The utility of a hybrid GEOMOD-Markov Chain model of land-use change in the context of highly water-demanding agriculture in a semi-arid region

Highlights

• We developed a hybrid spatio-temporal model of land-use/cover change.

• Orchard development was simulated and projected relative to water resources.

• Model assessment reported more spatial error than land-change quantity error.

• By 2030, orchard cover is projected to increase by 20% of its 2015 area.

Abstract

Land change simulation for highly water-demanding crops may prove a valuable tool to guide integrated land-and-water management in semi-arid regions facing water scarcity. We explored this premise by mapping and modelling past (1985–2015) and future (2015–2030) orchard development relative to water resources and other factors in Iran. We employed a hybrid GEOMOD-Markov Chain model whereby both the spatial allocation and quantity of orchard development were simulated. By 2030, orchard cover is projected to increase by 20% of its 2015 area, straining limited water resources. To gauge the accuracy of our projection of orchard gain to 2030, we assessed a comparable simulation of orchard gain for 2000–2015 according to the various components the Figure of Merit (FOM) metric. Misses, Hits and False Alarms of simulated orchard gain accounted for 1.84%, 0.45% and 0.74% of the study area respectively over 2000–2015 at a 200-m spatial resolution, for which the FOM was appreciable (15%) given the limited extent of simulated orchard gain and actual orchard cover across the study region (1.2% and 5.3%). With respect to orchard gain, spatial allocation error was more than land-change quantity error at 200-m resolution, at 1.48% and 1.10% of the study area, respectively. Predicting the location of agricultural change remains a priority and challenge for model utility, given scant agricultural footprints in semi-arid regions and their large draw on limited water resources. Results also indicate the importance of incorporating dynamic water availability and demand over the course of agricultural expansion, including shifts in the location preference amongst farmers. The integration of dynamic, agent-based models within our GEOMOD-Markov Chain framework is therefore methodologically appealing, but would adversely increase complexity for policymakers.

Introduction

There are increasingly acute pressures on water resources globally, such as population growth, climate change, and land change (LUC). An evaluation of these trends in the past, present, and potential future may support decision makers in managing water resources. Regarding LUC, the expansion of highly water-demanding land-use/covers will negatively impact surface water availability and regional watershed health (Schulz et al., 2010; Tourian et al., 2015; Andaryani et al., 2019a and b). For instance, in the arid and semi-arid regions of northwestern Iran, the increase of high-consuming agricultural land uses without land-use planning or water management may have irreparable consequences for Urmia Lake, a critical source of irrigation and drinking water threatened by climate change (Chaudhari et al., 2018; FAO, 2017; Ghale et al., 2018). For this region, we developed a hybrid spatio-temporal model of land change to project the continued expansion of highly water-consuming land use relative to critical water and soil resources.

Water scarcity has increased across Iran, the Middle East, and globally during recent decades (Ardakanian, 2005; Procházka et al., 2018; Vorosmarty et al., 2000; Wu et al., 2017). There is a need to understand the spatio-temporal dynamics of LUC in water-stressed regions in order to support sustainable integrated land-and-water management. This is particularly true for semi-arid and arid regions, such as the Urmia Lake Basin of Iran, given their inherent scarcity of water resources and frequent reliance on highly inefficient, traditional irrigation methods, which in Iran entail 40–60% evaporative water loss (Ahmadzadeh et al., 2016; Ardakanian, 2005; Hassanzadeh et al., 2012). Generally, 90% of freshwater in Iran is allocated to agriculture (Food and Agriculture Organization (FAO), 2008, Food and Agriculture Organization (FAO), 2016), a metric which allows little scope for greater water consumption but which has been and will remain progressively strained by growing agricultural and urban water demands. In this context, increasingly extreme climatic conditions will aggravate business-as-usual agricultural expansion and water management, demanding informed, conservative approaches to agricultural planning (Hoegh-Guldberg et al., 2018).

Several methods are now widely used to spatially simulate LUC dynamics and thus draw implications for integrated land-and-water management in semi-arid regions (Andaryani et al., 2019b; Hong et al., 2011; Mirzaei et al., 2020). These methods arguably vary more in complexity and design than overall accuracy and thus present a range of semi-specious options for policymakers. Amongst these methods are cellular automata–Markov chain analysis (CA-MCA) (Andaryani et al., 2019b; Keshtkar and Voigt, 2016; Mirzaei et al., 2020; Rimal et al., 2018), Land-Change Modeler (Nath et al., 2018), SLEUTH (Zhou et al., 2019), Dinamica (Sloan et al., 2018), and Geographical Model (GEOMOD) (Pontius Jr. et al., 2011; Sloan and Pelletier, 2012; Pontius Jr., 2018). The CA-MCA method predicts future spatial and temporal changes to land-use/cover based on the cellular automata (CA) model and Markov chain analysis (MCA), respectively (Kityuttachai et al., 2013; Xin et al., 2012). The SLEUTH model, like the GEOMOD, simulates spatial change only, based on trends typically observed over the recent past, and does not integrate the MCA or other such tools to estimate the quantity of change (Clarke et al., 1997). GEOMOD is relatively favored for policy applications, such as forest‑carbon conservation planning (REDD+) (Brown et al., 2007; Sloan and Pelletier, 2012), given its relative computational simplicity and straightforward binary output of ‘change’ or ‘no change’ for target land-use/cover classes. On the other hand, MCA, being well-documented as accurately simulating aggregate land change (Rimal et al., 2018; Varga et al., 2019), may estimate the area of land-change period to period, apart from changes to its spatial distribution. Consequently, the integration of models such as GEOMOD and MCA is arguably an optimal approach to describing spatio-temporal trends in ways that are accessible to non-specialist policymakers.

Here, we explore the utility of such a hybrid spatial projection methodology for the management of future land change. Building on the GEOMOD and MCA models, we test the performance of a hybrid GEOMOD-MCA model for projecting the spatio-temporal distribution of highly water-demanding orchard cover relative to critical water and soil resources as well as other determinants of orchard cover across the water-stressed Urmia Lake Basin between 1985 and 2030. Our findings describe a subtle redistribution of orchard cover relative to critical water and soil resources and suggest modes of model refinement accounting for small agricultural footprints and nonlinear dynamics of water access and availability over the course of agricultural expansion.