Land Cover Change Modeling Using a Hybrid Approach and Remote Sensing Data

Wetlands are vital for human survival and constitute 40.6% of the total value of global ecosystem services. This study analyzed land cover changes in the Golpayegan Shoor wetland from 1988 to 2018 and then modeled it with a combination of artificial neural network, Markov chain, and multi-objective optimization for 2030 and 2040 for obtaining environmental sustainability. To obtain the highest accuracy in transition potential modeling, four scenarios were used with various calibration periods and sub-models. The FOM index was used to evaluate the accuracy of various scenarios using the approach by overlaying three maps for the years 2008, actual 2018, and predicted 2018 maps. Effective variables in describing the changes occurred in the case study selected using Cramer's v. The highest accuracy was shown in a scenario with the calibration period of 1988-1998 and 9 sub-models in land cover change modeling 2018 year. The evidence likelihood of change map and distance from agricultural lands showed the highest impact in the changes of the study area. Residential areas and agricultural lands increased by 0.6% and 1.8% in 2030, while 1% and 2.2% grew up in 2040 compared to 2018, respectively. Dense rangeland decreased by 13.1% and 18.6%, respectively in 2030 and 2040 compared to 2018. The water bodies showed an increase of 0.18% and 0.27% due to the conversion of agricultural areas and rangeland to Lake Golpayegan and Kochery dams. Taking water from unauthorized wells, reducing the flow of water entering the wetland, and the presence of decorative ore has caused serious problems for Golpayegan Shoor Wetland, which requires managers and land planners to consider compensatory ways to restore, water requirement, and ecological sustainability for the wetland.
 
Extended Abstract
1-Introduction
Wetlands are the most important biodiversity areas in the world. Their ecosystems are unique and productive where land and water habitats meet and have more biodiversity, nutrient recycling, and special ecological niches than other ecosystems. It provides multiple ecosystem services that support water security and provides benefits and values to society and the economy (Gleason et al., 2011). The value of coastal and inland wetland ecosystem services is usually higher than other types of ecosystems. The size of global wetlands is estimated from 5.3 to 12.8 million square kilometers. However, wetlands are declining rapidly, and researchers estimate that more than 50 percent of the world's wetlands have declined (Tiné et al., 2019). Monitoring the changes in wetlands during the past decades until now and identifying the direction of these changes is very necessary for managing and how to exploit wetlands and providing solutions that will prevent them from being included in the Monterey list (Karami and Mirsanjari, 2018). Golpayegan Shoor wetland has dried up since 30 years ago due to various reasons such as lack of attention to the environmental water requirement, excessive exploitation of underground water resources, reduction in the volume of the wetland aquifer, and drought (Aazami et al., 2018). Considering the environmental importance of the Golpayegan Shoor wetland and the land cover changes that have occurred in it, there is a need to formulate strategic plans for the land managers in order to achieve sustainable ecological development of the watershed and to prevent further destruction of these areas and the continuity of ecosystem services provided by the wetland is felt more than ever. Monitoring and modeling the land cover changes of Golpayegan Shoor wetland is necessary in order to achieve sustainable management of this wetland. The aim of the current research is to design and apply a computational model combining artificial neural network, Markov chain, and multi-purpose optimization in simulating the changes of Golpayegan Shoor wetland for 2030, and 2040 and evaluating the accuracy of artificial neural network in transition potential modeling.
 2-Materials and Methods
Landsat 5 sensor TM satellite images for 1988, 1998, and 2008 and Landsat 8 OLI sensor for 2018 were used to produce land cover maps. In the pre-processing stage, geometric and atmospheric correction was done. For geometric correction, topographic maps with a scale of 1:25000 and 20 ground control points were used to implement the polynomial equation. Dark-object subtraction method was used for atmospheric correction (Rodríguez Eraso et al., 2013). Training sites for six classes including agricultural lands, residential areas, water bodies, dense rangeland, semi-dense rangeland, and low-density rangeland (Farajzadeh et al., 2010) using field visits, false and real color composite images, and Google Earth were selected. Then, the maps were produced using the maximum likelihood classifier.
The current research was conducted using the Land Change Modeler (LCM). Analysis of changes in the study area was calculated using LCM for the studied time periods. These changes include the transition from one class to another, gain, loss, and net changes that indicate the difference between decrease and increase for each land cover class.
The present study integrates the approaches of artificial neural network, Markov chain, and multi-objective optimization into a single model. In the present study, the artificial neural network was used to model the transition potential maps. In the present research, the number of executions is considered to be 10,000 repetitions to achieve the highest accuracy. Accuracy rate and skill measures were also used to evaluate the neural network. We used the Markov chain model to estimate the amount of change that can occur in 2030 and 2040 based on previous land cover maps and the probability of change from one land cover class to another. A forecast map for 2018 was produced using the Markov chain model to evaluate the accuracy of the model (Islam et al., 2018).
Four scenarios were considered to achieve the highest accuracy in this study. In all scenarios, 9 sub-models that include the transition from one class to another type were selected for modeling. The difference between the scenarios is in the used calibration period and different sub-models. The calibration period of the first and second scenarios is 1988-1998 and the calibration period of the third and fourth scenarios is 1998-2008.
 
3- Results and Discussion
Residential areas (1355 ha), water bodies (660 ha), and agricultural lands (6557 ha) increased, and the dense rangeland (1354 ha), semi-dense rangeland (5429 ha) and Low-density rangeland (1790ha) also showed a decrease during 1988-2018. During these 30 years, agricultural areas (76 ha), semi-dense rangeland (324 ha), and dense rangeland (956 ha) were completely changed to residential areas. The amount of net change from dense rangeland, semi-dense rangeland, and low-density rangeland to agricultural land is 5417, 1201, and 287 ha, respectively. Also, agricultural lands, semi-dense rangeland, and dense rangeland showed a net change of 287, 311, and 61 ha to water bodies.
4- Conclusion
In the border area of Golpayegan Shoor wetland, parts of dense rangeland were converted to low-density and semi-dense rangeland, and also areas of dense rangeland were transferred to agricultural lands, which indicates the drying and destruction of the wetland. A decrease in the flow of water to the Golpayegan Shoor wetland has occurred, which is one of the reasons for the drying up of this wetland.