3. MATERIALS AND METHODS
4.3 Analysis of land cover change trajectories
Land cover change trajectories are traced by codes and classified on the basis of the driving forces of the change. Moreover, rather than consider each land cover trajectory as a unique case, we are interested in grouping trajectories based on their characteristics. This study attempts to find main trajectories to represent the trend of land cover change influenced by natural disturbances at the landscape level. In this section, our study tries to find answers to the following questions:
1. What were the types and extent of land cover change trajectories in the Taimali watershed from 2005 to 2011?
2. Where did the hotspots of land cover change occur?
3. Which environmental variables were associated with land cover change?
4.3.1 Calculation and classification of land cover change trajectories Trajectory codes are established through formula calculation by applying the raster calculator in ArcGIS 9.3.1. For example, the trajectory of land cover change which converts from forest to landslide, and then to landslide on a pixel over three time points is marked as 211. Theoretically, there will be 64 (43) kinds of possible trajectory codes produced; however, practically some trajectory codes do not exist and some take very small percentages of the total study area so they can be omitted. To minimize potential analytic errors, the isolated trajectories are replaced with new values based on the majority of their neighboring cells through the generalization analysis of the majority filter. A total of 44 trajectories were identified, comprising 4 unchanged trajectories and 40 changed trajectories (Fig. 4.4). The selection of the main trajectories in the study area followed the criteria suggested by Mena (2008).
The trajectories covering less than 1 percent of the total watershed area were excluded from the following analysis because they might be derived from the artifacts of
classification error. The top 7 trajectories, each of which covered more than 1 percent of the total watershed area, were chosen as the main trajectories and shown in Table 4.5. These trajectories, comprising 95.07% of the entire study area, contained two unchanged and five changed trajectories.
First, the single most significant trajectory (222), comprising 80.32% of the landscape, was the persistence of forest cover. The second most important group of trajectories, comprising 11.75% of the landscape, consisted of five change trajectories that were 221, 211, 224, 212, and 244. These five trajectories selected as the main trend changes of land cover in the Taimali watershed contained 75.65% of the entire changed area (Table 4.6). Based on the cause of conversion, change trajectories could be divided into natural-induced or human-induced change classes at level 1.
According to the processes of conversion, all of the five natural change trajectories were grouped into three classes at level 2, inclusive of Forest-Landslide (FL), Forest-Channel (FC), and Vegetation Recovery (VR) (Table 4.6 and Fig. 4.5).
Human-induced change class involved change trajectories caused by human activities.
This class only occupied 2.84% of the entire changed area, so no level 2 class was further divided. It can be seen most land cover transformations in the study area resulted from natural changes induced by catastrophic typhoons. Therefore, typhoons played a pivotal role in the environmental change of the Taimali watershed. These three classes (or trajectories), inclusive of FL, FC, VR, covering 75.65% of the entire changed area, were considered the main trend changes and represented overall landscape (OL) changes in the Taimali watershed.
Based on the characteristic of natural effects, three level 2 classes can be separated into two groups, i.e., a positive and negative change group. For example, FL and FC, occupying 65.35% of the entire changed area, are negative change groups.
They demonstrate that forested lands had converted into landslides or channels, which
were caused by the torrential rainfall of strong typhoons. However, on the contrary, VR, taking 10.30% of the entire changed area and indicating vegetation recovery on landslides, belongs to a positive change group. From the point of view of soil and water conservation, disturbances mostly caused negative changes in the Taimali watershed, although there were some dispersed vegetation recoveries.
