3. Results

The results obtained by the visual interpretation of the SAR images and the comparison with the data obtained from the topographic maps of 1977 are reported in this chapter. Results refer to the area under study, that is the area covered by the satellite frames, and to the period of acquisition of the satellite data.

Table 5 shows the total area covered by the features of interest in the entire study area. This include various types of aquaculture and fisheries structures, plus the salt pans.

TABLE 5
Total area covered by the classes of interest (Pangasinan province)

¹ Identified in one image only, out of two or three.
² Uncertain assignment: may be a small island or a rough patch in the sea surface.
³ Polygons drawn around the areas on which fish traps were detected, to have an estimate of their extension.

The study area covers completely the Pangasinan province, plus approximately two-thirds of La Union and a small portion of Zambales provinces. All mapped aquaculture and fisheries structures occur in the Pangasinan province only, with the exception of some fishponds (90 units covering 18.762 km²) and of some fishponds classified as uncertain (13 units covering 2.613 km²) existing in the other two provinces. Table 6 summarizes the statistics on fish traps. These include all the segments composing the arrow-like traps, if detectable. The results obtained for each type of fisheries structure are presented in the next sections.

TABLE 6
Length of the fish traps detected in the study area

FIGURE 13
Interpreted RADARSAT-1 SAR image and the resulting map of the aquaculture and fisheries structures

3.1 Fishponds

Fishponds occupy the largest surface of all the structures occurring in the area. It is interesting to compare the surface occupied by fishponds in the year 2002 (excluding the class fishponds - uncertain) with their area coverage mapped in the 1977 cartography.

The cartography showed three types of fishponds: active, empty, and active containing nipa (section 2.2). Fishponds are routinely emptied for maintenance purposes, and nipa trees are cultivated inside active fishponds; thus, in order to perform a correct comparison, the cumulative surface covered by these three classes in the cartography of 1977 has been calculated.

The results are shown in Table 7: the cumulative area occupied by fishponds has increased. It must be noted that this does not correspond to a simple expansion of the fishponds existing in 1977. Some of the ponds mapped in 1977 were converted to other uses (e.g. commercial lots).

Fishponds appear in SAR images as shown in Figures 8 and 13.

TABLE 7
Comparison between the fishponds area coverage in 1977 and in 2002

FIGURE 14
Colour combination of the three SAR images used in the study. Red: RADARSAT-1, 4 February 2001; green: ERS-2, 2 December 2002; blue: ERS-2, 23 December 2002
(Note the different position of fish cages in 2001 and 2002)

3.2 Fish pens

Fish pens were detected only on RADARSAT-1 data, as explained in the preceeding chapter. The pens are located in the estuaries of the major rivers. In many cases, they are adjacent to fishponds.

The 1977 cartography does not include fish pens, thus a comparison with the actual position and location is not possible. Figure 13 shows the typical appearance of fish pens on SAR images.

3.3 Fish cages

Fish cages were detected both on the RADARSAT-1 image of 4 February 2001 and in the two ERS-2 images of 2002. They are more evident in the first ERS-2 image (2 December 2002), as in the second one (23 December 2002) the roughness of the sea surface and the low tide decreased the contrast between cages and the surrounding surfaces (Figure 14). Table 5 lists the number of elements (cages or groups of cages) found by visual interpretation of the RADARSAT-1 and ERS-2 images, and the corresponding total surface. The number of cages detected in the ERS-2 images of 2002 is greater than in the RADARSAT-1 image of 2001. However, the total surface of the cages in 2002 is smaller than in 2001. In fact, more small cages and less large cages were detected in the 2002 images with respect to the 2001 images.

Fish cages may be of several shapes (square, rectangular, circular) and of various materials. Those made up mainly of metal have a brighter appearance on SAR images. In fact, in addition to the factors described in section 2.2, the intensity of the SAR backscattered signal is proportional to the dielectric constant of the scattering surface; metals have high dielectric constants and generate a stronger return signal.

Figure 15 shows the appearance on RADARSAT-1 SAR and ERS-2 data of small metallic and non-metallic fish cages. Figure 16 shows the appearance of large cages on RADARSAT-1 SAR data.

Data on fish cages in 1977 are not available, thus a comparison with the actual area coverage and location is not possible.

FIGURE 15
Appearance on RADARSAT-1 (left) and ERS-2 (right) images of small metallic and non-metallic fish cages. Metallic cages appear as brighter spots

3.4 Fish traps

Fish traps were separated into two categories: offshore traps and traps inside major rivers. The areas occupied by fish traps were contoured only approximately to estimate their extension. Traps were detected only on RADARSAT-1 data of 2001.

Table 6 shows the statistics obtained from the polylines digitized from the detected traps. These data are not precise, as in many cases only the central structure of the traps is visible in the images; on the other hand, because of their dimensions (section 2.2) the uncertainty on the identification of traps is higher than that of the other structures. Figure 17 and 18 show respectively the typical appearance of offshore traps and traps inside major rivers on RADARSAT-1 SAR images. Traps were not mapped in the cartography of 1977, thus a comparison with the past situation is not possible.

