Assessing Fine-Scale Urban Green and Blue Infrastructure Change in Manchester, UK: A Spatiotemporal Analysis Framework to Support Environmental Land Use Management

Abstract

Understanding changes in urban green and blue infrastructure (UGBI) associated with land use management can inform planners on trends in environmental change that may impact urban resilience. While UGBI change resulting from land use conversion has received significant research interest, UGBI change within otherwise consistent land uses has received scant attention. This study developed a high-resolution spatiotemporal analysis framework to map fine-scale UGBI change across all land use classes in Manchester, UK, over a period (2000–2017) of significant population growth. The study found that UGBI declined in 17 out of 29 land use classes, with an overall city-wide UGBI loss of 11.9%, compared to UGBI gains for 6.4% of the city. Declines were most concerning in residential areas, which cover 33.6% of Manchester, as UGBI in these areas is important for delivering ecosystem services to citizens. Extrapolation of change rates indicate that two-thirds of future UGBI loss could occur in residential areas. These results provide insights into socio-economic processes which are likely to have similar implications for UGBI trends in other urban areas. Such knowledge is critical to inform land use planning and management to identify where UGBI is at risk and implement appropriate policies to reverse or minimise losses.

1. Introduction

Environmental features such as woodlands, street trees, rivers, ponds, wetlands, parks, shrubs, and hedges serve as a vital network of natural and semi-natural spaces, providing a wide range of recreational, cultural, and provisioning benefits to urban residents. As climate change is projected to exacerbate extreme weather events in the coming decades, this network of environmental features, or urban green–blue infrastructure (UGBI), will serve as an increasingly important resource in bolstering urban climate resilience [1,2]. Vital functions, such as stormwater absorption in canopy leaves and soils, and temperature cooling through evapotranspiration in vegetation and waterbodies regulate environmental hazards such as surface flooding and the urban heat island effect [3]. The quantification of such benefits, as ecosystem services with value to people [2], enables stakeholders to consider and contrast the relative advantages of UGBI to grey infrastructure adaptations for resident well-being [4].

Whilst many cities have adopted extensive greening programs in recent years, numerous studies suggest an overall decline in the extent and quality of UGBI in many urban centres around the world [5,6]. UGBI degradation typically occurs from infill development, whereby existing UGBI resources are replaced with impervious surfaces, or through the expansion of built infrastructure into habitats on the urban periphery [7]. As the pressure for economic development and housing continues, towns and cities become increasingly built-up, resulting in a growing population with diminishing access to ecosystem services [8].

Countering such degradation is therefore a key concern amongst many urban planning stakeholders, that benefit from the knowledge on the magnitude of UGBI change in relation to the management decisions and drivers behind it [9,10]. However, this is often difficult to measure as urban development is heterogeneous, occurring over varying spatial and temporal scales and affected by the local planning, socio-economic, infrastructure, and environmental context [11]. Complexity in urban development is typically organised through the concept of land use systems, whereby geographic extents of land are categorised according to the associated human activities and supporting land covers [12]. As a planning tool, the application of land use systems can ensure that the distribution of human activities is adequate to support economic and environmental policy goals [13].

Change in UGBI resources is therefore often approximated in land use change information, through association of an assumed proportion, configuration, or amount of UGBI per land use category [14]. For example, loss of UGBI may be assumed when converting from recreation areas, typically associated with high UGBI cover, to more built-up industrial or residential land uses. Land use therefore provides a conceptual framework to quantify structural change in an urban area and model impacts upon ecosystem services (e.g., urban cooling) and access to nature [15]. Comparison between land use, or more accurately land use land cover (LULC) map products at different time points, also provides indication of localised UGBI change according to socio-economic development pressures [14]. For example, the conversion of parkland to commercial land use may be quantified across a city, thus informing stakeholders on the environmental consequences of this change in relation to economic benefits from converting a public liability to a source of tax revenue [16].

