Urban greening resources and visibility–soil-moisture coordination in Dubai: Spatiotemporal patterns, spatially varying drivers, and planning implications

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 1 / 44

OPEN ACCESS
Citation: Li J (2026) Urban greening resources
and visibility–soil-moisture coordination in
Dubai: Spatiotemporal patterns, spatially
varying drivers, and planning implications.
PLoS One 21(10): e0360918. https://doi.
org/10.1371/journal.pone.0360918
Editor: Mitiku Badasa Moisa, Wallaga
University, ETHIOPIA
Received: July 24, 2026
Accepted: October 2, 2026
Published: October 23, 2026
Copyright: © 2026 Jianyi Li. This is an open
access article distributed under the terms of
the Creative Commons Attribution License,
which permits unrestricted use, distribution,
and reproduction in any medium, provided the
original author and source are credited.
Data availability statement: No new data
were generated by this study. The underlying
data (including land use, vegetation cover,
and topographic data) were obtained from
publicly available sources, such as the National
Geographic Information Public Service Platform
and government statistical yearbooks. All data
RESEARCH ARTICLE
Urban greening resources and visibility–soil-
moisture coordination in Dubai: Spatiotemporal
patterns, spatially varying drivers, and planning
implications
Jianyi Li *
China Urban Construction Design & Research Institute C.O., LTD, Beijing, China

• 2560731375@qq.com
  Abstract
  Urban greening in desert cities is closely related to atmospheric visibility and
  near-surface water conditions, yet its environmental performance depends on
  vegetation condition, landscape configuration, atmospheric context, and irrigation
  demand. Taking Dubai as the study area and 2010, 2015, and 2020 as reference
  years, this study integrates Landsat imagery, GLC_FCS30D land-use data, atmo-
  spheric visibility, precipitation, and surface soil moisture to examine the spatial and
  temporal coordination between visibility-related atmospheric conditions and water
  availability. It further analyzes the changes in the Normalized Difference Vegetation
  Index (NDVI), Fractional Vegetation Cover (FVC), Leaf Area Index (LAI), and the
  landscape patterns of different land-use types. An optimal-parameter geographical
  detector, ordinary least squares regression, and geographically weighted regression
  are used to identify the stage-dependent and spatially varying associations between
  urban greening resources and environmental coordination.The principal finding is that
  the relationship between urban greening resources and the visibility–soil-moisture
  coordination index was both stage-dependent and spatially nonstationary. FVC was
  the dominant individual factor in 2010 and 2020, whereas ED dominated in 2015,
  while the direction and magnitude of local effects varied among coastal built-up
  areas, urban expansion fringes, and inland desert areas. The results further show
  that: (1) Mean visibility generally declined during the study period, whereas surface
  soil moisture decreased from 2010 to 2015 and subsequently recovered from 2015
  to 2020. Areas with low coordination gradually shifted from the southeastern and
  southern parts of the study area towards the central urban area, indicating a spatial
  mismatch between atmospheric visibility and surface water conditions. (2) From
  2010 to 2020, the proportion of bare land decreased by 2.88 percentage points,
  whereas impervious surfaces and managed vegetation increased by 1.72 and 1.56
  percentage points, respectively. The edge density and patch density of managed
  vegetation increased continuously, indicating that the expansion of green spaces

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 2 / 44
was accompanied by increasing landscape fragmentation. During the same period,
NDVI increased by a dynamic degree of 1.15%, whereas FVC decreased by 0.93%
and LAI changed only slightly, with a dynamic degree of −0.08%. These results
indicate that local vegetation greenness improved without a corresponding increase
in continuous vegetation coverage or canopy structure. (3) The explanatory power
of the driving factors was strongly stage-dependent. FVC was the dominant factor
in 2010 and 2020, whereas edge density was dominant in 2015. The interaction
effects between vegetation indicators and landscape-pattern metrics were generally
stronger than the effects of individual factors, indicating that vegetation condition and
landscape structure jointly differentiated the coordination state. (4) The coefficients of
FVC, edge density, and patch density varied substantially across locations and years,
demonstrating spatially varying associations between urban greening resources and
environmental coordination. These findings suggest that urban greening should be
planned according to local vegetation conditions, landscape structure, atmospheric
visibility, ventilation requirements, and water availability. Coastal built-up areas, urban
expansion fringes, inland areas, and central urban districts should therefore adopt
differentiated vegetation configurations and irrigation strategies. This study provides a
basis for coordinating urban greening, atmospheric-environmental conditions, land-
scape structure, and water-resource allocation in arid cities.

1. Introduction
   Against the backdrop of intensifying global climate change and land degradation,
   sand and dust storms (SDS) have become one of the most significant environmen-
   tal risks in arid and semi-arid regions [1,2]. SDS are typically triggered by strong
   winds, which entrain fine surface particles into the atmosphere through wind erosion.
   These particles can subsequently be transported across regions or even continents
   by large-scale atmospheric circulation, thereby exerting substantial impacts on the
   global environmental system. According to an assessment by the United Nations
   Environment Programme, SDS are not only a typical meteorological hazard but also
   an important manifestation of the interaction between global land degradation and cli-
   mate change. At the global scale, SDS are primarily concentrated within the so-called
   “global dust belt,” which extends broadly from North Africa and West Asia to Central
   and East Asia and contains the world’s major dust-source regions. Nevertheless,
   long-range atmospheric transport can carry dust particles thousands of kilometres,
   affecting cities and ecosystems far from their source areas [3]. Previous studies have
   shown that dust aerosols not only alter the atmospheric radiation balance but also
   affect cloud formation, precipitation processes, and biogeochemical cycles, making
   them an important component of the global climate system.
   A substantial body of research has demonstrated that dust activity has under -
   gone marked changes in several arid regions worldwide in recent decades [ 4]. Cli-
   mate warming, declining precipitation, and vegetation degradation can increase the
   used are cited in the manuscript, and the
   analytical methods and processing procedures
   are described in detail in the Methods section.
   Requests for access to the original data should
   be directed to the third party: Geng Xing
   (ucfuxge@ucl.ac.uk).
   Funding: This study did not receive project-
   specific research funding. China Urban
   Construction Design & Research Institute Co.,
   Ltd., Beijing, China, provided salary support
   for JL through employment. The funder had
   no additional role in the study design, data
   collection and analysis, decision to publish,
   or preparation of the manuscript. JL’s specific
   contributions are described in the Author
   Contributions section.
   Competing interests: JL is employed by
   China Urban Construction Design & Research
   Institute Co., Ltd., Beijing, China. This employ-
   ment does not alter our adherence to PLOS
   ONE policies on sharing data and materials.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 3 / 44
extent of exposed land and reduce soil moisture, thereby intensifying wind erosion and increasing dust-emission inten -
sity. For example, Goudie and Middleton noted in their seminal study that changes in SDS frequency may represent
an important manifestation of global environmental change while also further destabilising the climate system, thereby
creating a feedback mechanism [ 5]. Meanwhile, the impacts of SDS on human societies have received increasing
attention. Epidemiological evidence indicates that airborne particulate-matter concentrations rise substantially during
SDS events, potentially triggering or aggravating respiratory and cardiovascular diseases. A systematic review of more
than 200 studies worldwide found that over 84% reported significant associations between SDS exposure and adverse
health outcomes, including increased morbidity and mortality from respiratory diseases [ 6]. SDS also exert
wide-ranging effects on socioeconomic systems. Dust deposition can reduce agricultural productivity, damage infra -
structure, and disrupt transportation and urban operations by impairing visibility. Relevant research has shown that
increases in SDS frequency may reduce agricultural productivity by 1.5%–24%, placing considerable pressure on food
security and economic development in arid regions [ 7].
With the continued advancement of urbanisation, cities located in arid and semi-arid regions are facing increasingly
severe dust-related risks. On the one hand, global climate change has increased the frequency of droughts and intensified
extreme weather events. On the other hand, human activities, including land-use change, overgrazing, and urban expan-
sion, have aggravated surface disturbance and further increased the potential for dust emissions. Research indicates that
the combined effects of climate change and land-use change are increasing the intensity and frequency of SDS in certain
regions [8]. As urbanisation expands in arid and semi-arid regions, desert cities face a compound challenge involving
atmospheric degradation, exposed land, and severe water scarcity. Urban greening can contribute to surface stabilisation
and environmental improvement, but its performance depends on the spatial arrangement of vegetation, the surrounding
built environment, and the water required to sustain planted areas.
Despite substantial progress, three gaps remain. First, existing studies often evaluate vegetation benefits and water
demand separately, providing limited quantitative evidence on their spatial coordination within desert cities. Second, many
studies report global average relationships and therefore provide insufficient evidence on how the effects of vegetation
cover and landscape structure vary across locations and stages of urban development. Third, vegetation expansion is
sometimes interpreted as evidence of dust mitigation even when direct particulate-matter or dust-event observations are
unavailable. These gaps matter because an intervention that stabilises exposed surfaces may simultaneously increase
irrigation demand or restrict local ventilation. Uniform greening targets may therefore conceal spatial trade-offs and misal-
locate scarce water resources.
This study addresses these gaps by treating mean visibility as an indicator of atmospheric clarity and surface soil
moisture as an indicator of near-surface water availability. The analysis combines land-cover structure, vegetation indices,
geographical detection, global regression, and local regression to identify the spatial and temporal conditions under which
greening resources are associated with different coordination states. Accordingly, the study asks: (1) How did visibility–
soil-moisture coordination vary spatially and temporally in Dubai? (2) Which vegetation and landscape-pattern indicators
showed the strongest and most stable spatial associations with the coordination index? (3) How can the results inform
spatially differentiated greening and water-management strategies under desert conditions?
This study makes four specific contributions. First, it applies a visibility–soil-moisture coordination framework to quantify
the spatial relationship between atmospheric clarity and near-surface water availability in a desert city. Second, it identifies
stage-dependent changes in the relative importance of vegetation cover and landscape structure rather than assuming
that a single indicator dominates throughout the study period. Third, it combines geographical detection, OLS regression,
and GWR to distinguish factor interactions, global associations, and spatially varying relationships. Fourth, it translates
these findings into differentiated planning strategies for coastal built-up areas, urban expansion fringes, and inland desert
areas. These contributions are intended to support exploratory environmental diagnosis and spatial planning; they do not
establish causal effects of greening on dust or particulate pollution.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 4 / 44
2. Research Background
2.1. Formation and management of dust storms in desert regions
International research has generally approached sand and dust storms from the perspective of their formation mecha-
nisms, with particular attention to the combined effects of climatic conditions, surface characteristics, and wind regimes.
Studies have shown that under conditions of limited precipitation, sparse vegetation cover, and high wind speeds, fine
surface particles are more readily entrained by wind and transported into the atmosphere, thereby generating large-scale
dust-transport processes. Recent research further suggests that the combined effects of climate change and land-use
change have increased the frequency of dust storms in some arid regions, with substantial consequences for urban air
quality.
In terms of mitigation measures, ecological restoration is widely regarded as an important means of suppressing dust
emissions. Increasing vegetation cover can enhance soil stability and reduce wind erosion, thereby limiting the gener-
ation and dispersion of airborne dust. Meanwhile, the establishment of shelterbelts and vegetative barriers along urban
peripheries has been shown to modify near-surface wind-field structures, reduce wind speeds, and restrict the transport
of particulate matter into urban areas. Recent studies have also begun to examine the long-term capacity of urban forests
to reduce atmospheric particulate matter. For example, a study of urban shelter forests found that increasing vegetation
indices was significantly associated with declining PM10 concentrations in urban areas and, to some extent, with reduced
health risks related to air pollution [9].
With the continued progression of global climate change, land degradation, and urbanisation, the impacts of dust
activity on urban ecosystems in arid and semi-arid regions have attracted increasing international attention [10]. In recent
years, sand and dust storms (SDS) have been recognised as a major environmental issue affecting urban air quality,
public health, and ecological security. Studies have demonstrated that SDS not only increase airborne particulate-matter
concentrations but also influence climate systems and ecological environments at regional and global scales through
atmospheric transport [11]. Against this background, international research has gradually shifted from isolated wind-
break and sand-stabilisation projects toward integrated ecological-governance approaches that combine urban greening,
ecological restoration, and water-resource management to strengthen the stability of urban ecosystems in arid regions.
Extensive research has been conducted in the Middle East, the arid regions of North America, and Australia, particularly
on dust-generation mechanisms, ecological protection systems, and water-efficient urban greening. These studies provide
important experience for the ecological management of desert cities.
2.2. Urban greening and water-resource allocation
Efficient water use has consistently been a central concern in the greening of desert cities. Because arid regions are
characterised by scarce precipitation and intense evaporation, urban greening must balance ecological benefits against
water consumption. Accordingly, international research has generally emphasised water-efficient landscape design and
drought-tolerant planting schemes to reduce the dependence of urban green systems on water resources. In recent years,
the concept of xeriscaping has been widely adopted in urban planning in arid regions. By reducing the use of water-
intensive lawns, introducing drought-tolerant plant species, and optimising irrigation systems, this approach seeks to
minimise the water demand of urban landscapes. Studies have shown that water-efficient green-space systems designed
to replace conventional lawn-dominated landscapes can substantially reduce irrigation demand and improve water-use
efficiency in urban greening [12]. Meanwhile, the use of drought-tolerant plants in urban greening has received increasing
attention. Recent research indicates that increasing the proportion of native and drought-tolerant species can not only
reduce irrigation demand but also enhance urban ecosystem stability and biodiversity [13]. Urban vegetation also provides
multiple ecosystem services, including air-pollutant capture, carbon sequestration, and regulation of the urban water cycle.
However, in water-scarce urban environments, a pronounced trade-off persists between the ecological benefits of green

