IntroductionMulticriteria assessment methods, simulation tools, and decision support tools for landscape assessment and design typically rely on two-dimensional Geographic Information System or satellite data and focus on the urban or regional scale (Pechanec et al. 2015; Koschke et al. 2012; Soares-Filho et al. 2002). The range of criteria they can assess is constrained by their two-dimensional plan view interfaces and data sets. On the contrary, assessments using three-dimensional geometric methods are limited to single-criteria assessments including privacy (Lonergan and Hedley 2016), shading/lighting (Zhang et al. 2015), ventilation and microclimate (Hong et al. 2011), views (Ervin and Steinitz 2003), financial cost or economic performance (Luttik 2000; Nowak and Dwyer 2007; Wang et al. 2016), water uptake (Javaux et al. 2008), and carbon sequestration (Le Roux et al. 2001). The use of photographs and renderings to assess household landscape preferences is also well-established and sometimes includes water use or cost heuristics (Kuper 2017; Hayden et al. 2015; Nassauer et al. 2009; Zheng et al. 2011).This modeling approach contributes to urban landscape and water planning research and practice in several ways. At present, no publicly available tool or published model known to the authors deploy a spatially explicit three-dimensional model to perform multicriteria assessment of landscape performance. This greatly improves the accuracy of simulations and allows the assessment and comparison of both existing conditions and new design options. The method is also novel in its use of multiresolution analysis to integrate (low-res) full-factorial analysis and pareto-front generation into (high-res) manual design to improve overall scores. Generating a full-factorial analysis at the higher resolution would increase the compute time by 200 billion times (or 400,000 years).Adaption of the proposed methods by municipal planning agencies would complement existing landscape design guidelines. It would help quantify the overall water and environmental performance of suburban and urban landscapes. Complementing the traditional heuristics of simplified planning guidelines with spatially and species-based explicit simulations would quantify and optimize the positive benefits of landscape plantings. This analytical approach helps to conserve limited water resources at multiple scales (O’Reilly 1996; Costello and Jones 2014; ADUPC 2017). It thereby helps to integrate spatial, environmental, and water resources dimensions of urban landscape planning. We test this model in the United Arab Emirates (UAE), a high water-constrained society with underperforming residential landscapes.Arabian Gulf Housing and LandscapesPremodern urban residences on the Arabian peninsula evolved to balance several intersecting and sometimes competing, needs for privacy, water conservation, passive cooling, shading, and views (AKTC 1996). These arid residential landscapes were refined over many generations to maximize the net benefits of a limited number of trees. During the late 1960s, planners in the UAE developed new housing typologies, such as the Sha’biyat, that reinterpreted traditional Arabic courtyard houses to provide many of the same spatial, cultural, and microclimatic benefits of premodern Arab houses (Ménoret 2014) (Fig. 1). Native drought and heat-resistant trees were planted to provide similar multifunctional benefits for courtyards, for privacy, and for esthetic beauty. During the rapid modernization of the United Arab Emirates, however, these neo-traditional houses gave way to large single-family villa houses starting in the 1970s (Elsheshtawy 2008). Gardens were initially irrigated from brackish aquifers and subsequently with desalinated or treated wastewater (Makhzoumi 2002).Aquifer depletion and increasing desalination costs and environmental impacts led to stricter regulation of outdoor domestic water use and wastewater reuse during the early 2000s. Public information campaigns, such as Waterwise, aimed to lower indoor and outdoor water use through voluntary household decision making (RSB 2014). These efforts resulted in reduced tree and grass plantings around villas that reduce water use but often at the cost of decreased privacy, shading, and landscape views (Fig. 1, Contemporary). Recent surveys in the region by Hakky (2012) show that reduced vegetation leads to less use of villa open spaces due to a lack of privacy and shading and less pleasing views. Residents identified the financial costs of landscape planting, maintenance, and irrigation as secondary concerns.In this paper, we use multiple criteria as views, indoor privacy, outdoor privacy, shading, landscape cost, and water use to develop an urban planting decision framework and tool (Table 1). These criteria were selected using Hakky’s surveys and the author’s expertise in Arab garden design. The methods and findings in this paper should find broad application in the Gulf region and beyond (Margolis and Chaouni 2014).Table 1. Description of criteria used in evaluating and ranking design optionsTable 1. Description of criteria used in evaluating and ranking design optionsNo.CriteriaDescriptionEquationMax. valueMin. value1ViewQuantifies the extent to which landscape features can be viewed from inside a houseView score = view hits/number of rays tested1.00.02PrivacyAssesses inside and outside privacy and computes obstruction of view between window and street vantage points.Privacy score = sum (occluded views × density K)