FIGURE 16
Appearance of large fish cages on RADARSAT-1 SAR data

FIGURE 17
Appearance of offshore fish traps on the RADARSAT-1 SAR image

FIGURE 18
Appearance of fish traps inside rivers on the RADARSAT-1 SAR image

3.5 Land cover changes

By comparing the results of the SAR mapping with the 1977 cartography, some interesting observations can be made.

The 58.2 percent of the present fishponds occupy areas which were already devoted to aquaculture in 1977. The new fishponds (31.5 percent) are mainly located on former agricultural land.

In contrast, only 6 percent of the land covered by fishponds in 1977 has now a different use. This can also be a direct consequence of the evolution between 1977 and 2002 of the drainage network, the major rivers having noticeably changed their course with subsequent flooding and siltation phenomena.

3.6 Field verification exercise

A team of the BFAR went in December 2003 to Lingayen Gulf to check the accuracy of the mapping results. They were equipped with all the necessary tools (GPS, compass, digital cameras, topographic maps and SAR mapping results at 1:50 000 scale) and checked the interpretation of SAR images. Before discussing the field validation results, the following aspects of the work should be considered:

1. Fish traps were recognized only on RADARSAT data which were acquired on February 2001, that is about two years before the field check. Some of them could have been removed or moved somewhere else in this timeframe;

2. Fish cages were easily mapped with both RADARSAT-1 and ERS-2 data. As the cages are floating, they can be moved to other places if there is a need. Actually, a group of cages located between Luzon Island, Santiago Island and Cabarruyan Island presents two distinct locations in RADARSAT-1 (February 2001) and in ERS-2 (December 2002), most probably as a consequence of a typhoon (Figure 14). Thus the field checking of fish cages was limited to ascertaining the presence of fish cages in the vicinity of the place indicated in our SAR-derived maps, as in a one year interval the cages could have been moved somewhat. However, in the majority of the cases, the fish cages were still at the places mapped from radar data.

3. Fish pens and fishponds, being semipermanent structures, were not influenced by the one year time interval and thus they were field checked at the exact coordinates reported in the radar-derived maps. In limited cases fishponds were converted to other uses in the most recent months, but interviews with local people confirmed their original nature.

The field work was thus aimed at mainly verifying the interpretation of fishponds and fish pens. The survey was conducted on 32 verification points, selected by means of a two-stage cluster sampling scheme (Figure 19). Some verification points were also located on the “fishponds, uncertain” class, as the results could have assisted in fine-tuning the interpretation keys. The results of the ground truth on the verification points and the corresponding visual interpretation are described in Table 8.

The ground verification included also six observation points located offshore and inside rivers, in order to confirm the presence of fish cages and traps in/or the proximity of the points in which they are located in the SAR images.

The analysis of the ground truth at the verification points shows that both the two points located on the class “fishponds, uncertain” are in fact other types of water-covered surfaces. This confirms the correctness of the main interpretation key, according to which the water-covered surfaces were assigned to the class fishponds only if the surrounding dykes were visible. Water-covered surfaces regular in shape but not surrounded by visible dykes were assigned to the class “fishponds, uncertain”; the results of the ground survey demonstrate that the class “fishponds, uncertain” should be removed from the final map.

Of the other 30 verification points, 23 are located on areas interpreted as fishponds, four on fish pens and three on salt pans. The actual land use was different at one point only; it had been assigned to fishponds by the visual interpretation, but the corresponding area is a marshland bordering salt pans.

The user’s accuracy of the verified classes, e.g. the ratio between the total number of points truly belonging to a class and the total number of points assigned to the same class by the visual interpretation procedure, is thus 100 percent for salt pans and fish pens, and 95.7 percent for fishponds. These figures give the probability that a point on the interpretation map truly corresponds to the class to which it has been assigned. However, the actual accuracy of the “salt pans” class might be lower, due to the fact that they may appear very similar to fishponds when they are completely flooded.

FIGURE 19
Position of the verification points and of some observation points

TABLE 8
Comparison of visual interpretation and in situ observations on the verification points

The accuracy cannot be calculated for fish cages and traps, as they may have been moved or removed in the time interval between the image acquisition and the field verification, as a consequence of severe weather conditions. In fact, differences in their positions are visible on the 2001 and 2002 images, as explained above. Thus, mapping accuracy for fish cages was estimated at 90 percent and for fish traps at 70 percent. The results of the ground survey show that offshore fish traps are still approximately located in the same areas in which they were when the images were acquired, while the traps detected inside rivers were not observed in the two surveyed points. The fish cages at the southern border of Santiago Island, clearly visible in all SAR images, are still located there, while the fish cages identified offshore Barangay Dori (Bolinao) are now replaced by fish pens built seaward from the rocky coastline (see appendix).