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 5 / 44
systems and the availability of water resources. Maximising the ecological functions of urban greening under water-saving
constraints therefore remains an important research priority [14].
In arid and semi-arid cities, greening configuration is not merely a means of improving landscape quality but also a form
of critical ecological infrastructure for strengthening ecological resilience, mitigating sand and dust hazards, and enhanc-
ing the human living environment [15]. Existing studies have demonstrated that well-designed urban green systems can
reduce airborne particulate-matter concentrations through multiple processes, including foliar interception, surface depo-
sition, attenuation of airflow, and regulation of local microclimates, thereby limiting dust dispersion and reducing residents’
exposure risks [16]. Vegetative barriers, multilayered tree–shrub–grass communities, and continuous green corridors are
considered particularly effective in reducing particulate matter and improving air quality along open roads, at residential
boundaries, and in protective spaces on the urban periphery [17].
Mechanistically, the regulation of dust and particulate matter by urban greening operates through three principal
pathways. First, plant leaves, branches, and aerodynamically rough canopies enhance the interception, adsorption, and
deposition of particles, thereby reducing near-surface concentrations of suspended particulate matter. Second, appropri-
ately configured plant communities can modify near-ground wind-field structures, attenuate peak wind speeds, and reduce
the potential for dust entrainment from exposed surfaces. Third, through transpiration and shading, green spaces improve
local thermal conditions and turbulent exchange, which in turn affect pollutant dispersion and particle resuspension [18].
Previous reviews have confirmed that urban vegetation can facilitate pollutant deposition and retention, although its actual
effectiveness depends strongly on plant morphological traits, community structure, spatial configuration, and background
meteorological conditions [19].
It should be noted that the capacity of urban greening to reduce particulate matter does not depend simply on increas-
ing the quantity of vegetation but, more importantly, on the suitability of its spatial configuration. Recent studies suggest
that urban green spaces generally reduce particulate-matter exposure at the city or district scale. In poorly ventilated
environments such as street canyons, however, inappropriate vegetation arrangements may obstruct air exchange and
cause local pollutant accumulation. Urban greening should therefore extend beyond the expansion of green-space area
and place greater emphasis on vertical vegetation structure, canopy porosity, greenbelt width, boundary continuity, and
spatial relationships with roads and buildings. Such considerations are essential for balancing dust reduction, ventilation,
and thermal-environment improvement.
The ecological value of greening configuration is particularly pronounced in desert cities. On the one hand, protective
greenbelts located along urban peripheries and major sand-transport corridors can reduce near-surface wind speeds, sta-
bilise exposed soils, and suppress wind erosion, thereby limiting the influx of windblown dust. On the other hand, internal
urban green-space systems can alleviate the combined effects of air pollution and sand and dust disturbance by improv-
ing local microclimates, increasing surface roughness, and enhancing particulate deposition. Urban greening should
therefore be understood not only as an amenity serving aesthetic and recreational purposes but also as a functional
spatial system for managing sand and dust risks.
2.3. Design and application of landscape coupling
In recent years, landscape-coupling design has gradually emerged as an important concept in urban ecological plan -
ning for arid regions. This approach emphasises the integrated consideration of natural factors, including climatic condi -
tions, topography, vegetation types, and soil structure, and seeks to enhance the multifunctionality of urban ecosystems
through multiscale landscape configuration. Studies have shown that composite landscape systems comprising wind -
break greenbelts, ecological corridors, and urban green-space networks can effectively improve the urban ecological
environment. For example, the establishment of continuous green infrastructure at the landscape scale can not only
strengthen ecological connectivity but also reduce the dispersion of air pollutants and improve the urban microclimate
to a certain extent [ 20].

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 6 / 44
Some studies have also examined the role of vegetation in pollution control from the perspective of plant func -
tions. Plants can capture airborne particulate matter and gaseous pollutants through foliar deposition and adsorp -
tion, thereby reducing ambient pollution levels. This plant-based approach to environmental purification, commonly
referred to as phytoremediation, has increasingly been recognised as a sustainable ecological-management strategy
[21]. Meanwhile, growing attention has been directed towards the influence of landscape configuration on the eco -
logical performance of urban green systems. Relevant studies indicate that the ecological benefits of urban greening
depend not only on vegetation coverage but also on spatial structure, community composition, and management
practices.
2.4. Research gaps and future directions
Overall, international research has made considerable progress in the management of dust storms and the configuration
of urban greening in desert cities. Several limitations, however, remain. First, existing studies have largely focused on
individual ecological interventions or local-scale analyses, whereas integrated management approaches at the urban or
regional scale remain insufficiently explored. Second, under water-resource constraints, systematic quantitative evidence
remains limited regarding how the ecological benefits of urban green systems can be enhanced without substantially
increasing municipal water consumption.
As urban environmental problems become increasingly complex, some scholars have proposed reconsidering
urban ecological governance from the perspective of the water–energy–food–environment nexus. Because these
components are linked through complex interactions, future urban environmental governance requires more integrated
and systems-oriented research approaches. Further studies may advance in three principal directions. First, interdis -
ciplinary research should be strengthened by integrating ecology, meteorology, hydrology, urban planning, and related
fields to establish a comprehensive framework for urban ecological governance. Second, greater emphasis should be
placed on quantitative simulation by combining modelling approaches with long-term monitoring data to evaluate the
dust-reduction performance of alternative greening configurations. Third, intelligent monitoring technologies should
be developed by integrating remote sensing, big data analytics, and sensor networks to dynamically monitor urban
green systems and dust processes, thereby improving the scientific robustness and spatial precision of ecological
governance.
3. Study area and data sources
3.1. Overview of the study area
Dubai is one of the seven emirates of the United Arab Emirates (UAE), located on the southeastern coast of the Persian
Gulf in the northeastern Arabian Peninsula, at approximately 25°16′ N and 55°18′ E (Fig 1). It borders the Persian Gulf
to the west, Sharjah to the north, Abu Dhabi to the south, and lies close to Oman in the southeast. With a coastline of
approximately 72 km, Dubai occupies a strategic position in regional shipping, trade, and urban development. The ter-
rain is generally flat, with most urban areas situated on low-lying coastal plains that gradually transition into the Arabian
Desert inland. The natural landscape is dominated by desert and sparse vegetation, resulting in a fragile ecosystem highly
sensitive to climate change and human disturbance. The UAE lies within a typical arid to hyper-arid climatic zone charac-
terised by limited rainfall, intense evaporation, and low vegetation cover. These conditions pose substantial challenges for
urban planning, ecological management, water-resource allocation, and greening development. Benefiting from its coastal
location and port infrastructure, Dubai has developed into a major international centre for trade, tourism, aviation, and
logistics. Dubai International Airport is one of the world’s busiest international aviation hubs, connecting Europe, Asia, and
Africa. However, rapid urban expansion within a desert environment has intensified pressures associated with extreme
aridity, land degradation, windblown sand, and ecological conservation.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 7 / 44
Dubai has a typical hot-desert climate, characterised by year-round aridity, limited precipitation, extremely hot sum-
mers, and mild winters. Annual precipitation is generally below 100 mm and occurs mainly during winter and early spring,
often as short-duration rainfall events. Summers, particularly from June to September, are extremely hot, with daytime
maximum temperatures frequently exceeding 40 °C and occasionally reaching 45–50 °C (Fig 2). Proximity to the Persian
Gulf also results in high summer humidity, further intensifying thermal discomfort. Winters are comparatively mild, with
average temperatures of approximately 20–25 °C, making this the principal season for tourism and outdoor activities.
The interaction between humid coastal air and the hot desert environment produces pronounced thermal stress. Strong
solar radiation, persistent drought, and evaporation rates far exceeding precipitation further aggravate water scarcity and
constrain vegetation growth.
Dubai has undergone rapid economic transformation over recent decades. Although its early development relied partly
on oil production and exports, limited reserves and exposure to oil-price fluctuations prompted the emirate to pursue
economic diversification from the late twentieth century onwards. Its economy is now dominated by trade, tourism, avia-
tion, financial services, logistics, real estate, and construction. According to the Dubai Statistics Center, non-oil industries
account for more than 90% of Dubai’s gross domestic product. Dubai has also become a major commercial and finan-
cial hub in the Middle East. The Dubai International Financial Centre provides regulatory, legal, and financial-service
infrastructure for institutions operating across the Middle East, Africa, and South Asia. Large-scale urban developments,
Fig 1. Location of the study area (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g001