/(number of rays tested × max density K)1.00.03ShadingFraction of total open-space area covered by shade of treesShading score = sum (shaded areas × shade K)
/(open space area × max shade K)1.00.04Water requirement and costNormalized landscape water volume required (this serves as a score for volume of water and its cost)((Water requirement = thin tree area × thin Kc × ETo) + (dense tree area × dense Kc × ETo))/irrigation efficiencyMax water demand = (landscaping area × max Kc × ETo)/irrigation efficiencyWater requirement and cost score = 1 − (water demand/max water demand)1.00.05Planting and maintenance costNormalized cost of installing and annually maintaining the plantings in landscapePlanting and maintenance score = 1 − ((thin tree area × thin tree cost K) + (dense tree area × dense tree cost K))/(landscape area × max cost K)1.00.0State of Water Resources in the UAEIn part due to the efforts mentioned previously, per capita, water consumption in the UAE dropped roughly 20% from 2010 to 2017 (ADWEA 2017). As of 2013, 70% of outdoor water use was in the agriculture, forestry, and public park sectors (2% desalinated, 6% recycled, and 91% groundwater, respectively). Domestic use was the next largest sector at 17% (100% desalinated). Government, commercial, and industrial water use accounted for the remaining 13% (100% desalinated) (EAD 2015). Since 2010, desalinated water is increasingly used for agriculture and potable use, whereas recycled water is increasingly used for outdoor landscaping (ADWEA 2017).The UAE Water Security Strategy for 2036 aims to “ensure sustainable access to water during both normal and emergency conditions” (UAE 2019). Measurable targets include 21% reduction in total demand, 69% increase in water productivity, and 95% increase in treated wastewater reuse. Over the coming decades, Abu Dhabi will replace existing gas-powered, multistage flash distillation plants with more energy-efficient, coupled renewable reverse-osmosis plants for desalination (Dawoud 2012). In 2019, Emirati households paid 2.09 AED/m3 for water up to 7 m3/day and 2.6 AED/m3 for water use above that; prices for non-Emirati residents started at 7.84 AED/m3 (ADDC 2018).New Landscape Technologies and OpportunitiesRecent transitions to water recycling, coupled with renewable energy sources, provide an opportunity to focus on the benefits and costs of water use for residential landscapes in the Arabian Peninsula (Amy et al. 2017; McDonnell 2014; Alobaidli et al. 2017). The potential to support a strategically limited number of trees and shrubs in villa open spaces can help to renew the historic social, cultural, and microclimatic benefits of Arab residential gardens in the region (Gharipour 2016). Updating urban landscape design in the region can also support local professional expertise and training (Mitchell 2016). The contemporary literature on urban residential trees benefits include improved home value, enhanced esthetics, improved privacy, microclimate cooling, air pollution reduction, noise reduction, increased walkability and open-space use, decreased soil erosion, and energy use reduction through shading (Nowak and Dwyer 2007; Roy et al. 2012; Tyrväinen et al. 2005). The potential for more sustainable water use for residential landscapes in the region has also been assessed (Birge et al. 2019).Current residential landscape design software tools (e.g., Lands for Rhinoceros) focus on esthetic, climatic, and/or irrigation factors in design. To the best of our knowledge, these tools do not incorporate shading, privacy, or views. The online i-Tree tool by the USFS (2019) assesses the environmental benefits of trees (e.g., shade, pollution mitigation, stormwater run-off reduction, carbon sequestration, and storage) but does not assess spatial or esthetic factors such as privacy and views.New Landscape ModelThis paper presents a new landscape design assessment tool (LDAT) that integrates multicriteria decision analysis (MCDA) with multiresolution spatial analysis (MRSA) for comparing villa landscape design options. The analytical framework in LDAT is presented and then applied to a prototype case study. LDAT is aimed at urban planning departments, water conservation programs, landscape designers, and their clients. It is based on industry-standard software to enable user-groups to access the same models and data, and answer questions such as follows:
•What landscape configuration provides the best performance for individual design criteria?•What landscape configuration provides the best performance across all criteria (here these include shade, privacy, view, cost, and water use)?•How sensitive are scores to water pricing policies? Do the same configurations consistently perform better?