Prediction of Land Use Changes Using the Markov Chain and Cellular Automata Model (Case Study: Hyrcanian Region)

Background and Objectives: Modeling and predicting land use changes is essential for sustainable land use and awareness of potential future changes. Detection and modeling of such changes using satellite image processing serve as effective tools for understanding environmental transformations associated with human activities. The use of Geographic Information Systems (GIS) and Remote Sensing (RS) provides systematic and accurate information about land surface phenomena. Monitoring the trend of these changes for sustainable land use can be successfully achieved using multi-temporal remote sensing data. Given the ecological significance of the Hyrcanian region- due to its genetic diversity and ecological and touristic functions- understanding the extent of land use changes and conversions is a key aspect of land use planning and sustainable development. Moreover, since land use change monitoring in the past and its future prediction have not yet been conducted across the entire Hyrcanian region, this study aims to simulate and predict land use/land cover (LULC) changes from past to future by integrating the Markov chain model with Cellular Automata (CA), known as the "Markov-CA" approach. Another novelty of this study lies in the implementation of image processing and modeling workflows within the cloud-based Google Earth Engine (GEE) environment.
Materials and Methods: To detect and model land use changes, Landsat 5 TM images from the 2000, 2005 and 2010, and Landsat 8 OLI images from 2015 and 2020 were analyzed and used. After applying spatial, temporal, and cloud filters (with cloud cover less than 5%), all time-series images were combined using a mean filter in the Google Earth Engine platform, and then classified into six classes (i.e. forest, rangeland, agriculture, built-up areas, bare land, and water bodies) using the supervised Random Forest machine learning algorithm. Classification accuracy was calculated based on the Kappa index and overall accuracy. To simulate land use for the year 2020, land use maps from 2000 and 2010 were utilized in conjunction with a combined Cellular Automata model and logistic regression-based transition rules, implemented in MATLAB 2021a. Spatial variables including Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI) and Normalized Difference Built-up Index (NDBI) indices, as well as elevation and slope, were used as influencing factors on changes in the Random Forest algorithm. To ensure the representativeness of the selected land use classes, efforts were made to achieve a homogeneous and well-distributed sampling across the study area. Seventy percent of the visually interpreted samples from satellite imagery were used for model implementation (training samples), while the remaining 30 percent were employed for accuracy assessment (test or validation samples). Additionally, the Markov-CA model was integrated to predict land use changes in the Hyrcanian region for the years 2070 and 2100.
Results: Based on the spatial accuracy assessment, for the period 2000 to 2020, the overall accuracy and Kappa coefficient were higher than 80%, indicating a high agreement between the model-predicted images and ground reality and also the efficiency of the Google Earth Engine platform in processing and classifying Landsat imagery. During the 2000-2010 period, a 3.8% decrease in forest cover was observed, primarily due to its conversion into rangelands and agriculture. In the subsequent period (2010–2020), the most significant land use conversion was from rangeland to agriculture, covering an area of 78,872 ha. Using the cellular automata model and logistic regression rules, the Kappa coefficient for land use predictions in 2070 and 2100 was estimated at 76% and 72%, respectively. The Kappa coefficient obtained from comparing these two maps was 87.31%, demonstrating the high capability of the Markov-CA model in simulating future land use changes. The predicted land use changes using the Markov transition estimator showed the degradation and reduction in forest, rangeland, and water bodies in 2070 and 2100. As a result of changes in land use/land cover between 2020 and 2070, forested areas are expected to decrease by 1,686,80 ha compared to the 2000-2020 period. Water bodies are also projected to decline by 2070, with approximately 8,667 ha expected to be converted into rangelands. By 2100, over a 30-year period, forest areas are projected to decrease by 8,900 ha, primarily transitioning into water bodies and barren lands. The most significant land-use changes in the study area include the conversion of forests to rangelands and agriculture, the transformation of rangelands into agriculture and built-up areas, and the conversion of water bodies into barren lands. In fact, if the current trend of land use changes and improper conversions continue, natural resource areas will be drastically reduced.
Conclusion: Overall, the accuracy assessment based on indices derived from the error matrix indicated that the Google Earth Engine platform, through processing multi-temporal Landsat data, proved to be an effective tool for evaluating and monitoring land use changes. The observed trend in the study area showed that agricultural lands, built-up areas, and bare lands had an increasing trend, primarily due to the conversion of other land use types into these classes. In contrast, among all land-cover types, water bodies experienced the most significant decline compared to forests and rangelands. The continuation of this degradation trend is expected to lead to adverse environmental consequences. The increase in built-up areas and agricultural lands can be justified by population growth and urbanization to meet the food needs of this population, as well as proper management in agriculture. Since by examining land use changes, trends related to degradation, deforestation, desertification, and loss of biodiversity in a region are identified, awareness of these changes over time, it is possible future challenges and implement preventive measures for better management.