2.7 Prediction of land cover/ use change
Land cover/ use change can be determined by biophysical, technological, economic, social and political drivers but also by trigger events (Bürgiet al. , 2004; Geist & Lambin, 2002; Hersperger et al. , 2010). Here we used biophysical site conditions to predict the spatial characteristics of LCLU change from 1985 to 2016 with the Terrset Land Change Modeler, where we developed several sub-models. Based on potential transition scenarios, following “business-as-usual” namely, rangeland to other land uses and agriculture to residences, time-dependent (dynamic) and time-independent (static) descriptive variables were determined in order to assess the transition potentials of pixels of one land use/cover to another (De Rosa et al. , 2016). Cramer coefficient, which ranges from “0” (no correlation) to “1” (strong correlation), was used to determine the correlation of model variables (independent variables) with LCLU classes (dependent variables). The prediction of LCLU transition was made separately for five transition classes, namely “rangeland to residences”, “rangeland to forest”, “agriculture to residences”, “rangeland to agriculture”, and “rangeland to industry”. LCLU change was predicted for 2016 and 2036 with a multi-layer perceptron (MLP) neural network. To determine the amount of change that will occur up to 2036, we used the Markov chain prediction process.
Several spatial biophysical and socioeconomic variables, namely, distance to water, residences, industries, agriculture and rangelands (based on classification results for 1985 and 2016), slope, distance to roads and flood potential, determined the likelihood of LCLU change transitions (figure 6). The slope was calculated based on a 30-meter shuttle radar topography mission (SRTM) digital elevation model (DEM) (figure 6). The Euclidian distance to roads was calculated based on the digital map of main roads. Similarly, the Euclidian distance to water was calculated with the account of the Zayandehrood River and the main water channels. We also assessed the susceptibility to floods (flood potential), which was determined based on the calculation of the distances to seasonal and permanent water flows based on hydrologic maps for the study area. All produced maps were calculated with a 30-meter pixel size to match the LULCC maps.
Results