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 8 / 44
including the Burj Khalifa and Palm Jumeirah, have strengthened the city’s global image and supported tourism and
real-estate growth. Although the property market experienced a substantial adjustment during the 2008 global financial
crisis, it subsequently recovered through policy intervention and renewed international investment. Real estate and con-
struction remain important contributors to Dubai’s economic growth [22].
Dubai is also a highly internationalised city with a predominantly expatriate population. Its resident population exceeds
3.5 million, of whom more than 85% are foreign nationals. Large migrant communities from South Asia, Southeast Asia,
Europe, and North America support the construction, service, commercial, and professional sectors. This demographic
structure reflects Dubai’s dependence on international labour and capital flows and reinforces its status as a global city
[23]. The resulting multicultural environment is evident in the city’s commercial activities, education, entertainment, and
lifestyles. Dubai has developed an extensive transport and infrastructure system. Its automated metro network is among
the largest in the Middle East, while continued investment in roads and public transport has improved urban accessibility.
In recent years, infrastructure development has increasingly incorporated smart-city and sustainability principles, including
intelligent transport systems, green buildings, and renewable-energy applications. Substantial investment has also been
directed towards green infrastructure and low-carbon urban development [24]. Through green-building regulations and
sustainability strategies, the government has promoted energy-efficient technologies and environmentally responsible
materials to reduce the environmental pressures associated with rapid urban growth.
3.2. Data sources
This study selected 2010, 2015, and 2020 as the reference years and constructed a multisource environmental dataset
(Table 1). Vegetation parameters, including the Normalized Difference Vegetation Index (NDVI), Leaf Area Index (LAI),
and Fractional Vegetation Cover (FVC), were derived from Landsat imagery at a spatial resolution of 30 m. Land-use
information, including bare land, impervious surfaces, and artificial vegetation, was obtained from the GLC_FCS30D
dataset. Visibility, precipitation, and surface soil moisture at a depth of 0–7 cm were derived from the Dubai-region subset
of the global ERA5 reanalysis dataset. Google Earth Engine was primarily used for remote-sensing image preprocessing,
annual compositing, and vegetation-parameter retrieval.
4 Research Methods
4.1. Calculation of the air–water coupling coordination degree
Coupling coordination analysis provides an effective means of examining interactions among environmental factors. In
desert cities characterised by extreme climatic conditions, the coordination between atmospheric conditions and water
Fig 2. Monthly weather conditions in Dubai. (Source: WeatherSpark, “Average Weather in Dubai, United Arab Emirates, Year Round”).
https://doi.org/10.1371/journal.pone.0360918.g002

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 9 / 44
availability is particularly important for sustainable urban development. In this study, the air–water coupling coordination
degree was assessed through the following three components.
(1) Spatiotemporal variation in mean visibility: Visibility is commonly used as an indirect indicator of atmospheric condi-
tions because it is sensitive to changes in airborne particulate matter, including PM10 and PM2.5. Analysing visibility
across different periods and locations can reveal the spatiotemporal evolution of atmospheric pollution under varying
climatic conditions. It also provides a basis for examining the relationships among dust storms, dust-weather events,
and urban air quality.
(2) Spatiotemporal variation in soil moisture: Soil moisture is a key indicator of water availability and directly affects vege-
tation growth, ecosystem functioning, and the environmental regulation capacity of urban green spaces. Its spatiotem-
poral variation can be used to evaluate the capacity of different greening configurations to retain soil water. In desert
cities, changes in soil moisture are closely related to irrigation demand and water-use efficiency, thereby influencing
the coordination between atmospheric conditions and water resources.
(3) Coupling coordination analysis: The coupling coordination model integrates atmospheric and water-resource indica-
tors to quantify the interaction and coordination between the two subsystems. By incorporating indicators such as visi-
bility, particulate-matter concentration, and soil moisture, the model evaluates the extent to which air quality and water
availability develop in a coordinated manner. The resulting coupling coordination degree can further be used to assess
the environmental effects of different urban greening configurations and water-resource management strategies.
4.2. Construction of the urban greening resource driver system
Urban green spaces play a critical role in the environmental management of desert cities. Quantitative assessment of
greening resources can clarify their contributions to dust-storm mitigation, air-quality improvement, and water-resource
regulation. In this study, urban greening resources were characterised using two groups of indicators: landscape-pattern
metrics and vegetation indices.
Table 1. Summary of data sources.
Data Type Specific Indicators Spatial
Resolution
Data Source
Remote-sensing vege-
tation data
NDVI, LAI, and FVC 30 m Landsat 5/7/8/9 Collection 2 imagery. Cloud removal, annual compositing, and
vegetation-parameter retrieval were conducted using the Google Earth Engine
(GEE) cloud-computing platform. USGS imagery download: https://earthexplorer.
usgs.gov; GEE platform: https://earthengine.google.com
Meteorological data Visibility and total daily
precipitation
0.25° ERA5 hourly data on single levels from 1940 to present were used to obtain
visibility, total precipitation, and volumetric soil water layer 1 (0–7 cm) for the Dubai
region in 2010, 2015, and 2020. The dataset was produced by the European Cen-
tre for Medium-Range Weather Forecasts (ECMWF) and distributed through the
Copernicus Climate Change Service (C3S) Climate Data Store. It provides globally
gridded hourly data at a spatial resolution of 0.25° × 0.25°. The hourly variables
were aggregated to the daily and annual temporal scales required for the analysis.
Official website: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-sin-
gle-levels?tab=overview; dataset DOI: https://doi.org/10.24381/cds.adbb2d47.
Soil data Surface soil moisture at a
depth of 0–7 cm
0.25°
Land-use data Bare Land, Impervious
Surface, Managed Veg-
etation, Sparse Vegeta-
tion, Water, and Wetland
30 m Land-cover maps for 2010, 2015, and 2020 were obtained from the GLC_FCS30D
global 30 m land-cover dynamics dataset, which provides annual land-cover maps
after 2000. The original land-cover subclasses were reclassified into six catego-
ries, namely impervious surface, managed vegetation, sparse vegetation, bare
land, water, and wetland, using a consistent crosswalk for all three years. The
same spatial mask, reclassification scheme, and raster-processing parameters
were applied to ensure temporal comparability.
https://doi.org/10.1371/journal.pone.0360918.t001

