Modeling changes in land use using the integration of MLP-NN, CA-Markov models and GIS for settlement development in Tembalang District

Abstract

The effects of dynamic land use change can be modeled by Artificial Neural Network (ANN) models with Multi-layer Perceptron (MLP) network architecture and backpropagation algorithms. The setting process and predictive capabilities testing will be generated by the models with a combination of MLP-NN, CA-Markov methods and Geographic Iinformation System (GIS) from several previous studies so the results is accurate. This study aims to analisys land use change in Tembalang District in 2010 as first period models, 2014 as period models, and land use in 2018 as validation data, make a model of land use change with ANN methods and projection land use in Tembalang District in 2026. CA-Markov models are used for future projections. In modeling land use change, several driving force variables are used, namely distance to roads, rivers, settlements, and population density. This research are using maps for 2010, 2014 and 2018 from digitization process of hight resolution satellite imagery and validation land use in the field. In this research, data on land use change from 2010 to 2018 is dominated by land use changes from vacant land to settlements and housing. Settlement increased by 2,13%, housing increased by 102,69%, and vacant land already allocate increased by 47,32% and vacant land decreased by -61,18%. The results of modeling validation have a Kappa index of 0.959, an the root mean square value is 2.579 m that means this value is accepted, and 85% of the area between the prediction map and the digitization map are said to be appropriate, so this model is classified as having very good similarities with existing land use conditions in 2018. Overall prediction results show that land suitability is 70.52% and 29.48% of land is not in accordance with Semarang City RTRW map for 2011-2031.