•What unintended consequences might develop from water policies?The answers to these questions can bring advanced methods to residential landscape design in arid regions.MethodsMCDA evaluates alternatives based on multiple decision factors (Blanchard and Fabrycky 2006). Different aggregation and weighting methods can be used to assess alternatives. In LDAT, the model begins with a 3 × 3 matrix of landscape spaces, with the villa in the middle, and the modeling involves three steps. Step 1 conducts a full-factorial enumeration of the landscape design options. Each design option has a unique combination of three numbers, locations, and types. A multicriteria analysis is done for each design, and the score for each design is obtained. Based on this analysis, a small set of good design options are selected. Step 2 increases the spatial resolution of the multicriteria assessment by four to eight times, generally leading to improved scores. The highest scoring results of Step 2 are used in Step 3 to create high-poly vegetation models for client-oriented visualization. We used McNeel and Associates’ Rhinoceros 3D (Rhinoceros 3D 6.35) and Grasshopper (Grasshopper 1.0) software packages to implement the method described in this paper, although other software packages that provide scripting interfaces could be used. LDAT’s method leverages weighted, multicriteria, geometrically explicit computation to identify the best-performing residential landscape designs.Criteria DescriptionThe following five different criteria were used in this analysis: view, privacy, shading, water demand, planting, and maintenance cost.The View Criterion quantifies the extent to which landscape features can be viewed from within a villa. Views from inside the house are measured by extending rays from the centroid of each window at y degrees incidence from the window normal (i.e., straight ahead). The first object the ray strikes is used to determine the view of that ray. If the view is of a tree or other landscape features or is of the sky (an upward ray with no hit), the views parameter is incremented by 1 unit. If the view is of empty ground (nonlandscaped), a wall, a neighboring house, or does not hit any object (for nonupward rays), the score is not incremented. Rays can be given a cut-off distance beyond which the default view is of nothing. Our study uses five rays from each window—four rays at 45° from the surface normal (i.e., projecting up, down, left, and right) and a fifth ray projected straight ahead at 0°, but any number can be used depending on the computational requirements.(1) Viewscore=viewhits/numberofraystestedThe Privacy Criterion is assessed both from inside the house and from the Villa’s open spaces. Inside privacy is calculated by testing for object obstruction of view lines generated between all neighboring window vantage points and from equally spaced vantage points along the street at a 1.75 m height. Open-space privacy is tested from the centroid of each landscape cell to all window and street vantage points. If a view ray intersects a tree, the occluded views parameter is incremented by 1 unit times the tree density coefficient (dense = 1.0; thin = 0.5) to account for varying levels of privacy provided by different tree types (i.e., the density of leaves). If the ray intersects a solid wall (building and perimeter wall), the occluded views parameter is incremented by 1 unit. Given the differences between inside privacy and open-space privacy, each type of privacy is scored and weighted as a separate criterion in the baseline model. Therefore, the(2) Privacyscore=sum(occludedviews×densityK)/(numberofraystested×maxdensityK)The Shading criterion estimates the percentage of total open-space area (minus driveways) shaded by trees. The shade in each landscape cell is projected vertically onto the ground plane and the area calculated. The tree shaded area for each cell is then multiplied by a tree shade coefficient K (dense = 1.0 or thin = 0.5). In low latitude countries like the UAE, vertical projection provides an adequate estimate of yearly shading potential. For higher latitudes, an angled projection could be used. Therefore, the(3) Shadescore=sum(shadedareas×shadeK)/(openspacearea×maxshadeK)Water requirement and cost criteria are estimated using the State of California’s WUCOLS method whereby reference evapotranspiration (ETo) is multiplied by a crop coefficient (Kc) for each planted area in the landscape (Costello and Jones 2014). This study used a Kc of 0.5 for thin trees and 0.75 for dense trees to account for the higher ETo rate of trees with dense canopies, although these values can be adjusted for different species. The base water requirement is divided by irrigation efficiency, which ranges from 0.5 for crude systems to 0.9 for drip systems, to give the net water requirement. Estimating landscape water-use volumes in arid regions is critical for municipal water planning. The estimated water requirement for each design is normalized to a score from 0.0 to 1.0, with 1.0 being the best possible score (no water use at all), and 0.0 being the worst one (100% coverage of a tree with the highest water coefficient possible).Water requirement estimates are also used to compute the monetary cost of water. In the simplest case of a uniform water pricing scheme, the monetary cost of water use would normalize to the same score as the water requirement. In that context, the water requirement score is a proxy for both water use and irrigation cost.