2 Methods

Figure 2. The phytogeographical regions of Monte and Espinal, showing the subregions of Algarrobal, Caldén and Ñandubay.

In order to advance an estimation of the native forest cover in the regions Monte and Espinal a sample of twelve Landsat TM satellite scenes¹ was sampled, six in each region. Due to high heterogeneity within the region of Espinal the sample was selected through stratified random sampling, dividing this region into three sub regions (Figure 2); Espinal-Ñandubay, Espinal-Algarrobal and Espinal-Caldén, with two scenes corresponding to each one of them.

Tables with paths and rows for those satellite scenes with a theoretical coverage within the regions/subregions of Monte and Espinal were created and the scenes were numbered for the sampling procedure, Table 1a-d.

Table 1a-d. Spatial distribution and enumeration of Landsat TM satellite scenes in the regions/subregions of Monte and Espinal presented by the paths and rows.

	MONTE									ESPINAL Caldén							ESPINAL Algarrobal							ESPINAL Ñandubay
	P233	P232	P231	P230	P229	P228	P227	P226		P230	P229	P228	P227	P226	P225		P230	P229	P228	P227	P226	P225		P230	P229	P228	P227	P226	P225
R078		3	12
R079		4	13																										306
R080		5	14																									303	307
R081	1	6	15																								301	304	308
R082	2	7	16																203	205							302	305	309
R083		8	17							101								201	204
R084		9	18	22						102	104							202
R085		10	19	23						103	105	108
R086		11	20	24							106	109
R087			21	25	27	30					107	110	111
R088				26	28	31	34							112
R089					29	32	35	36
R090						33

The MS Excel random function RAND()*(b-a)+a, were a is the first value in the population and b the last, was applied in each cell in a table designed for selecting the sample of satellite scene positions, (Table 2). The final sample of satellite scene positions is shown in Table 2.

Table 2. By random, selected numbers per Region/ Subregion.

Monte	Espinal
	Caldén	Algarrobal		Ñandubay
9	109	203		302
23	101	204		304
12
6
35
28

Satellite scenes corresponding to the selected numbers are presented in Table 3, indicating actual region/ subregion and further specified in Appendix 5. The spatial distribution of the selected scenes is shown in Figure 3.

Figure 3 Spatial distribution of the sample of satellite scenes, within each region respectively.

Table 3. Paths and Rows for corresponding Landsat TM scene.

N°	Path/Row	Region/Subregion
6	P232/R081	Monte
9	P232/R084	Monte
12	P231/R078	Monte
23	P230/R085	Monte
28	P229/R088	Monte
35	P227/R089	Monte
101	P230/R083	Espinal Caldén
109	P228/R086	Espinal Caldén
203	P228/R082	Espinal Algarrobal
204	P228/R083	Espinal Algarrobal
302	P227/R082	Espinal Ñandubay
304	P226/R081	Espinal Ñandubay

One of the twelve selected scenes, P231/R078, did not have any actual cover within the studied regions of Monte and Espinal and it was therefore excluded from the sample (figure 3).

2.1 Preparation of digital Satellite scenes for interpretation

The Native Forest Division in Argentina provided the selected Landsat TM scenes in digital version, BSQ format, including the six bands² 1, 2, 3, 4, 5 and 7. The chosen scenes had been picked out by the Consortium CAC for the national forest inventory and were shot mainly in the period of January - February with a low coverage of clouds.

For the interpretation of forest cover, the digital satellite scenes were converted into three band multispectral TIFF images by the image processing software PCI. The band combination 4, 3, 5 (Near IR, Red, Shortwave IR) was used to get the best discrimination of the dry forests in the actual regions.

The PCI TIFF format was modified by the image processing software Adobe PhotoShop 4.0 so that the satellite scenes could be georeferenced and analysed by the ESRI GIS software ArcView 3.2 with the extension Image Analyst 1.0. Analogue image maps (Instituto Geográfico Militar, Appendix 3) and a digital vector map (ESRI, 1995) were employed for the georeferenciation. Four control points were indicated in each scene and the geographical co-ordinates were specified by measuring in the analogue maps or by direct links to the digital map. The scenes were all prepared in the projection UTM 1983 zone 20.

2.2 Attempt to automatic interpretation

An effort was made to find out a method to interpret, classify and discriminate the forest cover in the digital satellite scenes through automatic classification and trough classification indicated by "training areas", but it failed due to high spectral heterogeneity within the forest areas. A wide variety in forest density and influence of grass and other ground cover with big effect on the spectral reflectance, implicated a high spectral resemblance between some native forests and brush lands, and with the GIS software used it was very difficult to separate them from each other.

2.3 Visual interpretation - phase 1

To initiate the visual interpretation of the multispectral Landsat TM scenes the bands 4, 3 and 5 were assigned the colours red, green and blue respectively. The scenes were enhanced by stretching the histograms using a Standard Deviation stretch, and further adjusted manually to optimise the appearance for visual interpretation and discrimination of forest cover. A Vegetative Index³ was calculated to assist the detection of forests in the scenes. The satellite images were also plotted⁴ in the scale of 1:250.000 using the band combination 4, 3, 2 (Near IR, Red, Green) with the colour assignment red, green and blue respectively to facilitate the visual interpretation.

The Argentinean Native Forest Division provided digital geographical data describing the extensions of the phytogeographical regions of Monte and Espinal. The original data was projected in the country projection Argentina zone 3 and was reprojected into UTM 1983, zone 20, using the ArcView 3.2 Shapefile Projection Utility. The geographical extensions of Monte and Espinal, completed with the subregions of Espinal; Ñandubay, Algarrobal and Caldén are shown in Figure 2.

