1. INTRODUCTION

Since the land use change affects many natural processes including soil erosion and sediment yield, floods and soil degradation and the chemical and physical properties of soil, so, different aspects of land use changes in the past and future should be considered particularly in the planning and decision-making. One of the most important applications of remote sensing is land use classification using satellite images. For this purpose, many algorithms have been developed. Since the generate maps with sufficient efficiency is the main purpose of processing images and thematic maps, therefore, selection of appropriate classification algorithm can play an important role in this regard. In recent years, it has been of interest for researchers to study land use changes, modeling and predicting these changes for the future due to the good performance of GIS systems and satellite data. Markov chain and cellular automata model is one of the current models to simulate land use map. This model, which is a combination of cellular automata model and Markov chain, is able to simulate land use changes with multiple features.

2- THEORETICAL FRAMEWORK

To evaluate the effectiveness of algorithms for mapping land use classification for past and present, three of Landsat satellite images were chosen from the years 1993 (TM), 2002 (ETM⁺) and 2014 (Landsat8) with almost an identical period. In this study, the middle of summer (July) was chosen as the time criterion for the selected images to minimize the effects of cloud cover and snow and also to improve the accuracy of training samples. At this time, the canopy of vegetation is maximum and cloud cover is minimum (Shahkooeei et al., 2014). One of the main steps is the prediction of the future land use map of basin by using Markov-cellular automata.

3- METHODOLOGY

In this study, firstly, the efficiencies of support vector machine (SVM), artificial neural network (MLP) and network self-organizing map (SOM) classification algorithms have been evaluated in the classification of Landsat satellite images. In addition, the efficiencies of the mentioned algorithms were compared. In the second step, for the prediction of future land use (2025), LCM model based on Markov chain and cellular automata was used. For this purpose, land use maps of 1993-2002 were applied for the calibration, and the 2002-2014 ones were implemented for model evaluation. Also, for producing the land suitability map, spatial variables involving distance from the road, distance from residential areas, distance from the mainstream, elevation and slope as the effective factors influencing on land use changes, were considered.

4- RESULTS

The results showed that SVM algorithm was superior with the total accuracy of 0.95 and kappa coefficient of 0.93 than the other two methods. In addition, the results of CA-Markov model for the period of 2002-2014 with the agreement of the 0.94 and Kappa index (Klocation) which showed the ability of the model to predict the position of the cells equal to 0.92, suggested a good performance of this approach to simulate future land-use map of the basin.

5- CONCLUSIONS & SUGGESTIONS

The estimated land use changes in 2025 attributes the exchanging of 60 percent of the gardens to urban areas. In other words, it is anticipated that approximately 420 hectares of orchards areas would be lost by 2025 in the study area. So, due to certain conditions of being in drylands, land management and land use changes must be further considered in comparison to those ones in the past.