(4) Waterdemand=((thintreearea×thinKc×ETo)+(densetreearea×denseKc×ETo))/irrigationefficiency(5) Maxwaterdemand=(landscapingarea×maxKc×ETo)/irrigationefficiency(6) Waterscore=1−(waterdemand/maxwaterdemand)Water pricing schemes are tested by adjusting the water cost coefficient functions that are used to score water demand. Zero water use in all cases produces a score of 1. Different cost functions change the rate at which an additional unit of water use lowers the score. This method allows the comparison of normalized costs for different pricing schemes without requiring knowledge of actual water pricing in a given region.Planting and maintenance cost criteria are estimated by multiplying the projected area of the tree canopy by a cost coefficient and normalizing those costs by the maximum cost. For this case study, we use a score of 0.75 for thin trees and 1.0 for dense trees to approximate the slightly higher cost of planting and trimming trees with dense canopies and more branches (McPherson et al. 2006). Accordingly,(7) Costscore=1−((thintreearea×thintreecostK)+(densetreearea×densetreecostK))/(landscapearea×maxcostK)Total ScoringThe aforementioned criteria assess landscape performance across a range of social, economic, and environmental dimensions. These can be aggregated in different ways. In this study, the values of each criterion are computed for each design, and the results are normalized and weighted using a general additive weighting technique.The details of the additive weighting approach, as used in LDAT, are as follows (Siddiqi et al. 2016). For n designs (generated through full-factorial enumeration), a set of m criteria are employed by decision makers for evaluation. For an m × n performance matrix, A is then defined as follows:(8) where aik = value (or performance) of kth design for ith criterion.The performance for each criterion is normalized, and for an m × n matrix, A¯ is obtained. For criteria to be minimized (such as cost), aik¯ is given as follows:(9) a¯ik=[max(ai)−aik][max(ai)−min(ai)]where max(ai) = the maximum value for the ith criterion; and min (ai) = the minimum value for the ith criterion across all k (i.e., set of designs). For a criterion to be maximized (such as shading), aik¯ is given as follows:(10) a¯ik=[aik−min(ai)][max(ai)−min(ai)]A weighted-normalized vector J is now defined as follows:(11) where wi = weighting factor of the ith criterion and where:(12) Note that J is a row vector and consists of weighted, normalized performance values of the n designs.We next formulate combinations of w for which each wi are systematically varied from 0 to 1 with step dw. Next, we select the sets of w where Eq. (12) holds. From within these sets of w (wherein wi sum up to one), we determine the number of cases when the ith project is determined to be i*,where(13) that is, the design with the highest normalized-weighted performance.Each individual criterion score is a percent of the perfect possible score. For example, a score of 0.95 for shading indicates that the landscape option achieves 95% of the best-possible performance for shading. There are tradeoffs between criteria. Higher shading scores require more landscaping, which increases watering, planting, and maintenance cost scores.The raw total score for a landscape option is simply the addition of each individual criterion score. To normalize raw scores to a maximum score of 1 and allow for differential weighting (preference) of individual criteria, each raw score is multiplied by a coefficient. For neutral weighting with five criteria, each coefficient equals 0.2. Coefficients can be adjusted as desired, as long as they sum to 1. For the purpose of this study, all weights were left equal. When applied by municipalities or landscape designers for specific regions or clients, the weighting of more important criteria would be appropriate.Spatial Design ModelingAs noted previously, the spatial modeling involves three steps, which begin with a 3 × 3 square site plan [Fig. 2(a)], and proceed to subdivide those squares into smaller planting cells [Fig. 2(b), and then into planting layout plans (Fig. 2(c)].
•Step 1—Nine-Square Full-factorial Model: Step 1 is based on the low-resolution division of a villa’s open space into nine landscape cells. The nine-square division results in one cell are occupied by the house (with flexibility in the location and dimensions of the house), and the eight remaining cells are available for landscaping [Fig. 2(a)]. Each landscape design alternative is evaluated against the performance criteria described previously. A full-factorial enumeration of design options is generated so that all permutations of tree planting numbers, locations, and types can be evaluated—from no cells to all cells filled and every option in between. The number of tree types determines the number of permutations that result: 1 species = 256 options; 2 species = 6,561 options; 3 species = 65,536 options; and 4 species = 390,625. This study uses two tree types (dense and thin canopy) to maintain tractability for our computer setup (Intel i7 4.0 GHz, 32 GB RAM). Full-factorial models are beneficial because they provide a full mapping of all design options and can offer insights into the sensitivity to different input values. This is helpful where designers or policymakers want to understand what leads to the best options, and what leads to inferior designs.•Step 2 aims to improve the scores of promising options identified in Step 1 by increasing the spatial resolution of planting design alternatives. Step 2 first subdivides landscape cells into smaller cells using min–max size criteria to ensure that the resulting subdivisions are roughly equal to one another and approximate a single tree size. Typically, this results in a division of each cell in Step 1 into four cells in Step 2. Step 2 also tests the immediately adjacent cells to those selected in Step 1 [Fig. 2(b)] to ensure that high-performing locations for landscaping are not missed due to the gross aggregation of landscape cells in Step 1. Step 2 removes low-performing landscape areas and identifies high-performing areas at a finer spatial resolution.