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 10 / 44
4.2.1. Landscape-pattern metrics. Landscape-pattern analysis uses remote-sensing imagery and land-use
classification to evaluate the spatial distribution and structural characteristics of urban green spaces. Green-space
patches were extracted from the classified land-use data and analysed using FRAGSTATS and related spatial-analysis
tools. The resulting metrics describe patch composition, fragmentation, geometric complexity, aggregation, and
connectivity, providing a quantitative basis for examining how green-space patterns affect air quality, dust control, and
water-resource use.
The principal landscape-pattern metrics are described below.
(1) Percentage of Landscape (PLAND): PLAND represents the proportion of the total landscape occupied by a particu-
lar green-space patch type. It can be used to assess the relative abundance of different green-space types and their
potential contributions to air-quality improvement and microclimate regulation.
(2) Patch Density (PD): Patch density refers to the number of patches of a particular land-cover type per unit area. A
higher PD generally indicates a greater degree of landscape subdivision or fragmentation. It is therefore an important
indicator of the fragmentation and spatial continuity of urban green spaces.
(3) Edge Density (ED): Edge density represents the total length of patch boundaries per unit landscape area, commonly
expressed as metres per hectare. It reflects the extent of edge development and the geometric complexity of green-
space patches. Higher ED values generally indicate more complex or fragmented patch boundaries, although their
ecological effects depend on the surrounding landscape context.
(4) Landscape Shape Index (LSI): The Landscape Shape Index measures the overall geometric complexity of landscape
patches. Higher LSI values indicate increasingly irregular patch shapes and more complex spatial structures. In urban
greening, patch shape can influence airflow regulation, ecological interactions, habitat provision, and the exposure of
vegetation to surrounding environments. Appropriate shape design may therefore contribute to wind attenuation and
dust interception in desert cities.
(5) Shannon’s Diversity Index (SHDI): Shannon’s Diversity Index measures the diversity of land-cover or green-space
types within a landscape. Higher values indicate greater compositional diversity and a more balanced distribution
among landscape types. In desert cities, a diversified green-space system may enhance ecological adaptability and
multifunctionality, particularly when drought-tolerant species and vegetation communities with different ecological func-
tions are appropriately combined.
(6) Aggregation Index (AI): The Aggregation Index measures the spatial concentration of patches belonging to the same
landscape type. Higher values indicate that patches are more strongly clustered. Aggregated green spaces can
improve habitat continuity, strengthen ecological stability, and facilitate the concentrated use of irrigation resources.
In desert cities, an appropriate degree of aggregation may also enhance wind protection, sand stabilisation, and local
microclimate regulation.
(7) Proximity Index (PROX): The Proximity Index evaluates the spatial relationships between a focal patch and neighbour-
ing patches of the same type within a specified search distance. Higher values indicate that nearby patches are larger,
more numerous, or more closely distributed. A high proximity level may support the formation of ecological corridors
and strengthen functional linkages among urban green spaces.
(8) Effective Area Index (EAI): The Effective Area Index characterises the relationship between the usable interior area of
a green-space patch and its boundary effects. Higher values generally indicate a larger effective core area and a lower
relative influence of patch edges. Green spaces with larger effective areas may provide more stable ecological func-
tions, including pollutant removal, dust interception, and efficient water use.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 11 / 44
(9) Landscape Connectivity Index (LCI): The Landscape Connectivity Index measures the degree of spatial or functional
connection among green-space patches. Higher values indicate a more continuous and integrated ecological network,
which can facilitate species movement, ecological flows, and the maintenance of ecosystem functions. In desert cities,
connected green-space networks may form protective corridors, improve dust interception, and enhance the overall
environmental performance of urban greening systems.
4.2.2. Vegetation characterisation indices. Vegetation indices are important indicators of the ecological condition
and ecosystem-service capacity of urban green spaces. They facilitate quantitative assessment of the role of urban
vegetation in improving air quality and reducing PM2.5 and PM10 concentrations [25]. The principal vegetation indicators
considered in this study are described below.
(1) Normalised2 Vegetation Characterisation Indices: Vegetation indices are important indicators of the ecological con-
dition and ecosystem-service capacity of urban green spaces. They facilitate quantitative assessment of the role of
urban vegetation in improving air quality and reducing PM2.5 and PM10 Difference Vegetation Index (NDVI)
The Normalised Difference Vegetation Index is widely used to characterise vegetation greenness and growth condition.
NDVI enables the quantitative assessment of vegetation distribution and activity and can therefore be used to examine
the potential effects of different greening types on urban air quality. In desert cities, it is particularly useful for identifying
spatial variations in vegetation condition and evaluating the ecological performance of irrigated green spaces.
(2) Fractional Vegetation Cover (FVC): Fractional Vegetation Cover represents the proportion of the ground surface
covered by vegetation. Higher FVC generally indicates greater vegetation continuity and a stronger capacity for soil
stabilisation, erosion control, and microclimate regulation. In desert cities, increased FVC may reduce exposed soil
surfaces, suppress wind erosion, and contribute to lower concentrations of windblown PM10 and PM2.5.
(3) Leaf Area Index (LAI): The Leaf Area Index is defined as the one-sided leaf area per unit ground area and is a key
indicator of vegetation canopy structure. Higher LAI values generally indicate denser foliage and a larger leaf surface
available for photosynthesis, evapotranspiration, carbon uptake, and pollutant deposition. The desert cities, LAI pro-
vides important information on vegetation growth and ecological functioning under extreme climatic and water-
resource constraints. It can be estimated using remote sensing, field measurements, and vegetation-growth models.
LAI data can support urban greening design and water-resource management by identifying areas with insufficient
canopy development or excessive irrigation demand. Appropriate enhancement of LAI may improve carbon seques-
tration, thermal regulation, and particulate-matter interception, although its effects depend on plant species, canopy
porosity, irrigation conditions, and local airflow.
(4) Vegetation Height: Vegetation height is a basic indicator of plant growth and the vertical structure of urban green
spaces. It influences airflow, shading, roughness, and the interception of airborne particles. Taller vegetation, particu-
larly trees and multilayered plant communities, can attenuate wind speeds and intercept windblown dust. However, its
actual effect on pollutant dispersion depends on planting density, canopy permeability, surrounding buildings, and local
ventilation conditions.
(5) Vegetation Biomass: Vegetation biomass represents the total quantity of plant material within a green space or study
area, including above-ground and below-ground components such as roots, stems, branches, and leaves. It reflects
the capacity of vegetation to provide ecosystem services, including carbon sequestration, microclimate regulation, soil
stabilisation, and pollutant interception. Biomass can be estimated by combining remote-sensing data, field sampling,
and vegetation-growth models. Monitoring vegetation biomass is particularly important in desert cities because plant
growth is strongly constrained by water availability, temperature, and soil conditions. Temporal changes in biomass
can therefore indicate the ecological adaptability and resource-use efficiency of urban vegetation under different

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 12 / 44
climatic conditions. Increasing biomass through appropriately diversified and water-efficient planting may strengthen
wind and sand control, thermal regulation, and pollutant-removal capacity.
(6) Vegetation Morphology: Vegetation morphology refers to external structural traits such as plant height, crown width,
branching form, leaf size, and leaf shape. These characteristics affect the interactions between vegetation and its
surrounding environment and consequently influence shading, airflow regulation, particle capture, and water consump-
tion. The assessment of vegetation morphology is especially relevant in desert cities, where plant form is closely asso-
ciated with climatic adaptation, soil conditions, and irrigation availability. Drought-tolerant species commonly possess
smaller leaves, specialised canopy structures, or extensive root systems that improve their ability to withstand water
stress. Trees with broad crowns can provide substantial shade and alleviate urban heat, whereas taller or multilayered
vegetation may serve as windbreaks and intercept airborne dust. Morphological analysis therefore provides an import-
ant basis for selecting plant species and optimising planting structures.
(7) Green View Index (GVI): The Green View Index measures the proportion of visible vegetation within streets or pub-
lic spaces and is commonly calculated from street-view imagery. Unlike remotely sensed indicators that primarily
describe vegetation from an overhead perspective, GVI reflects greenery as directly perceived by pedestrians. It is
therefore relevant to the assessment of environmental quality, visual comfort, psychological well-being, and urban
liveability. GVI is usually calculated by identifying vegetation pixels in street-view images using image-processing or
computer-vision techniques. In desert cities, where the expansion of green-space area is constrained by limited land
and water resources, improving visible greenery through strategic spatial design may offer a resource-efficient means
of enhancing the urban environment. Quantitative comparisons of GVI among neighbourhoods can identify disparities
in pedestrian-level greening and support the optimisation of vegetation location, vertical structure, and visual exposure
in densely used urban areas.
4.3. Global and local models of driving mechanisms
To investigate the relationship between landscape patterns and the air–water coupling coordination degree, this study
developed global and local driving models to quantify the effects of landscape-pattern characteristics on atmospheric con-
ditions and water availability. A geographically weighted regression model and the optimal-parameter geographical detec-
tor (GeoDetector) were employed to identify the overall and spatially differentiated effects of landscape-pattern changes
on urban ecological functions, thereby providing a quantitative basis for urban greening planning and
water-resource management.
The global analysis focused on the overall relationships between landscape-pattern metrics and the air–water coupling
coordination degree. XGBoost was used to assess the combined effects of indicators such as percentage of landscape,
patch density, and edge density on atmospheric and water-resource conditions, including particulate-matter concentra-
tions, soil moisture, and irrigation demand. This approach identifies the relative importance and nonlinear effects of land-
scape characteristics across the entire study area, thereby clarifying the overall environmental performance of alternative
green-space configurations.
The local analysis examined the spatial heterogeneity of these relationships. In desert cities, differences in climate, soil,
vegetation, and urban development may cause the effects of landscape patterns to vary substantially among locations.
The geographically weighted model was therefore used to estimate spatially varying relationships, while GeoDetector was
applied to identify the explanatory power and interactive effects of individual driving factors. These analyses reveal where
particular landscape characteristics exert stronger influences on air–water coupling coordination. For example, improving
green-space connectivity or optimising patch distribution may produce greater benefits in certain areas by enhancing dust
interception, air-quality regulation, and soil-water retention.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 13 / 44
4.3.1. Geographically weighted regression model. The geographically weighted Regression model integrates
spatial weighting with the Regression machine-learning algorithm, enabling the simultaneous assessment of global
nonlinear relationships and local spatial variations. By assigning distance-based weights to observations, the model
accounts for geographical differences in the effects of landscape-pattern characteristics on the air–water coupling
coordination degree. The bandwidth selected through spatial block cross-validation was used for the final GWR estimation
in each year.
The model was used to identify the relationships between landscape characteristics, including green-space types,
patch distribution, and landscape connectivity, and environmental conditions such as air quality and water-resource use.
It also quantified the relative contribution of individual landscape features across different locations. The results provide
a quantitative basis for optimising green-space configuration, controlling dust hazards, and improving water-resource
management.
To evaluate the generalisability of the GWR models, five-fold spatial block cross-validation was conducted separately
for 2010, 2015, and 2020. Spatially adjacent observations were assigned to the same block to reduce information leakage
caused by spatial dependence. Within each training fold, adaptive neighbourhood sizes of 50, 100, 150, 200, 250, and
300 observations were evaluated, and the bandwidth with the lowest validation error was selected. The model was then
refitted using the training observations only and used to predict the observations in the held-out spatial block. Predictive
performance was assessed using the cross-validated coefficient of determination, root mean square error, and mean
absolute error. The cross-validated R² was calculated from held-out predictions rather than from the fitted values of the
training data. Thus, the reported global R² and the spatial cross-validated R² represent sample-fitting performance and
spatial out-of-sample performance, respectively.
4.3.2. Optimal-parameter geographical detector. The optimal-parameter geographical detector was employed
to examine the spatial associations between landscape-pattern characteristics and the air–water coupling coordination
degree. Its principal advantage lies in its capacity to quantify the explanatory power of individual factors and identify
interactions among multiple drivers without requiring a predefined linear relationship.
Landscape metrics, including percentage of landscape, patch density, and edge density, were discretised using opti-
mised classification parameters before factor and interaction detection. This approach enabled the identification of the
landscape characteristics that most strongly explained spatial variations in coupling coordination. It also revealed whether
interactions between factors produced enhanced, nonlinear, or weakened explanatory effects. The results can support
spatially differentiated greening strategies; for example, improving green-space connectivity or landscape diversity may
enhance air-quality regulation and water-use efficiency in specific areas.
4.4. Validation through small-sample field surveys
To assess the reliability and practical relevance of the large-scale analysis, this study combined multisource big-data
analysis with small-sample field surveys. The city-scale assessment integrated remote-sensing imagery, environmental
monitoring data, and landscape-pattern metrics to evaluate urban greening and environmental conditions. Although these
analyses provided an overall basis for greening planning and environmental optimisation, field validation was necessary
because climatic, geographical, and socioeconomic conditions vary considerably across urban areas.
Representative small-scale sites were therefore selected for field observation and data collection. The surveys were
used to examine whether the spatial patterns identified through large-scale datasets were consistent with actual site con-
ditions. This process also supported the formulation of site-specific landscape-design strategies and improved the practi-
cal applicability of the analytical results.
4.4.1. Framework integrating big-data analysis and small-sample surveys. The study first conducted a city-scale
assessment incorporating green-space distribution, climatic conditions, atmospheric conditions, and water-resource
availability. Remote-sensing and geographic information system data were used to identify landscape characteristics