Table 4.5 The top 7 trajectories in the Taimali watershed
Rank Trajectory code Percentage (%) Cumulative percentage (%)
1 222 80.32 80.32
Table 4.6 The percentages of land cover change trajectories
Level 1 class Level 2 class Trajectory code
Channel corridor 444 3.01
Natural-induced
Vegetation Recovery (VR) 212 1.60 10.30 Human-induced
change 0.44 2.84
Fig. 4.4 Total trajectories of land cover in the Taimali watershed
Fig. 4.5 Main land cover change trajectories in the Taimali watershed
Unchanged FL FC VR
4.3.2 Change analysis in connection with environmental variables After the trajectory layer was acquired, two interesting questions were whether the distribution of change trajectories revealed some spatial characteristics and what environmental variables affected land cover change. To quantify the relationship between the change trajectories and the environmental variables, two steps are necessary. First, the values of selected main change trajectories in the trajectory layer were set at 1 and the values of other unselected trajectories were set at 0.
Consequently, a binary map of land cover changes were produced (Fig. 4.6). Second, the binary map was overlaid with the distribution maps of related environmental variables to measure the percentage and RCI of the land cover changes in each subregion of the environmental variables. This analytic process was achieved by carrying out the“Tabulate Area＂of spatial analyst tools in ArcGIS. The results of change analysis with the associated eight environmental variables were described as follows:
Fig. 4.6 Binary map of land cover trajectories
(1) Lithology
As shown in Table 4.7, two major formations, Pilushan and Lushan, comprise almost the entire watershed and each one covers nearly half of the land area. As presented in Fig. 4.7, RCI values larger than 1 found in the formation of Pilushan and Tananao Schist indicate that there was a highly concentrated land cover change occurring, especially for the Pilushan Formation. Besides, more than 80 percent of the entire changed area was also located in Pilushan Formation. Therefore, it seems that Pilushan Formation is prone to land cover change occurring as a result of relatively fragile geological conditions.
Table 4.7 The area distribution for lithological classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
Pilushan 46.70 98.78 20.33 20.58
Lushan 51.66 109.28 4.06 3.72
Tananao 1.64 3.47 0.51 14.58
Watershed 100.00 211.53 24.90 11.77
Fig. 4.7 Map of RCI and percentage of the entire area of change for lithology
1.75
(2) Distance to faults
Each of the five classes of distance to faults covers about 20% of the study area (Table 4.8). At first, both the values of RCI and percentage of the entire area of change ascended with the increment of distance to faults, and then both of their values showed a decreasing trend when the distance to faults was increasing (Fig. 4.8).
Overall, two indexes’ values reduced with the increasing distance to faults and this result suggests that the high intensity and large extent of land cover change were situated within 0-6000 m distance to faults. The region adjacent to faults underwent land cover changes to a large extent, probably due to the fragile geologic structure in the neighborhood of the faults.
Table 4.8 The area distribution for classes of distance to faults
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
Watershed 100.00 211.53 24.90 11.77
Fig. 4.8 Map of RCI and percentage of the entire area of change for distance to faults
1.29 1.49
(3) Rainfall
As shown in Table 4.9, the rainfall regions of 4000-5000 mm and 3000-4000 mm take up the two greatest proportions of the study area of about 35% and 30%, respectively. However, the two biggest RCI values were found in the rainfall regions of 4000-5000 mm and 5000-6000 mm with 1.85 and 1.37, respectively (Fig. 4.9).
Moreover, the two greatest percentages of the entire area of change were also located in these two regions and the total was up to 88%. These results demonstrate that two regions experienced concentrated and extensive land cover changes. A large amount of rainfall might cause the movement of soil and rocks to trigger landslides and channel expansion.
Table 4.9 The area distribution for rainfall classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%) 2000-3000 mm 18.06 38.21 0.19 0.49 3000-4000 mm 29.76 62.95 2.80 4.46 4000-5000 mm 34.78 73.56 15.98 21.73 5000-6000 mm 17.40 36.80 5.92 16.09
Watershed 100.00 211.53 24.90 11.77
Fig. 4.9 Map of RCI and percentage of the entire area of change for rainfall
0.04
(4) Distance to rivers
The change ratios presented a decreasing trend with the increasing distance to rivers (Table 4.10). As illustrated in Fig. 4.10, both the RCI and percentage of the entire area of change displayed a uniform trend where two indexes’ values decreased with the increasing distance to rivers. The greatest values of the two indexes were found within 0-300 m distance to rivers and this result reveals that the high intensity and large extent of land cover change were closely adjacent and tied to rivers.
