Revealing past and future land-cover transitions from 1985 to 2036 in the drylands of Central Iran
Bibizahra Mazloum1∗, Saied Pourmanafi1, Alireza Soffianian1 and Abdollrasoul Salman Mahini2, Alexander V. Prishchepov3
1. College of Natural Resources, Isfahan University of Technology, Iran
2 Department of Environment, Gorgan University of Agricultural Sciences & Natural Resources, Iran
3 Department of Geosciences and Natural Resource Management (IGN), University of Copenhagen, Øster Voldgade 10, DK-1350 København K, Denmark
* Corresponding author: bibizahramazloum@gmail.com, phone: +4535331386
Abstract
Land serves as a vital production resource, and therefore, land planning plays an important role in sustainable land-use design. Increasing the global population alters landscapes via land-use and land-cover change across different landscapes, including the drylands. Iran includes large areas of dryland, where the population increased by 60% from 1985 to 2016. Further population increase in Iran would require more land resources to be allocated for human needs. However, the pace and patterns of these changes remain unclear. The aim of this study was to map land-cover change from 1985 to 2016 and predict future land-cover change in the Zayandehrood ecologic sub-basins of Central Iran. By using multiseasonal Landsat imagery, nine thematic classes were mapped with a random forest classifier for 1985, 1998, and 2016 with an overall accuracy of 80% for each period. Classification results revealed that from 1985 to 2016 residential areas doubled and industrial areas increased at the expense of rangelands. Our study also revealed cropland expansion at the expense of rangelands, cropland abandonment and contraction of croplands due to residential and industrial development. Prediction of changes by 2036 with a multi-layer perceptron neural network and Markov chain analysis revealed further expansion of industries and residencies particularly nearby the protected areas such as Ghamashlu Wildlife Refuge. Predicted contraction of some degraded agricultural lands and concomitant agricultural expansion in the agricultural frontier by 2036, underscore the importance of sustainable land management in highly arid areas of Iran and improvement of the strategies for the protection of rangelands.
Introduction
Land serves as a vital production resource and therefore land planning plays an important role in sustainable land use design (Foley et al. , 2011; Lambin & Meyfroidt, 2011). Human-induced land-use and land-cover change (LULCC) are the key proximate causes that shape landscapes at different scales (Foley, 2005; Lambin & Geist, 2006). Evidence shows that LULCC increased over the last 100 years due to population growth, and subsequent increasing land demand and changing consumption patterns, among other factors (Gerbens-Leenes & Nonhebel, 2005; Goldewijk, 2001). It is envisaged that the purpose of land planning is to direct the agents that change land use (e.g., landowners, farmers) to a point that results in the balance between meeting societal needs and protection of the environment. Hence, the quantitative assessment of the state of land cover and land use and revealing the potential changes of land cover is crucial for better sustainable land-use planning (Herrmann & Osinski, 1999).
Earth observation, such as satellite remote sensing, is a prominent approach to monitor land changes and produce land-cover maps (Cohen & Goward, 2004; Coppin et al. , 2004; Hansen & Loveland, 2012). There has been substantive progress over the last three decades with the aid of remote sensing to monitor how humans appropriate land; for instance, the dynamic of residential and industrial areas (Reba & Seto, 2020), agriculture and forestry activities (Eisavi et al. , 2015; Hansen et al. , 2013; Rufin et al. , 2019), and also the intensity of appropriation of rangelands (Dara et al. , 2020; Jakimow et al. , 2018), including drylands (Dubovyk, 2017). In the past, scientists have relied on field surveys and interpretation of aerial photographs to map LULCC, often over smaller areas (Nebikeret al. , 2014). As the size of the area of study becomes bigger, field surveys and interpretation of aerial photos become very costly and time-consuming. Therefore, satellite remote sensing is an excellent source of information to study LULCC at local and regional scales because satellite images, from on-board of Landsat satellites for instance, can cover a large extent and have a higher temporal coverage compared to field based observations and aerial surveys (Hansen & Loveland, 2012; Loveland & Dwyer, 2012). Landsat images from the freely accessible United States Geological Survey (U.S.G.S.) archive are used widely to evaluate the quantitative parameters of land, especially land use/cover change due to long historical repetitive coverage, the spatial resolution of Landsat images that match LULCC phenomena and relevant radiometric resolution suited to map vegetation change, including drylands and rangelands (Eisavi et al. , 2015; Ridwan et al. , 2018; Roy et al. , 2014). Moreover, the change of paradigm about free accessibility of Landsat records made this data relevant for the scientific community to monitor LULCC (Wulder et al. , 2012).
