Land is the spatial carrier of all sorts of human life. It shapes a community’s socio-political and economic establishments through the interplay of use and value under a given tenure system [1^[1][2][3],2,3], wherein the rights and responsibilities of the individual, groups/communities and the duty of the government are defined within a context of an overall national development framework and regional/urban development aspirations, shaped within ideologically framed national development policies. At any scale, the spatial configuration of land use is a physical manifestation of the distribution of the structure of benefits and costs to a society [4,5]^[4][5]. Enforcement of fair distribution of such benefits and costs among communities and among the groups and individuals within a community is one fundamental reason that land use must be planned. This is especially true where the market fails, as is so often the case, to fairly balance benefits and costs ^[6]. On the other hand, changing the relation of use–value drives land use land cover change (LULC), where such alteration of ecosystem services (ESs) causes changes to the spatial relation of human activities (human–spatial interactions) [7,8]^[7][8]. This is the second fundamental reason that calls for the effective planning of land use. Land use planning is instrumental in alleviating the potential for incompatible changing of regional/rural land into land for urban use that detrimentally affects the productivity of the primary food supply and ecological services and alleviates land scarcity within the built environment through different use policies ^[9]. In general, as a human economic development carrier resource, as a natural endowment and ESs provider ^[10], and as an institutional entity that shapes the socio-political behavioral relations of humans by tenure conditions, both rural and urban land require effective planning for their sustained productive use. In other words, the degree to which human actions have an effect on changes in the environment has remained a key subject of study.

To this end, societies in successive generations have utilized land use planning, shaped in the context of corresponding dominant ideologies/thinking regarding development. Since the pioneering land use planning model proposed by von Thunen ^[11], the field of land use planning, especially in urban areas, has evolved through generations of models, including structured mathematical models in the 1950s and 1960s, where bid-rent and optimal firm location theories played a significant role in conceptualizing the spatial allocation of activities ^[12]. Spatial simulation models dominated the 1970s and 1980s, and since the 1990s, the focus has shifted towards sustainable development in the land use planning literature, where the hegemony of market forces culminated even in countries which are fundamental market advocators, such as the UK ^[9]. From a technical perspective, sustainable development-framed planning literature now integrates demand–supply quantity structure and spatial simulation, aiming to achieve a balance between economic, societal, and ecological outcomes [13,14]^[13][14].

Within sustainable development thinking, optimizing land use allocation based on the trade-off between different objectives goes beyond economic bidding. It considers the relative productivity of various land use structures for various capabilities (such as access to transport, recreation, shopping, etc.), or alternative function values (ecological reserve, built-up areas, mobility) [13,14]^[13][14]. With the shift in theoretical ground, the land use modeling task has become more challenging, with an increasing number of objectives, specific policy restrictions/requirements, and the need for stakeholder engagement at different stages of the planning process. As a result, traditional structured mathematical models have become less attractive, and self-learning machine learning algorithms (MLA) and dynamic metaheuristic algorithms (MHA) have gained prominence in the land use planning literature. These not only abruptly minimize the costs of processing data, but technological advancements in spatial data acquisition have further increased their capacity to determine patterns of land use and their prediction capacity regarding future use of both urban and regional/rural land.

Within the paradigm of sustainable development, the concept of ESs effectively bridges the gap between science and policy in regional/rural land use planning [8,15]^[8][15]. In urban land use planning, the sustainable built environment discourse, characterized by compact built-up areas, harmonious functionality, mixed-use development, and the relationship between physical structures and the natural ecosystem, serves as the major mainstream form [16,17]^[16][17]. Conceptualizing land use optimization as a contemporary approach to the classical highest-and-best-use value allocation holds promise for the achievement of sustainability ^[18]. Understanding the drivers of land use change and analyzing spatial measurements and the laws of spatiotemporal changes in land use are essential for modeling existing phenomena and simulating future spatial patterns of land use ^[6].

MHA and ML technologies have not only overhauled the technical capability of examining land use plan scenarios. They also have contributed to the shift in spatial development thinking by enabling the capacity to handle temporal perspectives and uncertainties. However, their effect over the fundamental spatial development theories and conceptual models looks unattended, especially in the built environment. The contemporary literature on land use planning tends hegemonically drifting away from spatial domain towards the domain of the technical capability of artificial intelligence in land use planning. This observation can easily be justified by the fact that land use optimization objectives are often formulated as being general management-oriented, especially in urban land use planning research. Only a limited number of works address the basic land use allocation conceptual modes spanning economic geography, utility theories, and spatial morphology. Nonetheless, the whereabouts of such theory-grounded conceptual models has not been explored in the optimization-based land use planning literature, including in review studies.