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 14 / 44
associated with environmental quality and ecological balance and to assess their contributions to the air–water coupling
coordination degree. These results provided a preliminary basis for optimising urban greening and water-resource
management.
Representative locations were subsequently selected for small-scale field surveys. Site observations and environmen-
tal measurements were used to collect detailed information on landscape patterns, atmospheric conditions, soil moisture,
and microclimatic characteristics. The field evidence was then compared with the large-scale analytical results to evaluate
their consistency and reliability.
4.4.2. Site selection and data collection. Survey sites were selected from key areas identified through the large-
scale spatial analysis. To capture differences in geographical and environmental conditions, the sample included
urban core areas, peripheral expansion zones, sparsely vegetated areas, and locations with relatively high exposure
to windblown dust. All the locations involved in our work are open to the public and do not fall under any restricted or
protected areas
Field data collection included the following components:
(1) Air-quality monitoring: Portable monitoring instruments were used to measure PM2.5 and PM10 concentrations. The
observed values were compared with the spatial patterns estimated from the large-scale environmental datasets.
(2) Soil moisture and water-resource use: Soil-moisture sensors and relevant water-use records were used to assess
soil-water conditions and the irrigation efficiency of different green-space types.
(3) Landscape patterns and green-space distribution: Field observations and GIS-based spatial analysis were combined
to document the location, form, composition, and connectivity of green-space patches.
(4) Microclimatic monitoring: Temperature, relative humidity, and wind speed were measured using portable environmen-
tal instruments to examine the influence of landscape configuration on local microclimatic conditions.
4.4.3. Comparison between field-survey and big-data results. Following the field surveys, site-level observations
were compared with the results of the large-scale spatial analysis. This comparison was used to identify potential
deviations caused by data resolution, model assumptions, or local environmental conditions and to refine the
interpretation of the analytical results.
Field observations also revealed fine-scale environmental variations that were difficult to capture using city-scale
datasets. For example, inappropriate planting density, discontinuous green-space configuration, or inefficient irrigation
could result in locally poor atmospheric conditions or low water-use efficiency. The surveys therefore provided evidence
for assessing the applicability of the proposed greening strategies. By linking large-scale spatial patterns with site-specific
conditions, the validation process strengthened the practical basis for landscape design and environmental management.
5. Results
5.1. Analysis of the air–water coupling coordination degree
5.1.1. Spatiotemporal evolution of mean visibility. Fig 3 shows that mean visibility in Dubai exhibited a clear spatial
reorganisation between 2010 and 2020. In 2010, relatively high visibility covered most of the study area. High-value
areas extended continuously from the southwestern coast towards the south-central inland region, whereas moderate-
and low-value areas were mainly concentrated along the northern coast and in several urban-development zones. By
2015, moderate- and low-visibility areas had expanded across the built-up areas along the Persian Gulf coast and the
central urban region. At the same time, high-visibility areas contracted to the southwestern coast and several locally
distributed areas in the north. The 2020 pattern remained broadly similar to that of 2015, with low and moderately low
values concentrated in the northern coastal urban belt and central region, while relatively high values persisted in the
southwestern coastal area and the southeastern desert interior.

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 15 / 44
These results indicate that visibility gradually changed from a relatively homogeneous high-value pattern to a more
differentiated coastal–inland structure. Lower visibility was concentrated in intensively developed coastal and central
urban areas, whereas higher visibility was generally observed in peripheral desert areas. This spatial contrast is com -
patible with the combined influence of urban development intensity, local atmospheric emissions, maritime humidity,
surface disturbance, and the openness of the surrounding desert landscape. The persistence of lower visibility in the
northern coastal urban belt also suggests that the deterioration of atmospheric clarity was spatially associated with
the main urban-development corridor rather than being distributed uniformly across Dubai. However, because visibility
is affected by dust, humidity, precipitation, atmospheric particles, and meteorological conditions simultaneously, the
observed pattern should be interpreted as an overall atmospheric-clarity signal rather than as a direct measure of dust
concentration alone.
5.1.2. Spatiotemporal evolution of soil moisture. The spatial distribution of surface soil moisture also changed
substantially over the study period (Fig 4). In 2010, soil moisture ranged from 0.049 to 0.089. Higher values were
concentrated in the western and southwestern coastal areas and gradually declined towards the central and eastern
inland regions, forming a pronounced coast-to-desert moisture gradient. In 2015, the overall range decreased to 0.027–
0.056. High-value areas became more localised in the northern and northeastern parts of the study area, while broad low-
value zones emerged in the south-central and southwestern inland regions. This pattern indicates that the middle study
period was characterised by a widespread reduction in near-surface moisture and a stronger spatial contrast between
relatively moist urban or coastal areas and dry inland areas.
Fig 3. Spatiotemporal evolution of mean visibility (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g003
Fig 4. Spatiotemporal evolution of soil moisture (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g004

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 16 / 44
By 2020, soil moisture recovered to a range of 0.041–0.062, and a distinct high-value centre reappeared along the
western coast. Nevertheless, the central urban area and eastern inland region remained relatively dry. The recovery
was therefore spatially uneven and did not restore the earlier moisture configuration across the entire study area. The
Hatta exclave exhibited comparatively high soil moisture in both 2015 and 2020, which differs from the main desert
plain of Dubai. This difference is consistent with the distinct topographic, geological, and hydrological conditions of
Hatta. The results suggest that the temporal variation in soil moisture was shaped by both broad regional conditions
and local landscape settings. In particular, the persistence of dry conditions in the central and eastern regions indicates
that the expansion of urban greening alone may not guarantee a corresponding improvement in near-surface water
availability.
5.1.3. Coupling coordination analysis. Fig 5 presents the spatial distribution of the air–water coupling coordination
degree. In 2010, moderate and high coordination values dominated Dubai. High-value areas were mainly distributed along
the western and southwestern coast, while the central and northern urban areas also maintained relatively favourable
coordination levels. Low-value areas were concentrated primarily along the southeastern desert fringe. In 2015, the low-
coordination centre shifted towards the southern inland region. At the same time, relatively continuous high-value areas
appeared in the northern coastal built-up zone and the northeastern region. Accordingly, the dominant spatial gradient
changed from a west–east pattern in 2010 to a more pronounced north–south differentiation in 2015.
In 2020, moderate and high coordination values remained prevalent in many peripheral areas, but a distinct low-value
core emerged in the central urban region and extended towards the east and south. The spatial location of the low-
coordination centre therefore shifted successively from the southeastern desert fringe to the southern inland area and
then towards the central urban region. This result demonstrates that the air–water coordination state did not follow a sim-
ple linear temporal trend. Instead, the spatial focus of environmental constraint changed with the reorganisation of visibil-
ity, soil moisture, land use, and urban development.
The coordination index should be interpreted as a measure of the degree of balance between atmospheric clarity and
near-surface water availability. A high value does not necessarily mean that both subsystems reached their maximum
levels; it indicates that their spatial states were relatively compatible. Conversely, a low value may result from poor visi-
bility, insufficient soil moisture, or a mismatch between the two subsystems. The emergence of low values in the central
urban area in 2020 is therefore likely related to the simultaneous presence of relatively poor atmospheric conditions and
persistent surface dryness. The migration of low-coordination areas further indicates that environmental management
priorities cannot be determined solely by the amount of green space. They must also account for the spatial relationship
between atmospheric conditions, water availability, and the surrounding urban landscape.
Fig 5. Spatial distribution of the air–water coupling coordination degree (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC
BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g005