There has also been progress in the application of semi-automatic classification methods to derive thematic classes from satellite imagery and produce LULCC maps. Machine-learning algorithms gained popularity recently to classify satellite imagery due to the ability to handle non-linearity among spectrally complex classes, such as various urban landscapes, agriculture and rangelands (Dara et al. , 2020; Khatami et al. , 2016; Kraemer et al. , 2015; Löw et al. , 2015). Random forest, which is probably the most popular machine-learning classification method, is an ensemble classifier, where, instead of one single tree (CART and C4.5), multiple trees from training data are produced. Such an approach allows the model overfit to be reduced (Belgiu & Drăguţ, 2016; Gislason et al. , 2006). At the same, compared to CART no pruning of produced trees is implemented in a random forest, which allows increasing classification accuracy with random forest (Belgiu & Drăguţ, 2016; Gislason et al. , 2006). High precision, ability to learn nonlinear relationships, ability to determine variables, non-parametric nature, high computation speed, and resistance to over-fitting are among the advantages of random forest classifier (Belgiu & Drăguţ, 2016; Gislason et al. , 2006; Millard & Richardson, 2015). It has been also noted, classification error decreases by increasing the number of trees; both the number of trees and the number of utilized variables for branches of each node can be parameterized (Belgiu & Drăguţ, 2016; Gislason et al. , 2006). The random forest classifier has been widely used over the last two decades for land-change mapping (Chen et al. , 2019; Jakimowet al. , 2018; Millard & Richardson, 2015). In sum, a random forest classifier can be a promising classifier to map LULCC in spectrally complex areas, such as the drylands of Iran (Eisavi et al. , 2015).
Various models have been used for investigating the dynamics of land use and predicting the land-use changes in order to improve decision-making (Castella & Verburg, 2007; Verburg et al. , 2006; Zhang et al. , 2015). Some models, especially dynamic models such as Markov, dynamic systems, automatic cells, multi-agent models, and conversion of land use and its effects at small regional extent (CLUE-S), can help the decision-makers to evaluate potential land-cover change patterns under the different scenarios (Mallampalli et al. , 2016; Sun et al. , 2016; Verburg et al. , 2002). Each of these models has advantages and disadvantages for predicting land-cover change. For instance, geometry modification technology (GEOMOD) can model only two land-cover change classes (Pontius Jr & Chen, 2006) and SLEUTH (slope, land use, exclusion, urban extent, transportation and hillshade) is primarily used for urban development and cannot capture the driving forces behind urban growth. Additionally, SLEUTH has limitations on the number of input variables and can be implemented only under the UNIX system (Chaudhuri & Clarke, 2013; Sekovski et al. , 2015). SLEUTH is deliberately focused more on form and dynamics, i.e., “where” could the development take place, but not “why” (e.g., demography and socioeconomic processes) (Sekovski et al. , 2015).
Land Change Modeler, available via Terrset software (former Idrisi software, Clark Labs) (Eastman, 2015), can model multiple land-cover change trajectories simultaneously in comparison to the above methods and tools. One of its advantages is that it is available via Terrset and can be run with Windows operating system, therefore it is more available for a broader community. It has no limitation on the number of input variables. It provides more capabilities for tuning than other methods, in addition to its simplicity. Land Change Modeler has been used to evaluate and model various land-cover changes in Iran (Ansari & Golabi, 2019).
