INFLUENCE OF FOREST COVER LOSS ON ORANG UTANS HABITAT 4. Discussion 4.1. The importance of quality differences of the inputsmaps The use of Kappa statistics to assess rater agreement in this research was based on the consideration that there are two functions of kappa statistics (Uebersax, 2002): (1) as a way to test rater independence (i.e. as a test statistic), in that it should test the null hypothesis that there is no more agreement that might occur by chance given random guessing; and (2) as a way to quantify the level of agreement (i.e. as an effect-size measure), in that kappa quantifies actual levels of agreement and uses a term called the proportion of chance (or expected) agreement. This is interpreted as the proportion of times rater would agree by chance alone. This study applied the second use of kappa. Results revealed that there were not relatively significant differences in the results as there was a high level of agreement of output maps. The 15 combinations of result maps are categorized into very good agreement, with kappa index agreement values more than 0.800. Moreover, the agreement between R1 and R3 and between R2 and R4 showed kappa index is equal to 1 meaning a perfect agreement of almost overall points on the maps in term of forest and non forest cover. The reason behind of the complete agreement came from the input maps. If it is checked from original forest maps: forest 2000b = forest 2000a, the difference is only that forest 2000b has many more polygons, that is why for scenario 1 = scenario 3 using the same 1985a but different F 2000a and F 2000b in that F 2000a = F 2000b. The same is true for scenario 2 = scenario 4 using the same 1985b but different F 2000a and F 2000b in that F 2000a = F 2000b. It can be said that between forest 2000a and 2000b have the same source (Landsat ETM imagery 1999 and 2000, modified by images from 2002, 30 m resolution) and the same provider (Table 2). The difference between them is that the number of polygons in that forest 2000b are more than those for 2000a (but they have the same forest boundary). The difference between F 2000c and F 2000a/b is that there is an area beyond F 2000c (only appearing in F 2000a/b called probability of F 4/6, Rm1_4) which is not in F 2000c 58 INFLUENCE OF FOREST COVER LOSS ON ORANG UTANS HABITAT so the intersection result is just for R1_R4 not including R5 and R6 which use F 2000c). Furthermore there is an area on eastern part of the map which only appears on F 2000c. The input of F 2000c has a different source (Landsat TM imagery 1998, 30 m resolution and obtained from MoF). The difference between F 1985a and F 1985b is mainly because they came from different providers, F 1985a was from UNEP-WCMC map 1996 (version I). On the other side, F 1985b was from WCMC-CIFOR map 1997, even though both maps were based on the same source (RePPProT data 1985, Landsat MSS 60 m resolution). We can say that the difference results were generated from different types of input maps. It is therefore important to note that the quality differences of input maps affected the results. The differences of input maps are inherent from different sources (sensors used), method of classification, and providers (experts’ knowledge). However, in this study, the assessment of the extent of the differences of the results (derived from different input maps) showed that all results have relatively very good agreement with each other, and four of them gave exactly similar results (R1=R3 and R2=R4). It means there were not significant differences among input maps. Another study about the effect of quality difference of input maps has different finding. Djajono and Weir (1994) found that assessment on two different qualities of slope maps having different scales showed high agreement for only one class, and low agreement for other four classes. This will generate different result since the high difference (low agreement) between the two input maps. 4.2. Forest cover Running all different input maps of forest cover as well as non forest resulted in four main components: (1) definite forest, (2) definite non forest, (3) probabilities of still either forest or non forest, (4) change areas (during 15 years there were changes of forest cover). It should be noted that both areas of all probabilities (3) and changes areas (4) indicated areas which definitely no change. It means to find out the forest loss area due to the main objective of this research both areas should be considered as changes areas (to be lost or to be afforested during 15 years, 1985-2000). Although the probabilities of forest and 59 INFLUENCE OF FOREST COVER LOSS ON ORANG UTANS HABITAT non forest differentiate probability levels, for instance, probability of 5/6 and 4/6 of forest could be said as areas having high probabilities to be forest (over 15 years) or probability level of 1/6 and 2/6 of forest could be said as areas having low probability to be forest, and tended to be non forest, but the research result showed that after processing loss and afforestation (table 9) high probability of forest (4/6) located in the loss 1990-2000. On the other hand, high probability of non forest (4/6) located in the afforestation 1985-1990. It can be concluded that high/low probability levels do not indicate tendency either forest or non forest. All levels of probabilities will indicate possibility of being forest or non forest. Table 10 clearly depicts forest cover area in 1985, 1990 and 2000 obtained from the particular components of definite forest, forest loss, and afforestation. The loss of forest cover during 5 years (1985-1990) and during 10 years (1990-2000) could also be assessed. It is important to note that during the first interval years (1985-1990) forest cover loss per year is smaller than the second interval (1990-2000). It is shown by the forest loss rate/year of first interval which is 0.24% per year (about 6,038 ha/year), while the second interval is 1.47% per year or about 36,527 ha/year (six times higher than the first one). The reason behind this high rate of forest cover loss is associated with the proximate causes of forest cover loss like roads and rivers, and also the influence of regional policy and regional development during those times. The elaborate explanation of that will be described in sections 4.5 and 4.6). The anual rate of forest cover loss of Kapuas Hulu during 1990-2000 was higher than the annual rate of forest cover loss of overall Indonesia which is 1.2%. At that period Indonesia was the sixth rank of the top ten losers of forest cove (Sengupta, 2004). In Kapuas Hulu itself, with 1.47% of annual rate of forest cover, it is nearly the annual rate of forest loss in Zimbabwe as the third rank of losers in the world which has 1.5%. 4.3. Orang Utans habitat fragmentation The condition of Orang Utans habitat loss (fragmentation) in 1989, 1999, and 2004 can be seen clearly from table 11. The changes of patches characteristics of the habitat are also be depicted in figure 15-17. In brief, during 15 years (1989-2004) there was huge 60 INFLUENCE OF FOREST COVER LOSS ON ORANG UTANS HABITAT fragmentation of Orang Utans habitat mainly in the second interval (1990-2004). The habitat fragmentation was indicated by habitat loss (area/size).