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 17 / 44
5.2. Urban greening indicators
5.2.1. Landscape-pattern metrics. The six land-use classes exhibited substantially different spatial configurations
and temporal trajectories. These differences indicate that changes in the quantity of green space were accompanied
by changes in patch structure, boundary complexity, and spatial continuity. Bare land was the dominant land-use type
in Dubai. As shown in Fig 6, high edge-density and patch-density values were concentrated mainly along the northern
coastal development fringe, around transport corridors, and at the interfaces between built-up areas and the inland desert.
By contrast, the south-central desert interior contained extensive and relatively continuous bare land, which produced
lower local edge complexity despite its large total area. Between 2010 and 2020, the spatial framework of high-value bare-
land areas remained broadly stable, but local patches became increasingly subdivided and more strongly aligned with
urban expansion boundaries and road networks. The reduction in bare-land area therefore occurred together with greater
fragmentation of the remaining patches. This indicates that the contraction of the desert matrix was spatially selective,
with the most pronounced structural changes occurring at urban–desert transition zones.
Impervious surfaces were concentrated along the northern and northwestern Persian Gulf coast and extended
inland along major transport corridors ( Fig 7). This distribution reflects the combined effects of coastal urban concen -
tration and corridor-based expansion. High edge-density values occurred along development fronts, built-up boundar -
ies, and linear infrastructure, whereas relatively high patch density was observed in the northern old city and coastal
Fig 6. Spatial distribution of landscape-pattern metrics for bare land (C1) (Boundary data: geoBoundaries, https://www.geoboundaries.org/,
CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g006

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 18 / 44
areas with dense road networks. From 2010 to 2020, the proportion of impervious surfaces increased from 15.27%
to 16.99%. However, edge density declined from 34.52 to 33.66 m/ha, and patch density declined from 3.61 to 3.18
patches/100 ha. The simultaneous expansion of impervious surfaces and decline in their edge and patch densities
indicates that urban development was accompanied by a degree of spatial consolidation. In other words, built-up land
expanded outward while some previously separated urban patches became more spatially connected or integrated
into larger developed areas.
Managed vegetation remained concentrated in the northern coastal built-up area, residential and public-service spaces,
roadside greenbelts, and several inland development nodes (Fig 8). It did not form a continuous citywide network. Instead,
it occurred as small patches, linear corridors, and isolated green spaces embedded within the built-up landscape. Its per-
centage of landscape increased from 3.30% in 2010 to 4.86% in 2020. During the same period, edge density increased
from 14.37 to 21.18 m/ha, and patch density increased from 2.85 to 4.10 patches/100 ha. The increase in all three metrics
demonstrates that the expansion of managed vegetation was accompanied by increasing boundary complexity and patch
subdivision. Therefore, the growth of urban greening was expressed primarily through dispersed parks, roadside planting,
residential green spaces, and small development-related patches rather than through the formation of a continuous eco-
logical network. This distinction is important because a larger green-space area does not automatically indicate stronger
connectivity, greater ecological continuity, or more efficient use of irrigation water.
Fig 7. Spatial distribution of landscape-pattern metrics for impervious surfaces (C2) (Boundary data: geoBoundaries, https://www.geobound-
aries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g007

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 19 / 44
Sparse vegetation displayed a more stable and semi-natural spatial pattern ( Fig 9). It was mainly distributed along
the outer northern coastal urban area, at the interfaces between built-up land and bare land, and along local desert
margins. A relatively concentrated distribution also occurred in the Hatta exclave. Compared with managed vegetation,
sparse vegetation consisted of more numerous but generally smaller patches and showed weaker spatial continuity.
From 2010 to 2020, its edge density increased from 14.26 to 15.58 m/ha, patch density increased from 3.45 to 3.91
patches/100 ha, and its percentage of landscape increased only slightly from 5.24% to 5.40%. These limited changes
suggest that natural or semi-natural vegetation experienced only modest spatial expansion. The overall greening tran -
sition in Dubai was therefore driven mainly by managed vegetation rather than by the recovery or expansion of sparse
natural vegetation.
Water bodies were concentrated along the Persian Gulf coastline, around offshore artificial islands and harbours,
and within several internal urban water systems (Fig 10). Their spatial configuration combined a coastal linear structure
with discrete artificial-water nodes. High edge-density values occurred mainly along the northern coast, where irregular
shorelines, artificial islands, and frequent land–water interfaces generated complex boundaries. Between 2010 and 2020,
water-body edge density declined from 2.38 to 1.03 m/ha, patch density declined from 0.55 to 0.50 patches/100 ha, and
the percentage of landscape declined from 1.31% to 0.77%. The main locations of water patches remained relatively sta-
ble, but their extent and boundary complexity decreased. Water patches in Hatta remained isolated from the coastal water
Fig 8. Spatial distribution of landscape-pattern metrics for managed vegetation (C3) (Boundary data: geoBoundaries, https://www.geobound-
aries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g008

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 20 / 44
system and exhibited a different spatial form, confirming the functional and geographical separation between the main
Dubai territory and the exclave.
Wetlands were mainly distributed in low-lying northern coastal areas, around urban waterways, and within artificial
waterfront spaces (Fig 11). They formed a combined point–line pattern associated with built-up areas and transport
networks. Compared with open water, wetlands extended further into the urban interior but generally consisted of smaller
patches with more complex boundaries. Their percentage of landscape declined slightly from 0.37% to 0.34%, while edge
density declined from 1.84 to 1.73 m/ha and patch density changed from 0.67 to 0.64 patches/100 ha. The relatively small
net changes, together with local expansion and contraction, indicate that the overall wetland framework remained stable
but experienced local spatial reorganisation.
5.2.2. Vegetation indices. The spatial distributions of NDVI, FVC, and LAI revealed distinct dimensions of vegetation
condition (Fig 12). NDVI and FVC displayed broadly similar patterns. Both indices were generally low in the densely
built-up areas along the northern Persian Gulf coast, whereas relatively continuous moderate- and high-value areas were
observed in the central-eastern and southeastern inland regions. The maximum NDVI increased from 0.42 in 2010 to 0.46
in 2015 and then declined slightly to 0.44 in 2020. Its minimum value decreased from −0.24 to −0.28 before recovering to
−0.21. FVC ranged from 0 to 1 in all three years, although the locations of high- and low-value areas shifted between the
northern urban periphery and the south-central inland region.
Fig 9. Spatial distribution of landscape-pattern metrics for sparse vegetation (C4) (Boundary data: geoBoundaries, https://www.geoboundar-
ies.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g009

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 21 / 44
LAI was substantially lower overall and exhibited a more concentrated spatial distribution. High LAI values were limited
mainly to northern coastal greening nodes, several urban parks, and the Hatta exclave. The maximum LAI increased from
1.187 in 2010 to 1.291 in 2020. The contrast between the relatively broad distribution of NDVI and FVC and the limited
distribution of high LAI values indicates that vegetation greenness or surface coverage did not necessarily correspond to
a dense or structurally developed canopy. In desert cities, moderate vegetation-index values may reflect sparse or discon-
tinuous vegetation, seasonal growth, or localised irrigated surfaces. LAI therefore provides additional information about
vegetation structure that cannot be obtained from greenness or fractional cover alone.
The change maps in Fig 13 show that NDVI variation was highly localised. Areas with marked increases or decreases
were mainly distributed in the northern coastal urban belt, development fringes, and several inland development nodes.
Between 2010 and 2015, increasing and decreasing NDVI patches were interspersed. From 2015 to 2020, NDVI became
more stable across much of the study area, although a notable decline occurred in the southern region. Changes in FVC
covered a broader spatial extent. Declining patches were more widespread during the first period, whereas relatively
extensive recovery occurred in the northern and central areas during the second period. Several inland areas, however,
continued to experience declines. Changes in LAI were concentrated mainly in northern coastal greening nodes and the
Hatta exclave, while most of the desert interior remained stable. These results indicate that vegetation change was spa-
tially uneven and was closely associated with managed green-space development, vegetation renewal, and local environ-
mental conditions.
Fig 10. Spatial distribution of landscape-pattern metrics for water bodies (C5) (Boundary data: geoBoundaries, https://www.geoboundaries.
org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g010

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 22 / 44
Table 2 quantifies the temporal differences among the three vegetation indicators. From 2010 to 2015, the mean annual
dynamic degree of NDVI was 0.62%, while FVC and LAI declined by 0.15% and 1.35%, respectively. From 2015 to 2020,
NDVI continued to increase at an annual rate of 0.51%, whereas the decline in FVC intensified to 0.79%. During the same
period, LAI shifted to a positive annual rate of 1.36%. Across the full 2010–2020 period, the cumulative dynamic degrees
of NDVI, FVC, and LAI were 1.15%, −0.93%, and −0.08%, respectively. The three indicators therefore described different
aspects of vegetation change. NDVI increased consistently, suggesting an overall improvement in vegetation greenness.
FVC declined over both intervals, indicating that the spatial extent or continuity of vegetation cover did not improve at the
same rate. LAI decreased initially and subsequently recovered, resulting in a nearly stable cumulative change over the full
decade. The results suggest that vegetation improvement in Dubai was expressed mainly through enhanced greenness
and localised canopy recovery rather than through a citywide expansion of continuous vegetation cover. The reversal of
LAI between the two periods may reflect vegetation maintenance, irrigation, or renewal in specific managed green spaces,
but the spatial concentration of these changes indicates that the recovery was not representative of the entire urban
region.
Fig 14 confirms the different trajectories of the three indicators. NDVI increased during both five-year periods, although
its rate of increase was slightly lower after 2015. FVC remained negative throughout the study period, with a stronger
decline during 2015–2020. LAI showed the greatest fluctuation, declining by 1.35% during 2010–2015 and increasing by
Fig 11. Spatial distribution of landscape-pattern metrics for wetlands (C6) (Boundary data: geoBoundaries, https://www.geoboundaries.org/,
CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g011