•The end result of Step 2 is a voxel planting template. Step 3 converts this planting template into a detailed, high-polygon model for visualization of high-performance design options. Rhinoceros’s visualization features generate axonometric and perspective views and renderings [Fig. 2(c)].Modeling ResultsThe results from each step are summarized and discussed subsequently.Analytical ResultsAs noted previously, the full-factorial analysis involves 6,560 design options. Table 2 presents individual criteria scores and total scores for a selection of those design options, including the top 5 ranked options along with medium- and lower-ranked options down to number 6,000. Green boxes indicate criteria with the highest scores, while red boxes indicate the lowest scores. The top-scoring options plant 3–4 cells out of the 8 that are available. In the middle-ranked options, the benefits of shading and privacy are weighed against water and landscape costs. The lowest ranking options sacrifice landscape benefits to save costs.Table 2. Ranked comparison of individual criteria and total scoresTable 2. Ranked comparison of individual criteria and total scoresRankIdxCell numberSpecies balanceTotal scoreViewShadingIndoor privacyGarden privacyWater useLandscape costWater use (m3/year)15543Dense0.580.420.350.480.380.650.6566721,6754Dense0.570.560.540.480.380.460.461,02631,6694Dense0.570.440.460.480.410.530.5388241,6784Dense0.570.440.460.480.410.530.5388251,6394Dense0.570.420.460.480.410.530.53882501262Dense0.540.290.230.350.280.770.774361001,6794Dense0.530.360.450.540.540.350.418672503,4235Dense0.510.480.600.370.340.390.371,1545003,2535Dense0.500.420.590.350.340.400.381,1341,0004,1126Dense0.480.310.580.350.410.400.361,1402,0003,1695Mixed0.460.370.520.280.270.450.411,0353,0003,6516Thin0.450.280.460.300.270.500.429424,00020105Mixed0.430.270.422.214.171.1240.449075,0002,8225Mixed0.420.300.4126.96.36.1990.449476,0001,3344Mixed0.400.190.3188.8.131.520.55753Spatial ResultsFig. 3 displays the spatial implications of these results. Fig. 3 presents the optimal landscape configuration based on all criteria, [Fig. 3(b)] shows the effects of increasing the weights for shade, privacy, and views, while [Fig. 3(c)] indicates that changing the species coefficient may or may not alter the optimal design.Fig. 4 graphs the full-factorial results from Step 1, mapping key relationships between total weighted score and number of filled landscaped cells [Fig. 4(a)], between total weighted score and total estimated annual water [Fig. 4(b)], and between social benefits (views, privacy, and shading) and environmental costs (water use and cost of landscaping) [Fig. 4(c)].Step 2 assesses and refines high-performing results from Step 1 at a higher spatial resolution. Once an option is selected, the designer sets up the three-dimensional Rhinoceros model using the tree configuration for that option (i.e., location and type) [Fig. 5(a)]. The selected cells from Step 1 are then manually subdivided into smaller cells [Fig. 5(b)]. Neighboring cells are turned on as well, but not given a type [Fig. 5(b), gray boxes]. Finally, the designer manually turns cells on and off in search of a better-performing landscape relative to the base typology from Fig. 5(a). The resulting landscape design has a higher score due to the more efficient use of landscaping [Fig. 5(c)]. For example, the privacy benefits of trees in front of the house are maintained with less water required. While this process could theoretically be automated, manual adjustment allows for more nuanced contextual adjustment and design intent.After the landscape design is refined in Step 2, the designer can lock the box geometries in Rhinoceros, make them translucent, and use the boxes as a guide to place high-polygon, three-dimensional trees. This step transitions from landscape performance to client visualization. When multiple options from Step 1 are developed further in Step 2, the result of Step 3 would be a set of renderings depicting each option with associated scores. Designers would use these to guide the final selection process with clients (Fig. 6).DiscussionIn this section, we will discuss how the best-performing options relate with each other; how model results vary under different score weightings and species coefficients; why some landscape designs perform poorly; and why our tool is most valuable in areas with significant water and planting costs.Best-Performing OptionsFig. 4(a) plots the number of trees against the total score. It shows that landscapes with three cells of dense landscaping achieve the highest score, while landscapes with 4, 5, and 6 planted cells out-perform those with 2, 7, and 8 cells. Interestingly, the worst- and best-performing landscapes both have three cells of landscaping, highlighting the model’s sensitivity to the location and density of planting.Relating scores to water use is useful for designers and planners in regions with water limits or quotas, as is increasingly the case in arid regions (ADUPC 2010). Fig. 4(b) displays the best-performing options at any given water-consumption level. The blue-hatched area shows how water-use boundaries can be demarcated on such figures to evaluate options in which water resources are used effectively.This type of analysis is particularly useful for cities where water sources are rapidly dwindling (McDonnell 2014). In Abu Dhabi, desalination and recycled water are the primary sources of residential irrigation. Household-scale