The better way to address a knowledge gap is conducting a literature study. As in any other disciplines, literature studies play a vital role in land use planning to trace the development trends of concepts/contexts and methods, develop hypotheses regarding new directions, identify current hotspots, and suggest potential research directions. However, review studies in optimization-based land use planning are relatively limited in number compared to the continuously accumulating volume of studies ^[18]. During data retrieval, for example, a broader title “land use optimization*” search parameter in the Web of Science core collection yielded only six articles (1.13% of 530) written in English. In addition, since any review work addresses only a few, if not a single issue, such as a single or certain group of optimization methods, a few objectives of land use optimization, etc., more reviews on land use planning are required. More importantly, reviews that assess the harmonization/synergy of theories/conceptual models/ are worth more given much of the available literature is heavily focused on the development of methods and their applications.

Only six among the retrieved 339 articles were literature studies, of which four were related to the use of metaheuristic methods in agricultural land optimization. Refs. [7,40,42,43]^{[7][19][20][21]} discussed the subject from different perspectives. Refs. [18,43]^[18][21] are about urban land use optimization. The authors of Ref. [42]^[20] are concerned that agricultural intensification is creating the challenges of pollution and biodiversity loss, which thus limits its capacity to meet food demand. According to the authors, land use optimization is the upcoming frontier solution to the issue of food demand. They emphasize the need for high-credibility optimization methods to achieve sustainable land use plans that balance the needs of the ecosystem and the economy. Accordingly, their study numerates the metaheuristic methods that have been applied in agricultural land use planning case studies, explores factors that determine the success of the methods, and explores the alternative mechanisms for the involvement of stakeholders in the planning process. They reviewed 50 articles (38 were case studies), following the PRISMA method [43]^[21]. Reportedly, simulated annealing (SA), taboo search (TS), evolutionary algorithms (EA), differential evolution (DEq), and swarm intelligence (SI) have been in use. Researchers often select methods based on previous success stories of similar problems. Posteriori stakeholders’ involvement, where trade-off solutions are filtered based on multicriteria analysis, has long been common practice. Yet, a great majority of other studies suggest that the stage at which stakeholders participate in the planning process is subject to the nature of the problem; ref. [43]^[21] discusses more on conditions of stakeholder participation. Countering a given land use planning problem, the key contribution of the paper is that it provides a general guideline for the selection of an optimization method. The suggestions of the paper are based on the characteristics of a given problem and the method applied in the studies reviewed. In cases wherein a greater number of constraints may limit the search space navigation, SA and TS tend to perform well. SA is preferred for its parallelization capacity, while TS is an effective local search for improved solutions in the immediate neighborhoods of the current solution by penalizing for revisiting already visited neighborhoods. On the other hand, evolutionary algorithms are recommended for handling multiple conflicting objectives. One cause of the scalability challenges of heuristics is that algorithms are problem domain-dependent; if not, algorithm design should be tailored down to the level of the problem’s specific context. Hybridizing two or more methods is a widely applied solution to deal with the scalability challenge arising from the problem-specific nature of metaheuristics [40]^[19]. The authors of Ref. [42]^[20] demonstrated that combining global optimizers and local search methods could effectively solve multidimensional combinatorial land use optimization problems. The hybridization complements the relative exploration strength of one and the relative exploitation strength of the other. Despite coupling being well-known alternative, parallelization and the use of heuristic-type operators are other alternatives available to deal with challenge of scalability. Ref. [43]^[21] extended the discussion of method selection by adding decision variable type as a criterion. They argue that exact methods are suited for discrete-type variables, and heuristics are compatible with continuous variables. Mixed-integer programming (MIP) is better suited for handling mixed variables. The approach of Ref. [43]^[21] to the optimization method selection is more scientific. Instead of relying solely on tabulating a matrix of problems and methods applied to solve them and then establishing a generic judgment regarding which method(s) could better solve a given nature of a problem, they approached the problem by examining the strengths and drawbacks of various methods (TS, SA, SI, genetic algorithm/GA, artificial immune system/AIS, linear programming/LP, and fuzzy programming/FP); objective formulation type (Pareto-based versus the variants-of-aggregation method; weighted sum/WS versus goal programming/GP); the constraint-handling mechanism (penalty function versus defining feasibility criteria); and the conditions of stakeholder involvement (priori, interactive, posteriori). Consequently, they suggest a well-structured hierarchical evaluation process to select an appropriate method. The first step is identifying the nature of the problem. If the problem does not require extensive trade-off analysis, priori methods (WS, GP/RP, TS, and SA) are preferred, and iterative stakeholder involvement is convenient. If the problem requires extensive trade-off analysis, the decision as to whether objectives have to be scalarized or the problem has to be converted to ε-constraints form is made. If possible, Pareto-based methods are recommended, with stakeholder involvement typically being posteriori. The scope of ref. ^[7] is broader, addressing both the method selection