Pearson correlation analysis was performed by regarding distance to rivers as the independent variable and RCI as the dependent variable. The results are presented in Table 4.11 and a scatter plot is also shown in Fig. 4.11. The result of Pearson correlation analysis demonstrates a strong negative linear correlation between distance to rivers and RCI (r = -0.937, Sig. = 0.05). It reveals that the RCI values became smaller when the distance to rivers became larger. The simple linear regression equation is shown below:
Y = -0.3516X + 1.8373 (4.1)
where Y is the RCI and its value ≥ 0; X is the distance to rivers and its value > 0.
Table 4.10 The area distribution for classes of distance to rivers
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
0-300 m 35.45 74.99 15.39 20.52
300-600 m 23.21 49.11 5.31 10.81
600-900 m 14.79 31.28 2.21 7.06
900-1200 m 9.69 20.50 1.00 4.88
＞1200 m 16.85 35.65 1.00 2.79
Watershed 100.00 211.53 24.90 11.77
Fig. 4.10 Map of RCI and percentage of the entire area of change for distance to rivers
Table 4.11 Linear correlation analysis between RCI and distance to rivers
RCI Distance to rivers
RCI
Pearson correlation 1 -.937*
Sig. (2-tailed) .019
N 5 5
Distance to rivers
Pearson correlation -.937* 1
Sig. (2-tailed) .019
N 5 5
* Correlation is significant at the 0.05 level (2-tailed).
Fig. 4.11 Scatter plot between distance to rivers and RCI value
1.74
(5) Elevation
Elevation of 500-1000 m takes up the greatest proportion of area, at about 45%, and the elevation of 0-500 m and 1000-1500 m also occupies considerable proportions, at about 20% and 21%, respectively (Table 4.12). As shown in Fig. 4.12, RCI values larger than 1 demonstrate that there were concentrations of land cover changes at elevations of 1000-2500 m, although the greatest percentage of the entire area of change was found at elevations of 500-1000 m. When explaining the relationship between elevation and land cover change, the correlation between elevation and rainfall and slope should be considered. As a general rule, the higher the elevation is, the greater the rainfall and slope gradient are.
Table 4.12 The area distribution for elevation classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
Watershed 100.00 211.53 24.90 11.77
Fig. 4.12 Map of RCI and percentage of the entire area of change for elevation
0.88 0.86
0‐500 m 500‐1000 m 1000‐1500 m 1500‐2000 m 2000‐2500 m ＞2500 m RCI Percentage of the entire area of change
(6) Slope
The proportions of area covered by slope gradients between 15°-45° are all larger than 21% and the greatest one is located at a gradient of 25°-35° (Table 4.13). Overall, the RCI index increased with the ascending slope grades, but fluctuated at the gradient of 5°-15° (Fig. 4.13). The percentage of the entire area of change also rose with the increasing slope grades, but declined at gradients of 35°-45°. Gradients of 35°-45°
and greater than 45°, with the highest RCI values of 1.23 and 1.46 respectively, indicate that land cover had experienced high intensity change in these two subregions.
Land cover change may result from forested land becoming landslides on steeper slopes.
Table 4.13 The area distribution for slope classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
Watershed 100.00 211.53 24.90 11.77
Fig. 4.13 Map of RCI and percentage of the entire area of change for slope
0.73
0°‐5° 5°‐15° 15°‐25° 25°‐35° 35°‐45° ＞45°
RCI Percentage of the entire area of change
(7) Aspect
The three greatest proportions of area occupied by slope aspects from the northeast, east, to southeast characterize the study area and the total proportion is about 48% (Table 4.14). RCI values greater than 1 were located in the slope of NE, E, SE, and S (Fig. 4.14). The top four percentages of the entire area of change were also situated in the above four slope aspects. Among them, the NE, E, and SE, which occupy 57.09% of the entire area of change, could be grouped into the eastward slope;
hence the eastward slope was the dominant region where land cover change occurred.
