Analysis

This research started with analyzing LULC change during the study period. Super­vised classification, the most widely used technique for quantitative analysis of remote sensing image data (P6cas et al. 2011), was used to perform image classi­fication. To evaluate the accuracy of the classified images, an accuracy assessment

Fig. 14.1 Batang Merao watershed showing the elevation at 25-m and 100-m contour intervals

or confusion contingency matrix was implemented. The error matrix compared the linkages between known reference data (ground truth) and the corresponding results of an automated classification (Congalton 1991). The kappa coefficient, the value for estimating how accurate the remotely sensed classification is in the reference data (Jensen 2004), was used for accuracy assessment. The kappa (Khat) is computed as follows:

Data	Description	Source
Landsat TM	Path 126/row 61 July 13, 1990 May 5, 2000 June 18, 2010	http://glovis. usgs. gov/
DEM	Aster GDEM 30 m	http://demex. cr. usgs. gov/
Administrative	Jambi Province	Geospatial Information
	Kerinci regency	Agency of Indonesia (BIG)
	Watershed boundary	Planning Agency of Jambi
Soil series map	Soil type 1990, 2000, 2010	Planning Agency of Kerinci
Rainfall	Monthly rainfall and distribution 1990-2010	Forestry office of Batanghari (BPDAS)
Land use planning map	Regional land use planning
Ground truth	Ground truth 2011	Field survey

Table 14.1 Data collection and its description

where r is the number of rows in the error matrix, xt is the number of observations in row i and column i, xi+ and x+i are the marginal totals for the row i and column i, respectively, and N is the total number of pixels. Value greater than 0.80 represent strong agreement or good classification performance, values between 0.40 and 0.80 indicated moderate agreement, and values less than 0.40 represent poor agreement (Jensen 2004).

LULC classification was modified from the LULC categories of the Indonesian National Standard No. 7645:2010 specified by the National Standard Agency of Indonesia, which refers to the FAO’s land cover classification system and ISO 19144-1 (BSN-National Standardization Agency of Indonesia 2010). Because of differences in scale, accuracy, and type of LULC categories, synchronization and generalization were performed, resulting in six categories: forest, mixed plantation, tea plantation, shrub/bush, agricultural land, and settlement (Table 14.2).

Potential soil erosion was determined using the USLE method (Wischmeier and Smith 1978). This is the most widely used method and the simplest model for predicting soil erosion and still remains the best known because of its scientific basis, ease of use, low cost, and direct applicability in watershed (Sharma et al. 2011). It can be described in the following equation:

A = R x K x L x S x C x P, (14.2)

where A is the annual soil loss (tons ha-1 year-1), R is the rainfall/erosivity factor (mm year-1), K is the soil erodibility factor (t J-1 mm-‘), L and S are slope length and steepness factors, respectively, and C and P are the crop management and conservation factors.

General description

Areas covered by dense trees with relatively darker green color An intensively managed plantation. It is characterized by a homogeneous canopy structure with single dominant species and regular spatial network and clear-cut boundaries with neighboring vegetation Areas covered by a combination of several woody and fruity plantations such as rubber, Cynnamomun, Mango sp., Durio sp., etc. It is indicated by irregular patterns

Areas covered by herbs, grass, and nonwoody herbs. These areas usually

correspond to recently opened areas, the first phase of land conversion into both mixed plantation and tea plantation Areas dominantly cultivated by paddy field and potatoes, characterized by inundating of fields from irrigation or rainfall Areas occupied by houses or buildings including road network and other facilities

Source: The Indonesian National Standard No. 7645:2010 by the BSN-National Standardization Agency of Indonesia 2010

Statistical analysis with Pearson’s correlation coefficient was applied to identify the relationship between LULC change and land degradation. In this analysis, the land use category served as the independent variable and the rate of land degrada­tion served as the dependent variable. Because of the difference in scale between land use type (nominal scale) and land degradation class (ordinal scale), the land degradation class needed to be aggregated using the statistics method “summated rating scale.” The next step was bivariate statistics correlation analysis between land use degradation to identify the relationship between LULC change and land degradation.

All image processing, classification, and change detection were performed using ERDAS Imagine 8.7; the GIS analysis was carried out using ArcGIS 10.1. Statis­tical analyses were conducted using PSAW SPSS Statistics ver. 18.0 for Windows.

14.2 Results