gray-water recycling can also offer a cost-effective and decentralized source for subsurface irrigation for trees and shrubs.High-Performing LandscapesAn analysis of the best-performing options (Fig. 4) shows that there is a set of high-performing landscapes (HPLs). These HPL options stand out as a sparsely populated region of the graphs, compared with the dense clustering of moderately performing options. The full-factorial results show that the HPLs are comprised of minor variations in species, cell number, and location around the optimal landscape option.Invariance of High-Performing OptionsHPL options remain relatively consistent under most water policies and different criteria weighting. For example, in Fig. 3, we see that three dense landscape cells positioned in front and on both sides of the house form the basis of all but one optimal landscape. Cross referencing with Table 2 reveals the reasoning behind the optimal scores: efficient use of water coupled with moderate scores across all social factors. Species coefficients have to be significantly adjusted before the optimal configuration significantly changes (Fig. 5). This invariance has a number of implications and uses. This finding suggests that there is a range of high-performing options, which should appeal to consumers and regulators.
•Even without setting absolute physical limits on water use, the model can predict the subset of landscape designs and plant types that comprise the HPLs.•Even in cases where water availability and financial cost are not limiting factors, city planning departments can ensure better landscape performance and effective use of water by setting lower bounds for landscape scores.•Applying the LDAT model to existing landscapes, cities can assess the current performance and future possible performance of public and private landscapes. Indicators could also be created to track city-wide residential landscaping performance over time.High Performing Versus Low PerformingWhile the majority of design options produced by the full-factorial enumeration process can be discarded, investigating the poor-performing options provides important information. For example, we can clearly see from Table 2 that landscapes perform poorly primarily because they are ineffective at generating social benefits, even though they do not use a lot of water. It is not the absolute use of water, but the inefficient use of water that leads to poor performance. Finally, due to their locational sensitivity, views and privacy scores have the most impact on the overall score. That said, higher-weighting of shading scores can lead to different outcomes for optimal landscapes.The low-resolution of Step 1 limits the ability for individual criteria scores to move independently of one another, which is a significant problem for multicriteria models (Brans and Mareschal 1994). The increased resolution of Step 2 resolves this problem by using small enough landscape cells such that a planted cell can simultaneously create privacy and attractive views.The higher spatial resolution of Step 2 improves overall scores by redistributing landscaping from lower- to higher-performing locations. These steps dramatically reduce the design space explored manually in Step 2. Redistributing smaller cells of landscaping within the schema of an option from Step 1 allows privacy, view, and shading scores to increase through the more precise location of a relatively fixed amount of landscaping while not incurring additional costs. Score improvement can also occur by removing redundant landscaping (e.g., editing landscaping that blocks views already blocked by other vegetation). In the example given, the beginning score of 0.51 is significantly improved to 0.72.Guidelines for Planting DesignStep 2 is also important for its ability to translate general typological guidelines to individual houses and ultimately to larger neighborhoods. For example, entry and window locations will ultimately determine the best location for trees along the front of a house. Gross landscape guidelines would never be able to account for such benefits.Step 3 on design visualization is briefly introduced in this paper. We include it to show what is possible when communicating design alternatives. There is enormous potential for future development along these lines, especially when considering multicriteria tradeoffs. It is important for planners, designers, developers, and clients to be able to imagine the difference that plant selection and location make both for landscape performance and cost. This is especially the case in arid contexts where the most common landscape designs still feature grass lawns and palm trees, which provide few of the potential shading, privacy, and esthetic benefits.While the results from Step 1 are intuitive in many respects—for example, planting trees for shading, views, and privacy in front of windows will perform best, all else being equal—what is far less obvious is how to prioritize the location and types of trees in a limited number of landscape cells. If one can only plant two landscape cells, where should they go and what types of trees would be best? Is it better to place two dense trees in front of half of a house’s windows or four thin trees in front of all windows? Steps 1 and 2 of the chained model help in answering these questions, and can inform landscape designers and their clients which options are best, by how much, and whether they fall within allowable limits of