Introduction

Introduction

Landscapes are dynamic with a mosaic of heterogeneous ecosystem elements that change in size, shape, and spatial arrangements due to complex and multi-scalar processes. Landscape dynamics involving changes in the structure of the landscape will have implications for ecosystem functions and processes. Land use land cover (LULC) changes leading to land degradation have escalated the carbon footprint while lowering the ability to sequester carbon, which is instrumental in accelerating global warming. Drivers of large-scale land use (LU) changes include regional economics, policies, culture, management, and environmental factors. LULC changes have been eroding ecosystem services, which is evident from the decline in biodiversity and alterations in the hydrologic regime, affecting the sustainability of natural resources (Vose et al. 2004; Polasky et al. 2011; Bharath et al. 2013; Vinay et al. 2013; Ramachandra et al. 2021). Alteration in bio-geochemical cycles has impaired the exchange of moisture, heat, and albedo at the local and global scales, intensely impacting climate feedback of the land surface. Degradation of forest landscapes results in lowered evapotranspiration, affecting the hydrologic regime and releasing the carbon stored in the soil and vegetation due to changes in the physical and chemical integrity of the ecosystem, thus contributing to higher levels of atmospheric greenhouse gases (GHG) (Ramachandra et al. 2020; Ramachandra and Bharath 2021). Unplanned developmental activities leading to LULC changes have affected the carrying capacity, which is evident from barren hilltops, conversion of perennial streams to intermittent or seasonal streams, and lower crop productivity.

The dynamics in landscapes, involving changes in the structure of forest ecosystems, have accelerated due to globalization (during the post-1990s) with the opening up of markets and consequent urbanization. During the past decade, global forests have witnessed a loss of 2.3 million sq. km. to cater to the demands of the burgeoning population coupled with unplanned developmental activities. Mitigation of these impacts entails actions toward the preservation of ecosystems at the local level. Knowledge of landscape dynamics and agents of LU transitions and LULC change analyses will provide valuable insights and aid in evolving prudent management strategies to sustain natural resources and mitigate climate change.

Urban expansions are taking place rapidly, with a higher rate of rural migration to the fringes and successively to core cities (Ramachandra and Aithal 2016). In recent decades, the urbanizing economy has contributed to urban sprawl. There is an expansion to peri-urban areas to develop residential layouts, create special economic zones, and transform ecological and agricultural spaces in cities into industrial setups (Ramachandra et al. 2012). LU models and advanced visualization of likely land use associated with policy decisions will help mitigate the impacts of rapid urbanization on vulnerable ecological spaces (Fu et al. 2018; Chandan et al. 2020; Dadashpoor and Panahi 2021). Scenario-based modeling helps in designing policies while taking into account the geomorphology (environment), connectivity (mobility), facilities, government (development plans), demographic and socio-economic factors, and sustainability (water bodies and green spaces) (Wahyudi and Liu 2013).

Spatially explicit data helps in understanding the heterogeneity within a landscape. This supports local and regional policy coordination by establishing a standard, agreed set of data and relationship between the environment and economic and human activity within the ecosystem accounting framework. The availability of coherent data supports the incorporation of environmental data into decision-making by business and finance sectors; this complements the wide range of initiatives underway in those sectors by recognising the importance of ecosystems and biodiversity. Thus, data from ecosystem accounts are used in conjunction with other methods and tools in modeling and scenario analyses to aid policy and decision-making. The exercise of informed decision-making and the use of scenarios to assess the outcomes and effectiveness of various policy intervention options is often termed policy scenario analysis (SEEA 2017).

Some of the simulation models used to visualize LU dynamics are Markov chain-cellular automata (MCA), SLEUTH, CLUE-S, multi-criteria evaluation (MCE), Fuzzy analytical hierarchy process (AHP) models, agent-based models (ABM), and CA-artificial neural network (ANN). The existing modeling techniques such as MCA, constrained-CA, MCE-CA, and logistic-CA have advantages such as the generation of richer forms of cells at individual class levels (Verburg et al. 2002; Batty 2005; Li and Liu 2006; Crooks 2010; Santé et al. 2010; Ramachandra et al. 2019). The lacunae with these models are calibration, optimization of factors (of cell state transitions), and intermixing of cells. CA models are simple, flexible, and intuitive (Eckhardt 1987; Batty and Xie 1994; Itami 1994; Torrens 2000; Bharath et al. 2021). Hence, CA is used widely in diverse fields ranging from landscape dynamics, flood modeling, forest fire modeling, medicine or biological modelling and urbanization, (Ermentrout and Edelstein-Keshet 1993; Bharath et al. 2014; Guidolin et al. 2016; Li and Gong 2016; Ramachandra and Bharath 2019b). The Fuzzy AHP CA is a cellular model tied to a system dynamics model and is considered very effective. It integrates qualitative knowledge with quantitative information, enabling the modeler to determine where the given LUs are likely to occur (Bharath et al. 2021). It represents individual decision-making with temporal and spatial dynamics more effectively compared to previous models (Mosadeghi et al. 2015). The rule-based MCA technique aids in modeling likely LU changes. Markov's approach provides information on transition probability between LU classes at a time (t) to LU at a time (t+1), and transitional area matrix concerning likely LU class (extent) changes. The transition probability matrix [P] and transition area matrix [A] are provided by equations 1 and 2.