The satellite scenes were interpreted at the visual scale of 1:100.000⁵ in an ArcView project using a View projected in UTM 1983, zone 20. Native forests and forest plantations within the study area of each satellite scene were delimited by screen digitalisation and saved as an ArcView Shape File, Forest Cover, with the attributes Area and Forest type, where the attribute forest type consisted in Native forest or Plantation. Smallest native forest formation interpreted and delimited was 3 ha for while for forest plantation it was 1 ha because of the organised structure of planted tree formation. Areas within the extension of actual region were interpreted for each satellite scene, independent of scenes overlapping.

2.4 Ground-truthing

Figure 4 Area for ground-truthing

A field visit was planned together with professional officers at the Argentinean Native Forest Division, to verify the visual interpretation of forest cover in the satellite scenes. Due to the rain season, it would have been too complicated to visit some of the areas, since the roads were in a very bad condition for transportation. To co-ordinate the mission with the Argentinean Native Forest Division a limited time available for field control was considered. The most remote areas were therefore excluded and along with the recommendations from forestry officers at the Argentinean Native Forest Division, representative areas with different types of forest were picked out for ground-truthing. The areas selected covered the Province of La Pampa and the southern part of San Luis and included the two satellite scenes P228/R086 and P230/R085, Figure 4.

A vehicle and several road maps were provided by the Argentinean Native Forest Division to carry out the field visits. Forest areas and areas with an undefined forest resembling ground cover indicated in the visual satellite scene interpretation were visited and the actual ground cover was verified and its vegetation described. Pictures were taken to illustrate the different types of vegetation and some control points⁶ were registered with a GPS⁷ for later evaluation and calibration of the visual interpretation using "training areas". Control points were also captured outside the extension of the actual satellite scenes to complement the evaluation with additional training areas.

2.5 Evaluation of visual interpretation using data from ground-truthing - phase 2

For the evaluation of the visual interpretation of forest cover in the satellite scenes the co-ordinates⁸ from the control points captured in field was presented as an event theme, Figure 5, in ArcView and the theme attribute table (Appendix 2) complemented with vegetation descriptions and vegetation classifications⁹. The training areas was defined by the control point co-ordinates and served to calibrate the visual interpretation of forest cover in all satellite scenes.

To complement the training area evaluation, the Argentinean Native Forest Division facilitated an extra sample of cut satellite scenes (Figure 5) covering the control points captured outside the extension of the actual satellite scenes.

The control points in dense shrubs, interpreted as forests in phase 1, were used to correct the visual interpretation of all twelve scenes. For the rest of the control points only misinterpretations in the visited area were corrected, since it was difficult to distinguish different wooded vegetation from each other without field controls.

Figure 5 Distribution of control points captured in field visit, presented on corresponding satellite scenes and complementary set of cut scenes.

2.6 Forest area estimation

2.6.1 Area calculation of forest formations in sample

The delimited forest area was calculated automatically with the area calculate script View.Calculate.Area, writing the area for every polygon to the "Area" field in the attribute table. The area is given in the map units of the View, which is meter in the UTM 1983 zone 20 projection, and thus resulting in square meters. The extensions of the regions/subregions of Monte and Espinal were here used as masks to exclude forests outside the study areas.

2.6.2 Estimation of total forest cover

Two methods were employed to estimate the forest cover in each region/subregion. The first method applied partial aggregation ([C₁/A₁+C₂/A₂+C₃/A₃+C_n/A_n]/n _* A_tot¹⁰) and calculated the local concentration of forest cover in each satellite scene; then it calculated the arithmetic mean of all the local forest concentrations and weighted it with the total area of the region/subregion. This method gives equal importance to every one of the interpreted satellite scenes without taking in consideration the actual studied area, meaning that a small area contributes with its forest concentration with the same weight as a big area.

The second method made a global proportional estimation (C₁+C₂+C₃+C_n)/ (A₁+A₂+A₃+A_n) _* A_tot¹⁰ and calculated the forest concentration for all satellite scenes together. It was based on the total forest cover of all scenes, divided by the total interpreted area, and then weighted it with the total area of the region/subregion. This method is more "straight on" and treats the sum of all interpreted areas without contemplating the contribution of each satellite scene, which gives higher importance to the scenes with a larger studied area.

¹ Scenes from the satellite Landsat 5 with the scan mirror spectrometer Thematic Mapper, TM.

² The Bands spectral range: 1 = 0.45-0.52µm, 2 = 0.52-0.60µm, 3 = 0.63-0.69µm, 4 = 0.76-0.90µm, 5 = 1.55-1.75µm, 7 = 2.08-2.35.

³ Normalised Difference Vegetation Index (NDVI) = (IR-R)/(IR+R), where IR is the Near IR band and R is the Red band.

⁴ Satellite images plotted by Earth Satellite Corporation in Rockville, USA.

⁵ The View map scale 1:150 000 was applied, which gives an actual scale of 1:100 000, employing a 19'' screen and the screen resolution 1024*768.

⁶ The co-ordinates Latitude/Longitude (WGS84) for the actual spot visited.

⁷ The Global Positioning System (GPS) Garmin 12XL was used for the capture and storage of map co-ordinates.

⁸ Co-ordinates in the format degrees (-DD,DDDDD).

⁹ Native Forest, Open Native Forest with pasture, Burned Native Forest, Shrubs, Meadow.

¹⁰ C₁ = forest area in interpreted part of satellite scene "1", A₁ = total area of interpreted part of satellite scene "1" and A_tot is the total area of the region/subregion.