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 23 / 44
1.36% during 2015–2020. The divergence between NDVI and FVC is particularly important. Although both indices were
strongly correlated spatially, they responded differently over time, indicating that vegetation greenness and vegetation
coverage should not be treated as interchangeable measures of urban greening performance.
Fig 15 shows that Dubai retained a pronounced coastal–inland land-use gradient. Impervious surfaces formed a con-
tinuous built-up belt along the northern and northwestern Persian Gulf coast, surrounding the urban core, ports, artificial
islands, and major transport corridors. Bare land dominated the central, southern, and southeastern areas and formed
the principal desert matrix. Managed and sparse vegetation were embedded mainly within built-up areas, development
fringes, and transport corridors, while water bodies and wetlands were concentrated along the coast and around artifi-
cial waterfront spaces. The broad spatial framework remained stable from 2010 to 2020, although built-up land gradually
expanded along the northern coast and was accompanied by an increase in managed-vegetation patches. This indicates
that urban expansion and artificial greening developed simultaneously within the same spatial corridor.
The land-use transition results in Fig 16 and Table 3 show that land-cover restructuring was most intensive from 2010
to 2015. During this period, 2,925.42 km² of managed vegetation remained unchanged, while 1,360.08 km² was classified
as changing to water bodies, 575.54 km² to impervious surfaces, and 402.19 km² to wetlands. In addition, 591.57 km² of
wetlands was classified as changing to water bodies. From 2015 to 2020, diagonal values became more dominant. A total
Fig 12. Spatial evolution of NDVI, FVC, and LAI (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g012

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 24 / 44
of 2,875.61 km² of managed vegetation and 614.40 km² of impervious surfaces remained unchanged. The main off-
diagonal transitions included managed vegetation to impervious surfaces, wetlands to water bodies, and managed vege-
tation to wetlands, with corresponding areas of 47.92, 43.33, and 37.25 km², respectively.
As shown in Table 3, the land-use transition matrix for Dubai during 2010–2015 reveals pronounced expansion of
impervious surfaces and a substantial increase in managed vegetation, alongside notable mutual conversions between
water bodies and wetlands, collectively indicating a highly active phase of urban land restructuring during this period.
Fig 13. Spatial distribution of changes in NDVI, FVC, and LAI, 2010–2020 (Boundary data: geoBoundaries, https://www.geoboundaries.org/,
CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g013
Table 2. Mean annual rates of change in NDVI, FVC, and LAI for individual periods.
Period NDVI Dynamic Degree (%) FVC Dynamic Degree (%) LAI Dynamic Degree (%)
2010–2015 0.62 −0.15 −1.35
2015–2020 0.51 −0.79 1.36
2010–2020 1.15 −0.93 −0.08
https://doi.org/10.1371/journal.pone.0360918.t002

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 25 / 44
Specifically, the largest transitions into impervious surfaces came from managed vegetation (281.20), wetland (132.83),
and water (26.88), while managed vegetation received considerable inputs from impervious surfaces (209.17), wetland
(188.96), and water (50.80). Water and wetland also exchanged extensively, with 110.54 from wetland to water and 2.46
from water to wetland. In contrast, the transition magnitudes during 2015–2020 are markedly reduced, with all land-use
categories exhibiting greater quantitative stability—particularly the sharply diminished influx into impervious and managed-
vegetation classes. For example, the inflow to impervious surfaces from managed vegetation dropped to 24.74, and the
inflow to managed vegetation from impervious surfaces fell to 47.92. The largest remaining transitions were internal reten-
tions, such as 614.40 within impervious surfaces and 2,875.61 within managed vegetation, suggesting a deceleration of
urban expansion and a progressive shift toward a relatively stabilised land-use configuration. Over the entire 2010–2020
period, the matrix further highlights a dominant transition from wetland to impervious surfaces (541.63) and from wetland
to managed vegetation (84.70), along with a notable conversion from impervious surfaces to wetland (355.32), indicating
complex long-term interactions among coastal, vegetated, and built-up land covers. Spatially, the changes remain con-
centrated in the northern coastal development belt, waterfront areas, and inland expansion fringes. Although the transition
matrix alone cannot definitively distinguish actual land conversion from classification uncertainties, boundary shifts, or
variations in image acquisition conditions, its temporal–spatial trajectories are highly consistent with the evolving trends
of landscape-metric indices. This consistency further corroborates that the primary transformation processes during the
study period involved coastal urban expansion, managed-vegetation development, and internal reorganisation of water-
front landscapes.
Fig 14. Grouped bar chart of the mean annual dynamic degrees of NDVI, FVC, and LAI, 2010–2020.
https://doi.org/10.1371/journal.pone.0360918.g014
Fig 15. Land-use distribution in Dubai (Boundary data: geoBoundaries, https://www.geoboundaries.org/, CC BY 4.0.).
https://doi.org/10.1371/journal.pone.0360918.g015

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 26 / 44
As shown in Table 4, from 2010 to 2020, the proportion of bare land decreased from 74.51% to 71.63%, and its edge
density declined from 37.90 to 34.74 m/ha. However, patch density increased from 3.55 to 3.86 patches/100 ha, indicat-
ing that the reduction in total bare-land area was accompanied by increased fragmentation. The proportion of impervious
surfaces increased from 15.27% to 16.99%, while edge density and patch density decreased by 0.86 m/ha and 0.43
patches/100 ha, respectively, reflecting the simultaneous expansion and spatial consolidation of built-up land. Managed
vegetation exhibited the most pronounced changes, with edge density, patch density, and percentage of landscape
increasing by 6.81 m/ha, 1.25 patches/100 ha, and 1.56 percentage points, respectively. This indicates that the expan-
sion of urban greening was accompanied by substantial patch fragmentation. Sparse vegetation increased only slightly,
whereas all three metrics for water bodies and wetlands generally declined, suggesting a contraction in the extent and
boundary complexity of waterfront landscapes.
As shown in Fig 17, the different land-use types exhibited marked differences across the three landscape metrics. Bare
land and impervious surfaces consistently had higher edge density than the other land-use types, although both declined
gradually over time. In contrast, the edge density of managed vegetation increased continuously, reaching 21.18 m/ha in
2020, indicating increasingly complex boundaries of artificially established green spaces. In terms of patch density, man-
aged vegetation showed the largest increase, while bare land and sparse vegetation also increased to varying degrees;
impervious surfaces, however, exhibited a gradual decline. Bare land remained the dominant land-use type in terms of
Fig 16. Land-use transition diagrams for Dubai, 2010–2015 and 2015–2020.
https://doi.org/10.1371/journal.pone.0360918.g016

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 27 / 44
percentage of landscape, although its share continued to decrease. Meanwhile, impervious surfaces and managed vege-
tation expanded steadily, indicating that land-use change in Dubai was primarily characterised by the gradual conversion
of desert bare land into built-up areas and managed green spaces.
Table 3. Land-use transition matrix.
Year Type Impervious Managed Veg Sparse Veg Bare Land Water Wetland
2010-2015 Impervious 256.34 281.2 2.47 0.07 0.6 0.95
Managed Veg 209.17 2418.39 0.85 0.09 8.15 2.39
Sparse Veg 19.39 17.24 23.04 3.38 0.75 0.04
Bare Land 17.23 15.44 3.7 11.25 0.13 0.07
Water 26.88 50.8 0.57 0.7 103.95 2.46
Wetland 132.83 188.96 3.86 0.82 110.54 201.19
2015-2020 Impervious 614.4 24.74 0.64 0.92 7.76 13.38
Managed Veg 47.92 2875.61 0.62 1.14 9.49 37.25
Sparse Veg 3.41 1.15 26.75 0.99 0.97 1.22
Bare Land 1.76 1.19 2.53 9.33 1 0.5
Water 10.56 8.15 0.33 0.86 197 7.22
Wetland 14.48 10.05 0.23 0.34 43.33 138.67
2010-2020 Impervious 182 0.82 0.76 36.11 15.83 541.63
Managed Veg 4.03 2619.26 0.12 0.26 5.21 10.16
Sparse Veg 7.34 19 25.75 3.4 7.25 1.1
Bare Land 18.72 9.94 3.7 8.42 5.93 1.11
Water 1.01 5.99 0.14 0.27 176.27 1.68
Wetland 355.32 84.7 0.57 0.47 28.78 168.36
https://doi.org/10.1371/journal.pone.0360918.t003
Table 4. Trends in ED, PD, and PLAND for different land-use types.
ClassName Metric_Short 2010 2015 2020 Total_Change
Bare Land ED (m/ha) 37.9 35.35 34.74 −3.16
Bare Land PD (patches/100 ha) 3.55 3.74 3.86 0.31
Bare Land PLAND (%) 74.51 72.89 71.63 −2.88
Impervious ED (m/ha) 34.52 34.27 33.66 −0.86
Impervious PD (patches/100 ha) 3.61 3.29 3.18 −0.43
Impervious PLAND (%) 15.27 16.24 16.99 1.72
Managed Veg ED (m/ha) 14.37 18.44 21.18 6.81
Managed Veg PD (patches/100 ha) 2.85 3.74 4.1 1.25
Managed Veg PLAND (%) 3.3 4.12 4.86 1.56
Sparse Veg ED (m/ha) 14.26 15.65 15.58 1.31
Sparse Veg PD (patches/100 ha) 3.45 3.85 3.91 0.45
Sparse Veg PLAND (%) 5.24 5.49 5.4 0.16
Water ED (m/ha) 2.38 1.39 1.03 −1.35
Water PD (patches/100 ha) 0.55 0.54 0.5 −0.05
Water PLAND (%) 1.31 0.86 0.77 −0.54
Wetland ED (m/ha) 1.84 1.79 1.73 −0.12
Wetland PD (patches/100 ha) 0.67 0.59 0.64 −0.03
Wetland PLAND (%) 0.37 0.41 0.34 −0.03
https://doi.org/10.1371/journal.pone.0360918.t004

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 28 / 44
5.3. Correlation analysis
The correlation matrices in Fig 18 reveal strong relationships among several vegetation and landscape metrics. NDVI and
FVC were strongly and positively correlated in all three years, with coefficients of 0.93, 0.89, and 0.89 in 2010, 2015, and
2020, respectively. This consistency indicates that the two indicators captured a common dimension of vegetation con-
dition. However, their different temporal trajectories demonstrate that a strong spatial correlation does not mean that the
two variables respond identically to land-use change. NDVI was more closely associated with greenness, whereas FVC
represented the proportion and continuity of vegetated surface.
ED was also strongly related to PLAND, with correlation coefficients ranging from 0.77 to 0.79, and its correlation with
PD consistently exceeded 0.70. These relationships indicate that landscape area, boundary complexity, and patch subdi-
vision were closely connected. Larger or more widely distributed patches can produce greater total boundary length, while
Fig 17. Temporal trends in ED, PD, and PLAND for different land-use types.
https://doi.org/10.1371/journal.pone.0360918.g017

