Forecasting Land Use Change

18 cell is mainly measured with straight-line distance to the specific location, where the straight-line is measured by using Euclidean distance to the target location. It is predicted that proximity factors become more dominant in this study, because of relatively smooth topography of the study area. Distance to facilities, and distance to road become important determinants in predicting land use change. Furthermore, elevation data is considered important in this area, although it has little variation. Elevation is significant for estate and settlement land use area, where it is associated with the distribution of irrigation water to agricultural crops and landscape for residential area. Another fact is that the density of population is measured by the relative density, not an absolute density. Density is calculated based on the number population person divided by the village area Km 2 . This is not a best-fit method to measure the density where density should be measured in urban area or named absolute density. However, the calculation of the absolute density of urban areas land use cannot be applied to logistic regression, because it will produce data that are bias, where the regression calculation applied to the same extent of areas.

2.3.3.3. Statistical Analysis

Logistic regression is divided in two types, including binomial and multinomial regression. Binomial regression, which is employed in the research, uses dichotomous value in the dependent variable, whereas the type of independent variable could be categorical or continuous. In land use analysis, logistic regression is used to examine the relation between land use and possible driving factors. The results of this analysis are coefficient values that show the contribution of each driving factor to land use change. The formula is: � − = � + � � + � � + ⋯ + � � � � Verburg 2002. Where : is the land use change probability � �… are independent factors � is coefficient value 19 To achieve the validity of land use change estimation, the model should be supported by the procedure to identify the driving factors that are statistically independent and to determine the significance of driving factors. One of methods to identify the driving factors that have significant contribution to land use pattern is the stepwise procedure. In the stepwise procedure, all of driving factors are involved in one step and eliminated according to their significance values. A driving factor that has a lower value than the significant threshold will be excluded from the analysis. In forward procedure, the analysis starts with one factor and continues to other factors respectively. ROC Relative Operating Characteristics is a method to measure the goodness of the statistical model. The probability of each land use resulted from logistic regression is compared to the real land use map to calculate the equal category of each grid cell between those maps. According to Pontius and Schneider 2001, this method will depict the capability of regression equation to represent land use characteristics. The range of ROC value is between 0 – 1, where ROC value below 0,5 is categorized in low or completely random, 0,5 – 0,6 is good, 0,6 – 0.99 is very good and 1.0 is fitperfect. The goodness of the logistic regression equation to represent land use condition indicates the suitability of driving factors as determinant of land use change.

2.3.3.4. Land Use Type Specific Conversion

Conversion setting for specific land use type is addressed to determine the temporal dynamic of the simulation by using reversibility of land use changes. This method will be implemented by using three different decision rules that represent the situation of study area: 1. Some land use types are unlikely to be converted into another land use type after first conversion. 2. Other land use types are converted more easily. Forest and grassland are more likely to be converted into another land-use type soon after their initial conversion without any restrictions.