Whilst the concept of land use is useful to investigate the effects of land conversion, consideration of UGBI change within otherwise static land use areas is also vital to understand the impacts of longer-term land management [17]. For example, numerous studies highlight increasing tendencies to convert garden UGBI (e.g., lawns, planters) to land cover types that are easier to manage and are more appropriate to support other household functions (e.g., tarmac driveways, house extensions) [18]. Land use management, described within land ownership boundaries or land use parcels [19], is often difficult to monitor due to small scales and limited planning control. Therefore, the process of UGBI removal may not be noticed by planning authorities. As demonstrated by studies of private garden land cover change, UGBI loss aggregated across numerous individual and small parcels can produce a significant overall environmental impact, such as increasing stormwater runoff rates and flood risks for the wider neighbourhood [20]. Uncontrolled loss in UGBI resources may indirectly degrade the impact of any adaptation strategy, increasing vulnerability amongst urban residents as a result.

Given recognised monitoring difficulties, UGBI changes in consistent land use parcels have received limited research attention. Since rates of land cover change can vary substantially depending on land use management [21,22], this is an important issue. For example, a study of temporal UGBI change in Berlin over a 30-year period found that greening policies in brownfields and street sides resulted in UGBI enhancement, whereas residential and parking areas suffered UGBI declines [22]. Whilst UGBI change may vary across individual land parcels, overall trends per land use can help to develop a broader understanding of the impact of specific land use pressures upon future UGBI resources and associated ecosystem services [23]. Trends considered across a wide range of land use classes can in turn aid spatial prediction of future UGBI change and therefore indicate hotspots of future UGBI decline to support the development of local UGBI strategies [24,25].

With the expansion of high-resolution digital geo-spatial data in recent years, a wide range of GIS and image analysis methods have been developed to investigate land use and land cover change in urban areas [26]. The application of advanced machine learning methods with very high resolution (<2 m pixels) multi-spectral imagery has been successfully applied to map land cover types at an appropriate patch-level scale and to improve analyses such as urban habitat connectivity modelling [27], urban heat island monitoring [28], and fine-scale land cover change detection [29]. Increasing availability of accessible datasets from public bodies and national/international mapping agencies enables consideration of broad-scale land use change [30,31], whilst also supporting finer-scale remote sensing analysis, providing ancillary data to enhance patch-level object-based land cover change comparisons [32,33].

Whilst spatial urban change analysis is an important and developing research activity, most studies focus on land cover change only or consider the concept of land use within hybrid land use land cover (LULC) categorisation systems. This study therefore attempts to address a current research gap by presenting a framework to map changes in individual UGBI patches (e.g., tree removal, grass lawn to paving) to consistent land use parcels across a city. In line with the previous discussion, this study therefore aims to develop useful urban planning information by demonstrating application of the framework to:

i.

Quantify and visualise spatial dynamics (loss and gain) in UGBI parcels at high resolution across an entire urban area.

ii.

Calculate UGBI loss/gain within each land use and extrapolate future UGBI change trends to understand risks to future urban environmental conditions.

iii.

Link vulnerability in UGBI resources (where UGBI has suffered losses) to land management practices and wider socio-economic trends.

The approach is applied to a post-industrial city (Manchester, UK) during a period of historic urban renewal and associated population growth (2000–2017). Given the similarities in socio-economic circumstances and urban configuration of Manchester to other post-industrial cities in the UK and Europe, the approach here also aims to provide comparative evidence of pan-urban trends to contribute to a growing evidence base on UGBI losses resulting from urban development.

2. Materials and Methods

2.1. Study Area

Manchester (NW England, UK) has a population of 550,000 [34]. Patterns of land use and UGBI reflect its former industrial heritage and several phases of growth and decline since the late 1700s (Figure 1). While the city is served by a network of open (e.g., parks, nature reserves) and private (e.g., gardens) green/blue space areas, evidence suggests that UGBI cover across the city may be in decline [6]. The city population has grown rapidly since 2000 (when pop. was approximately 400,000), with many former industrial brownfield areas converted to high-density residential developments and infill development of existing UGBI patches (e.g., garden paving, building extensions). Manchester therefore provides a useful case study of UGBI change which will identify linkages between socio-economic development and land use management. Manchester City Council has recently revised its Green and Blue Infrastructure Strategy and is actively seeking to improve knowledge on UGBI to achieve climate resilience goals [35]. Furthermore, the time period of recent population growth coincides with the ready availability of very high-resolution spatial data making it suitable for GIS analysis.