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 29 / 44
fragmented patches can simultaneously increase edge density and patch density. The high correlations among these met-
rics indicate that landscape structure should be interpreted as a set of interrelated properties rather than as a collection of
completely independent indicators.
The relationship between the air–water coupling coordination degree and the explanatory variables varied among
years. In 2010, coupling coordination was negatively correlated with FVC and NDVI, with coefficients of −0.42 and
−0.35, respectively. It was weakly positively correlated with ED, PD, and PLAND. In 2015, the correlations between
coupling coordination and most indicators weakened substantially. By 2020, the negative correlations with FVC and NDVI
re-emerged. LAI remained weakly correlated with the coupling coordination degree and with the other indicators in all
three years.
The negative correlations between vegetation indices and coupling coordination should not be interpreted as evidence
that vegetation directly deteriorated atmospheric or water conditions. A more plausible interpretation is that vegetation
was concentrated in developed or intensively managed areas where lower visibility, high irrigation demand, and spatially
constrained water resources could coexist. Because the coordination degree integrates atmospheric clarity and soil mois-
ture, the spatial presence of vegetation may be associated with a lower coordination value when its water requirement is
not matched by the environmental conditions of the surrounding area. The weak association of LAI with the coordination
degree further suggests that canopy leaf area alone did not adequately describe the spatial coupling between atmospheric
conditions and water availability.
5.4. Relationships between landscape-pattern metrics and coupling coordination identified using the optimal-
parameter geographical detector
The factor-detector results in Fig 19 indicate that the dominant explanatory factors changed substantially over time.
In 2010, FVC had the highest explanatory power, with a q-value of 0.90, followed by NDVI with a q-value of 0.60. ED,
PD, and PLAND had lower q-values of 0.42, 0.36, and 0.36, respectively. This result indicates that spatial differences in
coupling coordination during the initial study period were primarily associated with vegetation-cover conditions. In 2015,
ED became the dominant factor, with a q-value of 0.90. The explanatory power of PD and PLAND also increased to 0.73
and 0.60, respectively. The shift from FVC to ED suggests that landscape boundaries and patch structure became par-
ticularly important during the period of intensive land-use restructuring and urban expansion. Under these conditions, the
spatial arrangement of green and non-green patches may have described environmental differences more effectively than
vegetation amount alone. In 2020, FVC again became the strongest factor, with a q-value of 0.90, followed by NDVI at
Fig 18. Correlation analysis of the study variables.
https://doi.org/10.1371/journal.pone.0360918.g018

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 30 / 44
0.59. The explanatory power of ED, PD, and PLAND declined substantially compared with 2015. LAI consistently had the
lowest explanatory power, with a q-value of approximately 0.10 in all three years. The low and stable contribution of LAI
indicates that leaf-area structure was spatially concentrated and therefore had limited capacity to explain the broad spatial
differentiation of coupling coordination. Overall, the factor-detector results demonstrate that the dominant spatial correlate
changed according to the stage of urban and landscape development. Vegetation coverage was more important in 2010
and 2020, whereas landscape-edge characteristics became particularly important in 2015.
Fig 20 further shows that the explanatory power of factor combinations was consistently greater than that of individual
factors. In 2010, the strongest interactions were centred on FVC. The interaction between NDVI and FVC produced a
q-value of 0.98, while LAI–FVC and PLAND–FVC produced values of 0.94 and 0.88, respectively. These results indicate
that vegetation amount, greenness, canopy structure, and landscape proportion jointly described the spatial state of cou-
pling coordination more effectively than any single indicator. In 2015, ED became the central interacting factor. The inter-
action between FVC and ED reached a q-value of 1.00, while NDVI–ED, PD–ED, and PLAND–ED reached 0.90, 0.85,
and 0.80, respectively. This result indicates that the spatial influence of vegetation coverage was closely conditioned by
the configuration of patch boundaries during the period of rapid landscape restructuring. In 2020, FVC again acted as the
principal interaction factor. Its interactions with NDVI, PD, ED, and LAI produced q-values of 0.99, 0.98, 0.95, and 0.94,
respectively. The interaction results demonstrate that coupling coordination was associated with the combined configu-
ration of vegetation quantity and landscape structure. However, the q-value represents spatial explanatory power rather
than the direction or magnitude of an effect. A high interaction value therefore indicates strong spatial stratification or joint
Fig 19. Results of factor detection.
https://doi.org/10.1371/journal.pone.0360918.g019
Fig 20. Results of interaction detection.
https://doi.org/10.1371/journal.pone.0360918.g020

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 31 / 44
explanatory capacity, but it does not demonstrate that the two factors exerted additive or causal effects. The repeated
prominence of FVC–ED interactions suggests that the environmental meaning of vegetation coverage depended strongly
on the complexity and fragmentation of its spatial boundary.
5.5. Relationships between landscape-pattern metrics and coupling coordination identified using spatial
regression
5.5.1. Ordinary least squares regression. The OLS results in Table 5 indicate that FVC was significantly and
negatively associated with the air–water coupling coordination degree in all three years. The coefficients were −0.3117 in
2010, −0.0236 in 2015, and −0.3589 in 2020, with all relationships significant at p < 0.001. The magnitude of the coefficient
was substantially weaker in 2015, indicating that the global association between vegetation coverage and coupling
coordination became less pronounced during the period when landscape structure and urban expansion were changing
most rapidly.
ED had a significant positive association with coupling coordination in all three years. Its coefficient declined from
0.00021 in 2010 to 0.00018 in 2015 and 0.00007 in 2020. Although the positive direction remained stable, the declining
coefficient indicates that the average association between boundary complexity and coupling coordination weakened over
time. This change is consistent with the increasing complexity of managed vegetation and the changing spatial role of
urban development boundaries. ED may therefore have captured different environmental configurations in different years,
rather than representing a stable landscape function.
PD was negative and statistically insignificant in both 2010 and 2015, with coefficients of −0.00001 and −0.00002,
respectively. In 2020, its coefficient became positive and significant, reaching 0.00007 with p < 0.001. This temporal
shift suggests that patch number became more relevant after the expansion and subdivision of managed vegetation
had become more pronounced. Nevertheless, the coefficient sign describes a statistical association with the composite
Table 5. Coefficients and significance tests for variables in the global OLS regression models.
Variable Statistic 2010 2015 2020
Constant Regression coefficient 0.8832 0.5716 0.8232
Standard error 0.0041 0.0045 0.0054
(t)-value 215.160 126.394 153.056
(p)-value <0.001*** <0.001*** <0.001***
FVC Regression coefficient −0.3117 −0.0236 −0.3589
Standard error 0.0052 0.0057 0.0072
(t)-value −60.327 −4.101 −50.074
(p)-value <0.001*** <0.001*** <0.001***
VIF 1.367 1.377 1.424
ED Regression coefficient 0.00021 0.00018 0.00007
Standard error 0.00001 0.00001 0.00001
(t)-value 39.366 33.377 11.457
(p)-value <0.001*** <0.001*** <0.001***
VIF 2.185 2.141 2.312
PD Regression coefficient −0.00001 −0.00002 0.00007
Standard error 0.00002 0.00002 0.00002
(t)-value −0.306 −0.860 4.576
(p)-value 0.760 0.390 <0.001***
VIF 2.001 1.929 2.074
https://doi.org/10.1371/journal.pone.0360918.t005

PLOS One | https://doi.org/10.1371/journal.pone.0360918 October 23, 2026 32 / 44
coordination index and should not be interpreted as evidence that increasing patch number directly improved environmen-
tal conditions.
The variance inflation factors of all three retained predictors were below 2.5 in each year, indicating that severe multi-
collinearity was not present in the OLS models. This result is important because the correlation analysis showed strong
relationships among several vegetation and landscape indicators. The retained variables therefore provided sufficiently
distinct information for the estimated global associations, although the high correlations among the broader indicator set
still require caution when interpreting individual coefficients.
The spatial distribution of standardised OLS residuals in Fig 21 reveals pronounced spatial clustering and clear inter-
annual variation. In 2010, positive residuals were concentrated along the western and southwestern coast and in several
northeastern areas, whereas extensive negative residuals occurred in the central-eastern and southeastern regions. In
2015, positive residuals dominated the northern coastal and central areas, while negative residuals were concentrated
in the southern inland region and along several coastal margins. In 2020, a large negative-residual core emerged in the
central urban area, while positive residuals were mainly distributed along the western coast, southern fringe, and several
northern locations.
These residual patterns indicate that a single set of global coefficients could not adequately represent the differences
between intensively urbanised coastal areas, inland desert regions, and local development nodes. The residual clusters
may reflect spatially structured conditions that were not fully captured by FVC, ED, and PD, including differences in land-
use context, irrigation practices, microclimate, topography, atmospheric exposure, and urban morphology. The residual
pattern therefore provides empirical support for applying a local spatial model. It does not, however, identify which unmea-
sured factor generated the remaining spatial dependence.
The OLS goodness-of-fit statistics in Table 6 further demonstrate the limitations of the global model. The sample size
remained stable at approximately 15,700 observations in each year, with 15,695 observations in 2010, 15,696 in 2015,
and 15,693 in 2020. The model achieved its highest explanatory power in 2010, with an R² of 0.3337 and an adjusted
R² of 0.3336. The R² declined to 0.1020 in 2015 and partially recovered to 0.1681 in 2020. The adjusted R² values were
nearly identical to the corresponding R² values, indicating that the difference was not caused by excessive model com-
plexity or artificial inflation associated with the large sample size.
The residual sum of squares was 118.87 in 2010, 142.26 in 2015, and 247.52 in 2020. The AICc values were
−32,091.46