Unsustainable land-use activities may result in substantial negative ecological consequences with cascade effects even when land-use change occurs at the local level (Newbold et al. , 2015; Yaghobi et al. , 2019). Reduction of the negative environmental impacts of land-use change while maintaining economic viability and social acceptability is a particular challenge for most developing countries in Asia, including Iran (Zhao et al. , 2006). For instance, in Iran, the population increased by 60% from 50 million to 80 million people just within a 30-year course from 1986 to 2016 (Danaei et al. , 2019). Over the same course, the urban population increased and by 2016, it comprised 70% (Danaei et al. , 2019). Most of the territory of Iran can be considered dryland with vast rangelands and irrigated croplands lying in semi-desert conditions (Mesgaran et al. , 2017; The Cambridge History of Iran , 1968). There has been documented evidence of environmental degradation caused by unsustainable urban and agricultural expansion (Minaei et al. , 2018; Yaghobi et al. , 2019; Zareie et al. , 2016). This all increased the concern of the scientific community on the negative impact on the environment and the relevance of quantification and prediction of such impacts.
The ultimate goal of this study was to assess land-cover change with the aid of multiseasonal Landsat satellite imagery from 1985 to 2016 in Zayandehrood ecologic sub-basins of Central Iran, an area of rapid agricultural and urban/industrial development and to determine the required land amount for various land uses in the future. Our specific research questions were:
  1. What are the major trajectories of land-cover change in Zayandehrood ecologic sub-basins of Central Iran from 1985 to 2016?
  2. What will be the future land-cover/ use transitions in the study area by 2036?
Methods
2.1 Study area
The study area includes the three sub-basins of Zayandehrood: Lanjanat, Najafabad, and Mahyar Shomali, which are located at 50 57’ to 51 51’ of eastern longitude and 31 45’ to 32 50’ of northern latitude with an area of 527,148 hectares in the Isfahan province of Iran (figure 1).
The population in the study area increased from 321,000 in 1996 to 2.26 million people in 2016 (https://www.isfahan.irib.ir). The climate ranges from arid and semi-arid (east and southeast) to cold semi-humid (west and southwest) with an elevation range from 1,569 m (plains, hills) to 3,235 m (mountains). Rainfall increases with elevation, while temperature decreases. The mean annual temperature is 13 °C, and the mean annual rainfall is 209 mm. The Zayandehrood River is the biggest and most important river in the study site, which originates in the Zagros Mountains (figure 1). This river provides water for farming, residential areas and industries in the study site and finally goes to the Gavkhoni wetland in the east of the watershed (Akbari et al. , 2015; Iranmehr et al. , 2015). The Ghamashlu National Park in the north and Kola Ghazi National Park in the Southeast host charismatic herbivores, Capra aegagrus and Gazella subgutturosa . Among protected native plants in the study area are Fritillaria imperialis and Pistacia atlantica . Therefore, despite semi-arid conditions, the study area has high biodiversity and importance for environmental protection. Different land uses, such as residential areas, agriculture, industries and mining, emphasize the importance of obtaining accurate information about the state of land use and of developing strategies for sustainable land use in the future.
<Figure 1>
2.2 Data
In this study, U.S.G.S. Landsat multiseasonal imagery for the years circa 1985, 1998, and 2016 were used to match the development of land-planning policy over the last decades in Iran. Landsat images are well-suited to distinguish different land-cover types due to the spectral and spatial resolution of the sensors (TM, ETM+, OLI) on board Landsat satellites, and also due to the temporal revisits. To determine different phonological stages of spring and summer crops, image-dates for May and August were selected. Systematically corrected (topography and atmosphere) level 2 Landsat imagery from TM and OLI sensors were acquired from U.S.G.S. for Word Reference System -2 (WRS-2) path 37 and 38 and row 164. We then mosaicked satellite imagery from different paths.