ANNEX 1
Country tables

Note:
NA indicates data not available
All percentage rates of change are compound rates

¹ Sub-totals and total gross national product exclude those countries for which annual growth data is not available. The values for these countries are indicated with an *.

(*) Survey reliability class

(1) High Reliability: forest inventories based on high resolution satellite data (LANDSAT TM, SPOT) or aerial photographs, supplemented with extensive field checking or sampling.

(2) Average Reliability: forest inventories based on medium resolution satellite data (typically LANDSAT MSS) with limited ground truthing.

(3) Low Reliability: Surveys or maps based on heterogeneous material like vegetation maps, land use surveys, etc. generally at coarse resolution and often out of date. The reliability class 3 represent the cases with insufficient information and need for a reliable baseline in future.

(**) Change assessment reliability classes:

(1) High Reliability: this class is assigned to the countries for which two or more coherent (in terms of classification of forest types) inventories exist, and the observed forest change is used to perform a ‘local fit’ of the model parameters.

(2) Average Reliability: this class is assigned to the countries for which some partially reliable multi-date observations are available and are used to control the model estimates.

(3) Low Reliability: this class is assigned to countries for which no reliable multi-date information is available. The estimation of the change in forest cover area is based on the general model.

Reforestation refers to Annual plantation rate; Deforestation refers to Table 5A - Annual Change

ANNEX 2
Deforestation modelling

1. INTRODUCTION

As pointed out in the main text, the forest inventory data available with the countries have different reference dates ranging from 1955 to 1992 and need to be standardized to a common reference date (i.e. year 1990) for the global assessment purposes. The adjustment requires an estimate of the rate of change of the forest cover during 1980–90.

There are two ways of doing such adjustment:

1. use of ad hoc expert's estimates based on personal experience and knowledge of the study area; and

2. use of statistical models correlating forest cover and deforestation with population density and its growth and auxiliary variables.

The second approach is objective, repeatable and avoids subjective estimates, whose quality is necessarily variable and difficult to control and to document. In particular for the area under present assessment, given the scarcity of national or even partial forest inventory, the estimates of changes would require considerable guess work.

The model approach is related with the availability of statistical and spatial databases such as: sub-national units (province, district, etc.) and demographic and ecological variables.

The estimates obtained by modelling are expected to become more precise in future when additional forest surveys data become available together with socio-economic and ecological variables. The process of modelling enables a continuous upgrade of the estimates by successive approximations using improved models. The modelling will benefit from the enlargement of the multi date and single date observations. In general the model approach can be considered a useful methodology for present and especially for future global forest resources assessments.

The models are applied at subnational level and results are successively aggregated at national, regional and global levels. Keeping in view the law of propagation of errors the global estimates are expected to be more precise than the subregional; and the subregional estimates more precise than the national and subnational estimates. Nevertheless some important deviation are expected at district level.

The present results are based on a minimum set of data and should be considered mainly indicative of the current processes of deforestation and degradation. Certainly they constitute a useful starting point, covering the totality of the study area, for further analysis and understanding of dynamics of forest resources.

2. MODEL BUILDING

The techniques used in the present modelling exercise are derived from the experience acquired with the study of tropical countries. The main finding was that forest cover (as percent of land) of a given administrative unit is a function of (i) population density; and (ii) ecological conditions. The above hypotheses have been tested on 5 sub-regional data sets for North Africa, Middle East, South America, South Africa and China.

Each subregional data set includes the following variables : population density and growth; forest cover (as percent of land); potential forest cover (forest as percentage of potential forest area) where potential forest land is obtained by subtracting desert area (hot and cold); ecological zoning (percent of land in each ecological zone within the subnational unit). For some subregions additional variables were also available and have been used (e.g. rural population density, animal density, Weck Climatic Index, etc.).

Each subregion was analyzed separately using a cross-sectional analysis approach, using population density as proxy of time, this assumes that within a given ecological type the less populated district would be more forested, representing early stages of development and that with the growth of population the forest cover would decline progressively.

An illustration of data scatter is given in the following graph. above the example only the influence of population is shown. Ecological data are further included in the dataset and a multiple regression analysis performed using forest cover (or relative forest cover) as dependent variable and population, ecological zoning (% of land in each ecological class) and other factors, indicated in each case, as independent variables. The regression analysis was performed using a stepwise approach to determine the best fit.

Population density and relative forest cover example: South America

A worked out example for North Africa is given hereunder to illustrate the methodology used.

Step 1: Correlation of individual variables

Step 2: Variable selection

Montane moist area is found to be the most significant variable with an R² of 0.585.

The second variable added is Rural population density and the combined effect leads to an R² of 0.714 (adjusted to 0.686 for the degrees of freedom).

After the inclusion of rural population the other variables are no longer significant as their F values and partial correlation are low.

Step 3: Model fitting results

The model selected is as follows:

Y = a + bx + cz

Where

Y = relative forest cover
a = constant
b = percentage of area in montane moist zone
c = rural population density

and the values computed for the parameters are:

a = 16.913
b = 0.164
c = -0.138

Model fitting : R-squared (adjusted) : 0.6862 ; d.f. = 20

The selected model expresses the relative forest cover of a given unit as a function of a = constant; b = montane moist area; and c = rural population density. The first two parameters are positive while the parameter for population is negative. Montane moist zones have a positive effect on forest cover probably due to inaccessibility and limited suitability of land for agriculture and grazing; since the residuals forests are located mainly in montane zones, the environmental implication of their possible further reduction should be analyzed. On the other hand the growth of rural population leads to forest degradation and deforestation.

The above model can be used to estimate the forest area change between two dates for which the corresponding rural population density values are known. The procedure is as follows.

If:

Y1 and Y2 are the relative forest cover at time 1 and 2; and
x is the portion of land in the montane moist zone; and
z1 and z2 are the rural population density at time 1 and 2;

Y1 = a + b*x + c*z1

and

Y2 = a + b*x + c*z2

then

(Y1-Y2) = (a+b*x+c*z1) - (a+b*x+c*z2)

since a,b,c and x are constants the change function can be simplified as follows:

(Y1-Y2) = c* (z1-z2)

The dynamic aspects of the model are controlled by population size and growth steered by the estimated parameter. Nevertheless the auxiliary parameters (of ecological nature) are influencing the value of the ‘c’ parameter and are necessary to improve the correlation with forest cover and to control the effect of the population.

The methodology described for North Africa is adopted for the other subregional models also and, thus, the implications discussed before are similar. In practice for each subregional data set, one model has been statistically developed. The selected combination of variables have been analyzed to test the consistency of the results. Special attention was paid to significance of the population parameter which is crucial for change assessment. In all models except one (Mongolia) the population was the statistically most significant variable (probability > 95%). In case that the population parameter is not included in the selected model the forest cover appears to be determined only by ecological factors and the decline due to human activities appears to be not significant.

A summary of the statistical results is presented in the following table.

5. RELIABILITY ASSESSMENT

The modelling approach, though useful, is not exempt from criticism. The main question is related to the reliability of the estimates. The easiest way to assess the reliability of estimates would be to compare independent multi-date observations for given districts with the corresponding model results. Unfortunately the virtual absence of reliable multi-date observation does not allow such comparison. Moreover, the models are obtained by regression analysis and are intended to represent average trends; local deviations in individual districts can lead to underestimation or overestimation. From a statistical view, the most important requisite is the absence of bias which has been tested with residuals analysis. Even in absence of bias the model estimates are necessarily associated with an error.