issue and issues of spatial measurements and drivers of land use change. From a spatial development theoretical standpoint, the authors highlight the importance of considering the mutual feedback between humans’ socioeconomic activities and the eco-environment in the context of land use change. This mutual feedback conceptualization recognizes that socioeconomic activities could cause land use change, which in turn affect the eco-environment. The reverse is also true. Changes in the eco-environment could influence the spatial patterns of humans’ socioeconomic activities on land. This mutual feedback conceptualization helps to identify and characterize factors contributing to spatial land use change. Increased attention paid to the effect of ecological limitations on human activity on land means that the driving factors of the analytic paradigm of land use spatial change are of equal weight in terms of economic growth and restraint. At the other end of the spectrum, mutual feedback also involves optimizing land use for a given activity, considering land use change as a factor. In terms of methods, the authors argue that the CLUE-S and Markov models are effective tools for integrating socioeconomic and natural driving factors. These coupled methods are capable of handling conflicting land use relationships. Planning models such as LP, system dynamics (SyD), and multi-objective programming (MOP) facilitate analysis by quantifying economic and social driving factors and determining the equilibrium between demand and supply. Since different problems require different spatial resolutions and different spatial resolutions return different outputs, the authors suggest appropriate level of spatial abstraction and use multiple data types/sources to redress possible gaps while analyzing the land use change. Incomplete individual datasets may further trigger the use of multiple data sources/types. The study conducted by ^[7] focuses specifically on the application of GA in multi-objective land use optimization. Using CiteSpace 5.8.R3 software, the authors map the knowledge collaboration among countries and institutions based on 1154 articles retrieved from multiple databases. They also traced the temporal development of GA knowledge by mapping the frequency of the first-appearing keywords. According to their findings, the period from 1995 to 2004 marked the mainstreaming of GA methods for land use optimization. From 2004 to 2008, much research focused on optimizing the GA itself. Hybrid application of GA dominated the years 2009 to 2016. Since 2017, the literature has progressed towards deepening the integration of GA with big data. Consequently, the authors suggest three future research directions including further integration of GA with emerging AI capabilities, incorporating a temporal dimension into land use optimization to cope with the dynamic nature of land use change, and integrating GA with broader knowledge and practices in the field of land use planning. Both Ref. ^[18] and Ref. [43]^[21] assessed the literature on urban land use planning. The focus of ref. ^[18] is the representation and formulation of objectives of the sustainable built environment, including contiguity, compactness, and compatibility. The authors examined 55 articles following the PRISMA protocol. Reportedly, the compactness objective appeared in 16.67% of the studies, the contiguity objective in 13.67%, and the suitability objective in 11.9%. Aggregating the objectives according to the three pillars of sustainable development, economic objectives appeared in 46.67% of the studies and ecological/environmental objectives in 43.33%, signifying a comparable emphasis on economic development and ESs in urban literature. The representation of social objectives was only 10%. The majority (42.86%) of studies applied Pareto-front-based objective construction; 36.73% applied weighted sum, and the remaining 20.41% followed goal programming. With regard to methodological approaches, 80% of the studies surveyed applied GA in its various customized forms, followed by PSO (12.73%) and ant colony optimization (ACO) at 9.1%. About 80% of the cases depended on raster data models. Given the generic nature of the bibliometric method, the paper, however, falls short of evaluating concepts/theories that influence the identification and formulation of objectives and the performance of the various optimization methods encountered in the articles studied. The review by ref. [44]^[22] is more specific. It examines the establishment of compactness and contiguity objectives that characterize the structure/configuration of a sustainable built environment. Supposedly, compactness is achieved through consolidation of the same uses to form large clusters [27]^[23] enforced by various mathematical formulations including perimeter-to-area ratio, diagonal length of the minimum bounding rectangle, the weighted average ratio of area to the square of a perimeter, core–buffer cell assignments, minimizing the number of clusters, or maximizing the largest cluster. Alternatively, constraints such as setting a minimum threshold of a certain land use cell in a neighborhood may be imposed as a precondition to allow additional allocation of a specific use type to a neighborhood. The modeling of contiguity, on the other hand, is often based on graph theory operationalized using a path-based model, order-based model, or network-based model alternatives. However, whether promoting larger clusters of the same uses is in line with the conventional wisdom of sustainable built environments is questionable, and is considered in next subsection. Collectively, the review articles provide a wide-coverage summary of the land use planning literature both in the context of methods and in planning knowledge. Yet, there exist critical gaps left unaddressed. The first limitation is that the literature heavily dwells on methods, rather than on planning theories and conceptual discussions. Among the six review articles, only ref. ^[7] gave meaningful weight to planning concepts, which underscores that the effect of human socioeconomic activity on the ES is not the only aspect of the spatial planning domain. The mutual-effect feedback concept seems very important in deepening understanding of sustainable development. Yet, whether the spatial