personal financial costs (planting, maintenance, and water use) and/or absolute water-use limits.Urban planning and environment departments can use this tool to understand the landscape implications of absolute water limits, water pricing schemes, and other design regulations, including the potential unintended consequences of policy decisions. Cities can bound the number of conforming landscape designs while still leaving room for personal preference and refinement at the parcel scale. Cities can provide the LDAT model to landscape designers and require that it be used to submit planting proposals for permitting purposes. Agencies can also use the LDAT model to generate three-dimensional design guidelines, augmenting existing design guidelines such as Abu Dhabi’s Public Realm Design Manual (ADUPC 2017).ConclusionThe LDAT simulates landscape performance at multiple resolutions for indoor and outdoor privacy, shading, and views of the landscape. It measures the financial costs of landscape planting, maintenance, and water use. The tool can thus help to integrate decision making across multiple stakeholders including planning departments, designers, and families. It was tested in the UAE but has wider applicability in water-sensitive urbanizing regions. Water scarcity is increasing in all urbanizing regions, especially during drought events. Hydroclimatic variability and climate change will further challenge landscape planting design in ways that can be modeled to enhance the robustness and strategic value of landscape design alternatives. Climate change will likely alter species choices as well as irrigation design. It is important to recognize that transferring water to urban regions to meet new demands and offset uncertainties comes at a cost to other users and environmental systems and must therefore be optimized and not wasted. Prohibitions against waste exist in many jurisdictions but are rarely implemented due to the absence of quantitative tools such as LDAT. This tool can help municipalities to better quantify their needs, benefits, and value in changing climates. In regions with greater rainfall, integrating rooftop rainwater harvesting with landscape planting alternatives could be incorporated in this model as can various graywater and treated wastewater alternatives.LDAT is aimed first and foremost at exploring and evaluating the social and hydrological dimensions of residential landscapes so that families living in arid regions can benefit from optimally designed landscapes. The social, physical, and psychological benefits of residential landscaping are well established. Villa homeowners in the UAE and surrounding regions have directly expressed a desire for increased shading, privacy, and views of landscaping. LDAT combines realistic visualization with weighted scoring so families can make the best design decisions according to their preferences that also meet or exceed the minimum requirements. Future research could address additional landscape benefits such as energy benefits for heating and cooling that could involve adding a rooftop space for green roof alternatives. Green roofs are increasingly adopted, especially in mesic climates, and have significant residential benefits. Another extension of this research could link modeled landscape benefits with various measures of well-being or quality of life.At the larger scale, cities could test various technology, policy, and zoning options to assess costs and benefits across the multiple interacting criteria of water use, shading, privacy, open-space regulations, tree density, and species selection at various levels of spatial aggregation. Previous work using systems analysis methods to integrate urban planning for buildings, transportation, water, energy, and waste management in new regions (Alfaris et al., 2010) connects with the specific capabilities of LDAT. Cities can use LDAT to set the minimum standards of landscape performance that balance the environmental, social, and economic sustainability of landscape designs. Cities can also aid designers by highlighting the set of high-performing landscape options for the most prevalent housing typologies in a region. Designers can use LDAT to design residential landscapes for clients in ways that balance the performance criteria measured by our model.Finally, cities can also use the LDAT model to refine the density, species selection and mix, and performance characteristics of city-wide landscapes over time. Air-pollution removal, microclimate buffering, walkability, and neighborhood-wide home values are all parameters that are impacted by street, neighborhood, and city-scale landscaping. The framework (and tool) presented here provides an approach for systematically exploring and identifying parcel-level landscaping options, which at neighborhood and city scales can combine to create positive urban social and environmental benefits.References AKTC (Aga Khan Trust for Culture). 1996. Sustainable landscape design in arid climates. Geneva: Aga Khan Trust for Culture. Alfaris, A., A. Siddiqi, C. Rizk, O. de Weck, and D. Svetinovic. 2010. “Hierarchical decomposition and multidomain formulation for the design of complex sustainable systems.” ASME. J. Mech. Des. 132 (9): 091003. https://doi.org/10.1115/1.4002239. Alobaidli, A., B. Sanz, K. Behnke, T. Witt, D. Viereck, and M. A. Schwarz. 2017. “Shams 1—Design and operational experiences of the 100MW—540°C CSP plant in Abu Dhabi.” AIP Conf. Proc. 1850 (1): 020001. https://doi.org/10.1063/1.4984325. Amy, G., N. Ghaffour, Z. Li, L. Francis, R. V. Linares, T. Missimer, and S. Lattemann. 