PRk(i,j) is 1.Hence, multiplying PRk(i,j) by PAk will produce the threat Ak(i,j) for each cell, thatwill average over the entire area to a value equal to PAk The methodology for thenormalization is described below.

Normalizing the Parameters

Given initial values for weights P(i,j) for a given threat k, the normalized values

PR(i,j), or relative weight, is given by:

PR(i,j)⳱P(i,j)

<P> Equation (3)where <P> is the weighted average of the P(i,j) within the map We will nowexplain how to compute <P> for different spatial distributions

Factors Covering a Well-Defined Area This is the simplest case that applies toslope classes and aquifers Once a weight Pl is assigned to each class l within athreat k, a scale factor <P> is calculated from the area occupied by each class lrelative to the total area of interest Given the set of class weight Pl, each having

an area Sl, we obtain:

ST⳱∑l

Sl Equation (4)and

<P>⳱∑

l

Pl(i,j)Sl

ST Equation (5)where P1(i,j) is the weight of each class l at cell (i,j) Within IDRISI, the areas Slare computed with area

Factors That Vary Continuously in Space This is the case for populationdensity and threat from distance to roads The problem can be illustrated by asimple example Let us consider an infestation by insects at a given point of theACCVC At a given time, the density of insects will decrease with the distancefrom the infestation nucleus Insects could be found at any place in the area butare scarce far from the nucleus If we want to use equation 5, we have todetermine the area of influence, ST, of the insect infestation If we consider all theACCVC (the area of interest) as the area affected by insects, the scaling factor

<P> as in equation 5 will be very small since the insects are concentrated nearthe infestation nucleus (i.e., Slis small) On the other hand, if we set a threshold

114 Leclerc and Rodriguez

to define ST, saying that if the density of insects is lower than x percent it isconsidered negligible, then the value that STwill take will depend on the choice

of x

Therefore, there is no well-defined area to weight a distribution that variesslowly in space, and the task of normalizing this distribution is by no meanssimple There is, however, a way to use equation 3 with such distribution For acontinuous distribution, equation 5 is represented by a volume integral:

<P>⳱ 兰AOI

P(x,y)dS

ST Equation (6)

where AOI denotes the area of interest of surface ST Such an integral can beevaluated numerically if the distribution is discretized In GIS, discretization isdone by reclassifying the continuous map in a large number of discrete classes.This is already done if the map is represented by integer numbers For these newclasses, one can calculate the average value of the threat (obtaining the Pl), andtheir areas (Sl), and compute <P> with equation 5 With IDRISI this is done withthe modules reclass (to obtain the discrete map from the continuous map),extract (to extract the average value of the continuous map within each discreteclass), and area (to obtain the area of each discrete class) Then one can compute

<P> with equation 5

Details of the Threats

Roads and Trails In Latin America, building a new public road results in thecolonization of the area within a short time along the road The natural tendency

is to develop agricultural and cattle production and wood extraction simplybecause it is easier to bring the products to market It also results in an increase

in the standard of living in the region and increased demand for products

We digitized roads from 1:50,000 scale maps and made the assumption thatthe pressure of the human activity along them is decreasing exponentially withthe distance to the road All types of roads (from highway to gravel) have beengiven the same weight The choice of the exponential comes from the hypothesisthat the probability of penetrating a given distance in the forest (and therefore todeforest at that point) is a constant Consequently, the probability P(dⳭ ⌬d) ofgoing from a point at d to a point at (dⳭ ⌬d) within the forest is proportional tothe probability P(d) of being at d (i.e., having logged up to that point)

P(dⳭ⌬d)⳱P(d)(1ⳮ⌬d

d0) Equation (7)where d0is a constant called the characteristic distance Therefore:

P(d)⳱expⳮd

d0 Equation (8)

Hence the probability of entering the forest up to a distance d0 is P(d0)⳱ 1 / e

⳱ 0.37 This function (figure 9.3) applied to roads is shown in the map in figure9.4 We estimated that a characterestic distance of 1 km was a reasonable figurefor the area

Forest Management Plans The forest management plans in the ACCVC proved by the DGF after 1989 have been mapped at a 1:50,000 scale, whichprovides an overview of where the logging activity was concentrated between

ap-1986 and 1992 Some of the plans have already been executed within the BraulioCarrillo protected area, as a consequence of the private ownership of these lands

In these areas the loggers (principally locals) have contracts with the owners ofthe land covering a period of from one to five years The management plans have

to be approved by DGF This is a long and tedious process which involves severalrevisions, but approval is eventually conceded

To simulate the pressure that the logging activity can bring to an area, weselected a buffer zone of 1 km around the actual management plans and gave it

a weight of 1 The management plans supervised or approved by FUNDECORhave been given a weight of 0, which implies that these plans do not present any