(1)

where P_ij is the probability of i^th LU converting into j^th class during the transition period and n is the number of LU classes.

(2)

where A_ij is the area of i^th LU transition into j^th LU class.

Cellular automata aids in simulating and predicting land use (LU) changes based on transitional rules depending on the state of cell changes according to the neighbourhood cells (and the current cell's previous state), with interactions at the local and regional levels. LU change patterns follow Markovian random properties with various constraints that include the average stable transfer state of LU structure.

A different Markov is used for deriving the LU change probability of the study region. LU is predicted using CA with Markov chain as per equation 3. MCA provides better results by collectively incorporating transitional rules and probabilities.

(3)

where L _(t+1) = LU at t+1 and L _(t) = LU at t.

The MCA model-simulated LU is compared with the actual LU to compute prediction accuracy with the revised Kappa statistics (Pontius and Malanson 2005).

The relative weights for a group of factors considered in the model are developed through AHP by providing a series of pairwise comparisons of their relative importance to pixel suitability for the respective activity (Bunruamkaew and Murayam 2011; Ramachandra et al. 2021). AHP incorporates a measure of consistency for the individual comparison matrix of the decision problem (Ramachandra et al. 2017a). Pairwise comparison is made by considering two factors, and weights are assigned 0 to 9 scale. The consistency index computed as per equation 4 depends on the principal eigenvector (λ max). The consistency of pairwise comparisons or the consistency ratio (CR) is given by equation 5.

CI = (4)

CR = (5)

where λ max = ∑ (e𝑖𝑔𝑒𝑛 𝑣𝑒𝑐𝑡𝑜𝑟 ∗r𝑒𝑐𝑖𝑝𝑟𝑜𝑐𝑎𝑙 𝑚𝑎𝑡𝑟𝑖𝑥 𝑐𝑜𝑙. 𝑠𝑢𝑚)

Lower values of CR are ideal for high accuracy; CR of 0.10 is preferred to attain minimum accuracy (Saaty 1980). MCE helps generate site suitability maps for future LU prediction in CA models (Clarke 2008; Spencer 2009).

Through an appraisal of changing behaviour, ABM considers individual actions of various agents in the simulation (Holland and Sigmund 1995; Franklin and Graesser 1996; Axtell 2000; Macal and North 2009; Crooks and Heppenstall 2012). Agents need to learn, engage in dynamic relationships with other agents, adding a spatial component to identify behavior and interactions of agents in the real-world environment, adapt and change behaviours. The interaction of agents with the landscape leads to likely relations such as agents’ self-influencing behaviour, their effect on the landscape, self-influencing state of a landscape at any given time, and landscape’s effect on agents (Crooks and Heppenstall 2012). The advantages of ABM compared to traditional modeling techniques are that it: (a) captures the emerging spatio-temporal footprint of LU changes; (b) provides a domain-specific environment of the region under investigation; and (c) is highly flexible for developing geo-spatial models (Bernard 1999). The advantages of Fuzzy AHP CA over conventional modeling techniques are that it: (a) encompasses dynamic spatial transitions through the integration of influence of distances of each factor on respective LU, unlike considering a LU category; (b) links macro to micro driver’s responses, taking into account social, economic, dynamic and spatio-temporal dimensions; (iii) prioritises comparison ratios for decision-making, based on site suitability (probability of a cell changing to a given class in future) and state of the neighbouring pixels; and (iv) is simple and provides improved visualization by translating qualitative assessment into quantitative data and delivering more logical and precise results.

The objectives of the current research endeavour are to account for the spatio-temporal LU changes for Karnataka state and evaluate the likely LU changes under five policy scenarios, using the Fuzzy AHP MCA model for improved decision-making.