change in activities caused by ESs change ends with sustainable solutions seems an emerging new direction for discussion and inquiry. Despite being heavily method-focused, the reviews also left many methodological concerns unaddressed. First, they lack standardized frameworks to assess and compare the performance of the methods they encountered while reviewing studies. Second, they did not notice that different performance indicators may rank the same set of methods differently. It is also doubtful whether computational time cost will prevail over planning quality. There is no theoretical ground to benchmark the trade-off between computational time and solution quality. Third, there is a tendency to assume tools/methods that evaluate data efficiently according to defined instructions are end-solutions in and of themselves (see ref. [42]^[20]). The six reviews discussed above explored various land use planning concepts and methods. These studies systematically analyzed the strengths and weaknesses of different global and local land use optimization methods spanning evolutionary algorithms (EAs), SA, LP, and integer programming (IP). By consolidating and summarizing the findings, these studies provide valuable insights and guidelines for selecting an appropriate method. Regarding method selection, two distinct approaches have been identified. The first approach is establishing an association of the problem domain and the applied method type based on previously published studies. This approach is, however, overly simplistic, and the success rate of obtaining appropriate methods is lower. The second approach involves a scientific analysis of the nature the problem domain and the construct behavior of the optimization methods. Additionally, decision making time constraints and solution quality criteria, such as diversity and reliability, are important considerations for the method selection [45,46,47,48]^{[24][25][26][27]}. Another factor that influences method selection is problem size. Exact methods are less suitable for larger problem sizes, e.g., exceeding 200 [24]^[28]. This is primarily due to the complex interrelationships between geographic and ecology, as well as the characteristics of the geographic units (e.g., shape, distance, adjacency, etc.). The type of variable involved is also a determining factor in method selection ^[18]. However, it is important to note that an objective and comprehensive analysis considering all these parameters is still lacking for method selection. Since the contextualization of a certain land use planning problem may belong to a certain conceptual/theoretical/framework and the mathematical formulation of a problem is often distinct, the method selection strategy suggested by ref. [42,43]^[20][21] still remains one including generic guidelines. This fact is supported by the fact that different studies reporting different findings and presenting different interpretations offer the same method(s) applied to similar problems. A study by ref. [49]^[29] assessed the performance of a nondominated sorting genetic algorithm II (NSGA-II), PSO, and a multi-objective evolutionary algorithm (EA/D) based on solution dispersion, the diversity of the solution space, and the number of dominant solutions in Pareto-front parameters. Reportedly, PSO exhibited the best diversity of solutions, while the EA/D outperformed the other two in terms of computational time. In the study by ref. [50]^[30] that compared LP, SA, mult-objective land use allocation (MOLA), and multidimensional choice for land use planning (MDCHOICE), reportedly, MOLA performed better in landscape metrics, LP returned the best demand requirement convergence level, and MDCHOICE had the best ability to harmonize land uses (i.e., diversity of the solution space). It is also very interesting to encounter a situation wherein a certain method may be biased to a problem type. The same study reported that SA was the weakest for the metrics mentioned above in all the test problems, but excelled in solving the forestry land use allocation problem. Ref. [51]^[31] assessed the performance of GA, Cuckoo Search, and PSO in agricultural land allocation problems, aiming to diversify crop plantation plans. The authors reported GA’s superior capabilities, as evidenced by 103% simulated crop yield growth and 97% simulated profit growth, while simulated water consumption still reduced by 5%. In a study that compared Markov–cellular automata and Markov–cellular automata–CLUE-S hybrid models, both were applied to a land use problem with the aim of maximizing ESs values; in [52]^[32], the authors reported that the non–stationary Markov state transition and cellular automata (CA) probability hybrid returned a high level of layout precision. Despite the fact that the review articles are devoted to optimization methods, they also contribute to the advancement of concepts in spatial development planning. Ref. [52]^[32] suggests that the depth and breadth of sustainable urban land use optimization are barely explored. One of the limitations highlighted is the low (only 10%) representation rate of social development-related objectives. It is worth noting that studies that encompass objectives that represent the three dimensions of sustainable development are rare; only two (of total 55) studies included social development dimension-related objectives (in a review by ^[18], for instance). In studies that consider the social dimension, the representation is often vague, such as a form of social security service values or accessibility to main roads [36]^[33] and spatial compactness [53]^[34], wherein the representation creates confusion with geographic compactness objectives that address the ecological environment. Another significant limitation worth mentioning is the lack of standardization or consensus concerning the use of proxy variables to calculate the benefits of the objective parameters of each sustainable development pillar. These gaps provide opportunities for further research, especially that which establishes consensual objectives that represent social development and proxy measurements to gauge the performance of each pillar of sustainable development. This can be made more concrete by developing protocols that standardize the measurement of ES values and social development.