F IG. 9.3 Graph of the threat due to the proximity to a road in function of the tance of the latter, following equation 8 The arrow shows the value of the functionfor the characteristic distance d0 (1000 m)

dis-116 Leclerc and Rodriguez

risk to the environment (figure 9.5) Adjacent to these areas, however, COR has a responsibility because the logging activities, although well managed,imply road construction which opens the way to uncontrolled logging in thebuffer zone

FUNDE-Population Density The forest and other natural resources that are close topopulation centers are under constant pressure More families will need moreland for construction and agriculture, and will represent a larger threat to theforest In addition, water pollution is likely to increase closer to the villages.Locating population centers and estimating the population density will helpfocus efforts where human activity is greatest

Since 1945 Costa Rica has put tremendous effort and money toward ing education, and as a consequence it has a lower rate of illiteracy than theUnited States As soon as there are more than six children in an area, a school isbuilt and a teacher assigned Numbers of schools allowed us to generate a betterestimate of population density than from the villages that appear on the outdatedmaps The schools, digitized as points from the 1:50,000 scale maps of theMinistry of Education, were assigned a weight equal to the number of registeredstudents Then, by applying successive passes of a 3 ⳯ 3 average filter, wegenerated a Gaussian distribution centered on the school (which representsstudent density) and converted this to population density by knowing the num-ber of family members and students per family Here, however, we are onlyinterested in the normalized distribution

improv-F IG. 9.4 Map of the normalized threat due to proximity to roads (1000 m tic distance, exponential model)

characteris-In the remote areas of the ACCVC the children will walk a maximum of 2.5

km to go to school (those that have to walk a longer distance will eventuallyhave a school built closer to their homes) Knowing this fact, we can estimate theextent of the population distribution The average filter that IDRISI provides(filter) computes, for each cell (i,j), the average of the values of its nearestneighbors Given E(i,j), the registration for a school located at cell (i,j), the valueafter 1 pass of the filter will be

E(i,j)⳱1

9k⳱iⳭl∑k⳱iⳮl l⳱jⳭl∑l⳱jⳮlE(k,l) Equation (9)

Each time the filtering operation is done on the map resulting from thepreceding filter, the population distribution becomes more Gaussian, is flattened,and extends radially On a raster map with a square pixel of 100 m, one hundredsuccessive passes of the average filter produces a distribution with a width athalf height RE/E0⳱50%of about 1,000 m and a maximum radius of approximately2.5 km This distribution represents the density of students per hectare (since thepixel is one hectare) The normalized map resulting from this operation is shown

in figure 9.6 Table 9.2 shows how the distribution changes with the number ofconsecutive passes of the average filter In this table, E/E0represents the height

of the center of the distribution (for example, the height of the distribution afterone hundred passes is 0.0024 of the original height E, which is the number of

F IG. 9.5 Map of the normalized threat due to forest management plans (1 km bufferzone)

118 Leclerc and Rodriguez

registered students) Rmax is the value where the distribution is close to zero,although the absolute zero is reached at a distance equal to the number of passestimes the size of the pixel (i.e., for one hundred passes, 100⳯ 100 m ⳱ 10 km;see figure 9.7)

IDA Land Distribution As an important criterion to determine a potentialthreat to natural resources, FUNDECOR considered the proximity occasioned bythe IDA’s land repatriation plan In fact, IDA has given extensive areas of forestfree to families in order to promote agriculture and development A way to add

F IG. 9.6 Map of the normalized population density (one hundred passes of a 3 x 3average filter) The map was produced starting from a point coverage of the schools,where the value of the school is equal to the number of registered students

T ABLE 9.2 Distribution Changes Produced by Passes of a 3 ⳯3

note: Effect of consecutive passes of the filter in the IDRISI program (filter) on

the width of the distribution at 50 percent of maximum height, its maximum

width (1 percent of maximum height), and the height of the distribution.

value to the land, however, is to cut down the forest To simulate the fact thatnew IDA colonies can be located in the neighborhood of existing colonies, weconsidered a simple 1 km buffer The IDA colonies and the buffer zone have beengiven a weight of 1 (figure 9.8).

Slopes The criterion of reduced slopes as a threat to the natural resources isbased on the fact that steep slopes are a natural barrier for logging and forexpansion of the population Flat areas, to the contrary, are prone to be invadedrapidly The threat for slope classes has been computed using the inverse of theaverage slope in a given slope class, and has been normalized using the area ofeach class (equation 5) The calculated weights for slopes are 5–15 percent slope

⳱ 3.838; 15–30 percent slope ⳱ 1.706; 30–45 percent slope ⳱ 1.023; 45–60 percentslope⳱ 0.731; 60–75 percent slope ⳱ 0.569; 75 percent slope ⳱ 0.465 Note thatthere is no area of less than 5 percent slope in the ACCVC The resulting map isshown in figure 9.9