Since the outlet of Taimali watershed is eastward and typhoons often strike Taiwan from the east, typhoons that brought torrential rainfall and wind gave rise to intensive land cover change on the eastward slope.
Table 4.14 The area distribution for slope classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%)
N 8.90 18.82 1.59 8.43
NE 15.21 32.17 4.50 14.00
E 17.87 37.81 5.36 14.19
SE 14.60 30.89 4.34 14.06
S 12.16 25.73 3.77 14.65
SW 11.73 24.81 2.45 9.87
W 11.40 24.12 1.57 6.51
NW 8.11 17.16 1.31 7.64
F 0.02 0.03 0.00 0.00
Watershed 100.00 211.53 24.90 11.77
Fig. 4.14 Map of RCI and percentage of the entire area of change for aspect
(8) Curvature
The curvature classes of the study area are primarily comprised of concave and convex landforms that occupy 48% and 49% of the entire watershed, respectively (Table 4.15). As shown in Fig. 4.15, the RCI values in these three subregions of the curvature distribution map were relatively more even than the other environmental variables. RCI values decreased with the negative values transforming into positive ones. The concave area, with a RCI value larger than 1, indicates that higher land cover change intensity occurred in the depression or valley. The greatest percentage of the entire area of change was also found in the concave area but the difference in RCI and percentage of the entire area of change between the concave area and the convex area was not significant.
0.72
1.19 1.21 1.19 1.24
0.84
0.55 0.65
6.37% 18.09% 21.55% 17.45% 15.14% 9.83% 6.30% 5.26% 00%
0.0 0.5 1.0 1.5
N NE E SE S SW W NW F
RCI Percentage of the entire area of change
Table 4.15 The area distribution for curvature classes
Class % Area covered Area (km2) Area of change (km2) Change ratio (%) Concave (-) 48.00 101.53 13.19 12.99
Flat (0) 2.71 5.74 0.61 10.67
Convex (+) 49.29 104.26 11.10 10.64
Watershed 100.00 211.53 24.90 11.77
Fig. 4.15 Map of RCI and percentage of the entire area of change for curvature
4.3.3 Summary
Through formula calculation, there were 44 trajectories recognized, inclusive of 4 unchanged trajectories and 40 change trajectories. By referring to selection criteria where the area was larger than 1 percent of the entire study area, the top 7 trajectories were chosen as the main trajectories. These trajectories comprising 95.07% of the entire study area contained two unchanged trajectories and five change trajectories.
According to the cause of conversion, five change trajectories belonged to natural-induced change classes at level 1. Based on the processes of conversion, all of the five natural change trajectories were grouped into three classes at level 2,
1.10
inclusive of FL, FC, and VR. These three classes, covering 75.65% of the entire changed area, were considered main trend changes and represented overall landscape (OL) changes in the Taimali watershed. These results reveal that land cover changes in the study area almost always arose from natural forces. Therefore, natural disturbances play a crucial role in the environmental change of the Taimali watershed.
In addition, based on the characteristics of natural effects, three level 2 classes can be divided into two groups, i.e., the positive and negative change group. In terms of soil and water conservation, natural disturbances mostly brought about negative changes in the Taimali watershed, although there were some positive changes of dispersed vegetation recoveries.
To explore the relationship between the change trajectories and the environmental variables, the area of change in each subregion of environmental variables was measured by the percentage of the entire area of change and RCI. The results indicate that a large-area land cover change occurring was subject to the geologic condition of Pilushan Formation and the vicinities of faults and rivers. A great quantity of average annual precipitation, 4000-6000 mm, may lead to the movement of soil and rock to trigger landslides and channel expansion. In addition, land cover experienced highly concentrated changes at elevations of 1000-2500 m, gradients greater than 35°, and on the eastward slope, while the change intensity between the concave and the convex was not significant.