2017. “Membrane-based seawater desalination: Present and future prospects.” Desalination 401 (Supplement C): 16–21. https://doi.org/10.1016/j.desal.2016.10.002. Birge, D., S. Mandhan, W. Qiu, and A. M. Berger. 2019. “Potential for sustainable use of trees in hot arid regions: A case study of Emirati neighborhoods in Abu Dhabi.” Landscape Urban Plann. 190: 103577. https://doi.org/10.1016/j.landurbplan.2019.05.008. Blanchard, B. S., and W. J. Fabrycky. “Systems engineering and analysis.” In Bringing systems into being, 4th ed. 36–37. Upper Saddle River, NJ: Pearson Prentice Hall. Costello, L. R., and K. S. Jones. 2014. “WUCOLS IV: Water use classification of landscape species: A guide to the water needs of landscape plants.” Univ. of California Cooperative Extension. Accessed December 21, 2018. https://ucanr.edu/sites/WUCOLS/. Dawoud, M. A. 2012. “Environmental impacts of seawater desalination: Arabian Gulf Case Study.” Int. J. Environ. Sustainability 1 (3): 22–37. https://doi.org/10.24102/ijes.v1i3.96. Elsheshtawy, Y. 2008. The evolving Arab City: Tradition, modernity and urban development. London: Routledge. Environment Agency – Abu Dhabi (EAD). 2015. “A water budget approach for the Emirate of Abu Dhabi: Policy brief.” Accessed January 23, 2018. https://www.ead.ae. Ervin, S., and C. Steinitz. 2003. “Landscape visibility computation: Necessary, but not sufficient.” Environ. Plann. B: Plann. Des. 30 (5): 757–766. https://doi.org/10.1068/b2968. Gharipour, M. 2016. Contemporary urban landscapes of the Middle East. London: Routledge. Hakky, R. I. 2012. “Residents’ satisfaction with the villa as a housing type in Saudi Arabia.” J. King Saud Univ. Archit. Plann. 24 (2): 153–161. Hayden, L., M. L. Cadenasso, D. Haver, and L. R. Oki. 2015. “Residential landscape aesthetics and water conservation best management practices: Homeowner perceptions and preferences.” Landscape Urban Plann. 144: 1–9. https://doi.org/10.1016/j.landurbplan.2015.08.003. Hong, B., B. Lin, L. Hu, and S. Li. 2011. “Optimal tree design for sunshine and ventilation in residential district using geometrical models and numerical simulation.” Build. Simul. 4 (4): 351–363. https://doi.org/10.1007/s12273-011-0056-1. Javaux, M., T. Schröder, J. Vanderborght, and H. Vereecken. 2008. “Use of a three-dimensional detailed modeling approach for predicting root water uptake.” Vadose Zone J. 7 (3): 1079–1088. https://doi.org/10.2136/vzj2007.0115. Koschke, L., C. Fürst, S. Frank, and F. Makeschin. 2012. “A multi-criteria approach for an integrated land-cover-based assessment of ecosystem services provision to support landscape planning.” Ecol. Indic. 21: 54–66. https://doi.org/10.1016/j.ecolind.2011.12.010. Margolis, L., and A. Chaouni. 2014. Out of water—Design solutions for arid regions. Basel, Switzerland: Birkhäuser. McPherson, E. G., J. R. Simpson, P. J. Peper, S. E. Maco, S. L. Gardner, S. K. Cozad, and Q. Xiao. 2006. Midwest community tree guide: Benefits, costs, and strategic planting. General Technical Report. United States Department of Agriculture. Accessed August 13, 2019. https://www.fs.usda.gov/treesearch/pubs/25927. Ménoret, P. 2014. The Abu Dhabi Guide: Modern Architecture 1950s–1990s. Abū Ẓaby, United Arab Emirates: FIND. Mitchell, K. 2016. “Beyond greening: Approaches to the contemporary landscape in the United Arab Emirates.” In Contemporary Urban Landscapes of the Middle East, edited by M. Gharipour, 198–214. London: Routledge. Nowak, D. J., and J. F. Dwyer. 2007. “Understanding the benefits and costs of urban forest ecosystems.” In Urban and community forestry in the northeast, edited by J. E. Kuser, 25–46. Dordrecht, Netherlands: Springer. O’Reilly, W. 1996. Sustainable landscape design in arid climates. Geneva: Aga Khan Trust for Culture. Roux, X. L., A. Lacointe, A. Escobar-Gutiérrez, and S. L. Dizès. 2001. “Carbon-based models of individual tree growth: A critical appraisal.” Ann. For. Sci. 58 (5): 469–506. https://doi.org/10.1051/forest:2001140. Roy, S., J. Byrne, and C. Pickering. 2012. “A systematic quantitative review of urban tree benefits, costs, and assessment methods across cities in different climatic zones.” Urban For. Urban Greening 11 (4): 351–363. https://doi.org/10.1016/j.ufug.2012.06.006. Siddiqi, A., F. Ereiqat, and L. D. Anadon. 2016. “Formulating expectations for future water availability through infrastructure development decisions in arid regions.” Syst. Eng. 19 (2): 101–110. https://doi.org/10.1002/sys.21337. Soares-Filho, B. S., G. Coutinho Cerqueira, and C. Lopes Pennachin. 2002. “Dinamica—A stochastic cellular automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier.” Ecol. Modell. 154 (3): 217–235. https://doi.org/10.1016/S0304-3800(02)00059-5. Tyrväinen, L., S. Pauleit, K. Seeland, and S. de Vries. 2005. “Benefits and uses of urban forests and trees.” In Urban forests and trees, edited by C. Konijnendijk, K. Nilsson, T. Randrup, and J. Schipperijn, 81–114. Berlin: Springer. Wang, Z., B. Yang, S. Li, and C. Binder. 2016. “Economic benefits: Metrics and methods for landscape performance assessment.” Sustainability 8 (5): 424. https://doi.org/10.3390/su8050424. Zhang, X., Y. Lv, J. Tian, and Y. Pan. 2015. “An integrative approach for solar energy potential estimation through 3D modeling of buildings and trees.” Can. J. Remote Sens. 41 (2): 126–134. https://doi.org/10.1080/07038992.2015.1043004.