Conflicts Between Threats The parameters used by the model are globally asindependent as possible, but there may be areas where two parameters areredundant For example, where population is concentrated there is sometimes a

F IG. 9.7 Graph depicting the effect of the number of passes of an average filter

on a point distribution having an initial value of 1 The resulting distribution isvery close to Gaussian and hence becomes smaller and broader with increasedfiltering

120 Leclerc and Rodriguez

F IG. 9.8 Map of the normalized threat due to IDA colonies

F IG. 9.9 Map of the normalized threat due to reduced terrain slopes

greater density of roads and the region is usually flatter, and the simple sum ofthe weights resulting from these parameters will overestimate the threat Toovercome this situation, we combine population and roads into a single threat:

⳱ MAX[P(population,P(roads)]

This operation is done after the normalization of the respective parameters It

is this combined threat that is used as a specific threat PRkin equation 2

Results and Discussion

Map of Critical Areas

The critical areas map is the result of combining the distinct layers representingdifferent threats, as in equation 1 (for the total threat maps, see figures 9.10 and9.11), and of overlaying the total threat map with the map showing the naturalresources of interest (normalized after prioritization) Figure 9.12 shows the

F IG. 9.10 Map of the total threat to forest which results in combining the precedingnormalized threat maps with the weights appearing in the text (under the section

“Prioritizing the Threats”), according to equation 1 Darker areas have already beendeforested or are more prone to deforestation

122 Leclerc and Rodriguez

F IG. 9.11 Map of the total threat to aquifers which results from combining the ceding normalized threat maps with weights appearing in the text (under the section

pre-“Prioritizing the Threats”), according to equation 1 Darker areas are more prone tocontribute to water contamination

F IG. 9.12 Map of critical areas (1986 forest cover mask of the total threat map of ure 9.10)

fig-critical areas map for forest, in this case a simple mask of the total threat mapwith the 1986 forest cover map (see folowing section).

These maps contain quantitative real values that we can now reclassify toobtain a qualitative map that is easier to interpret visually For example, weearlier defined five levels of criticality:

Land Use Change and Model Validation

We used Landsat TM satellite imagery and aerial photographs for cloudy areas

to estimate land use changes in the ACCVC for the period 1986–1992 (figure9.13) To validate the model, the critical areas map (real values) from figure 9.12was reclassified in twenty levels in intervals of 0.05 For different regions of theACCVC out of the national parks (buffer zone), the area deforested was deter-mined for every level of threat (figure 9.14)

For low levels of threats, deforestation was erratic, probably due to errors in

F IG. 9.13 Forest cover changes for the period 1986–1992, based on digital tion of Landsat TM images and on the interpretation of aerial photos Due to persis-tent cloud coverage in 1992, three 1992 scenes of different dates were combined withinformation from several aerial photos in order to obtain a better coverage Still,there were considerable areas with cloud cover, located mainly over protected areas

classifica-on very steep terrain and likely undisturbed

124 Leclerc and Rodriguez

the land use change map (slight misregistration of the 1986–1992 maps) and todifferences in the pixel size of the source imagery For the intermediate levels ofthreat the precision of the prediction improves considerably, and a linear relation-ship is found (which is to be expected if the model is correct) In the zones withthe greatest threat, the prediction of the model loses precision as evidenced bythe divergence of the curves for each region This is an indication of other factorsthat differ among regions The present analysis can help to identify the factorsthat do not contribute to deforestation For example, for high threat levels, theaverage deforestation rate seems to become constant or decrease This phenome-non could be explained in some areas by the proximity of the forested land tonational parks, where the owners prefer to maintain the forest cover in the hopethat the government will purchase the land to enlarge the core areas of the parks

In general, the model correctly predicts land use change or deforestation in

F IG. 9.14 Graph of the deforestation percentage by 0.05 threat range, as a function ofthe threat, in the buffer zone and excluding the protected areas The linear behaviorfor threats between 0.1 and 0.7 confirms the validity of our model For lower threats,erratic behavior is probably due to errors in the registration/classification of the for-est cover maps For very high threats, the model again loses precision, probably due

to social factors not considered in the analysis

the buffer zone GIS technology has enabled the model to be used as a flexibletool that can be easily improved, modified, and updated.

This model is now part of FUNDECOR’s activities For operational purposesthe critical areas map has been reclassified in three levels in order to obtain alinear relationship for deforestation (figures 9.15 and 9.16), showing low-, aver-age-, and high-priority areas When a farmer requests help and financing fromFUNDECOR, the farm is mapped on the critical areas map and the average level

of priority for that farm is obtained Farms are then ranked with respect to theiraverage priority levels, and preference for support is given to high-priority areaseven if their commercial value is not as high

F IG. 9.15 Graph of the average annual deforestation rate as a function of the totalthreat (reclassified threats) in the buffer zone (excluding the protected areas)