2.2. Overview of Methods

The approach follows three key stages:

• Object-based image classification, and subsequent validation, to produce a very high-resolution map of UGBI change patches.
• Semi-automated land use mapping to identify topographic parcels and sub-parcel features that have remained consistent in land use over the study period.
• Integration of stages (1) and (2) to identify UGBI change trends across the city, and for individual land use types, using error-adjustment methods. Visualisation of predicted future change in UGBI across the city.

Key components of the approach are summarised in the following sub-sections, with detailed notes on method processes provided in the appendices.

2.3. Mapping UGBI Change Patches

Image classification was undertaken to categorise UGBI change patches at high resolution to assess the impact of small-scale land use management processes (i.e., garden paving, re-greening of derelict buildings). For analysis purposes UGBI was defined as all identifiable vegetation features (e.g., tree canopies, grassland, planted shrubs/crops, natural shrubs) and all identifiable water features (including rivers, canals, ponds) in the imagery, irrespective of the associated land use. In this definition all sources of vegetation (green) and water (blue) serve as infrastructure, as opposed solely to explicitly planned natural resources, to provide ecosystem service benefits [37,38]. The removal of any patch of vegetation or water may have a detrimental effect on ecosystem services and local climate resilience, and therefore the same patch or water retained will continue to provide benefits as infrastructure.

Cloud-free very high-resolution (≤2 m pixel size) imagery was obtained for the years 2000 and 2017. For the year 2000, a true colour three-band (RGB) aerial image composite (0.25 m pixel size; acquired in the month of June) was purchased from commercial vendors [39]. Multi-spectral (RGB and near infrared) images were acquired from the Spot-7 (1.5 m pixel size) and Pleiades-1A (0.5 m pixel size) satellite sensors [36], respectively, for May and October 2017. Multi-date image composites are advantageous since they provide additional temporal difference information enabling enhanced vegetation classification [40]. High-resolution multi-spectral imagery was not available for the year 2000, therefore the true colour imagery was identified as the most suitable data source.

Object-based post-classification change detection was used to map UGBI change. This method compares outputs of independent classifications and is suitable for information derived from different sensors [41]. A limitation of this approach is that errors in either input classification dataset will compound within the final change detection layer [42]. A simple two-class (UGBI and non-UGBI) scheme was therefore used to generate an overall four-class (UGBI stasis, UGBI gain, UGBI loss, non-UGBI stasis) change detection map. This approach constricted error that can arise from the multiplication of unique change detection instances [43]. A rigorous manual geo-rectification process was applied to implement image co-registration to the recommended level of accuracy to minimise error from spatial mis-registration. Images were respectively cross-examined to concurrent UK Ordnance Survey MasterMap^TM topographic features [44] to ensure registration of both images to a consistent spatial model. The year 2000 imagery was then downscaled to match the resolution and alignment of year 2017 image pixels to enable consistent comparison of classification outputs.

A general framework was applied to ensure similar outputs for the 2000 and 2017 land cover classification exercises. Appendix A provides a detailed overview of methods for each image date. Overall accuracy was >94% for each classified map, which enabled intersection to generate the four change detection classes (Figure 2). A separate test sample dataset was used to assess the accuracy of this output and generate final rules to clean spurious change detection patches [45], which resulted in the removal of 4.6% of the total change detection area. The final error matrix (see Appendix A) informed statistical error adjustment of change detection rates in the final stage.

2.4. Consistent Urban Land Use (ULU) Features

For the 2000–2017 study period there are no existing map products that offer consistent fine-scale land use information. Whilst the Urban Atlas [46] provides an invaluable product to analyse urban land use, the earliest version of this product is for the year 2006. Other available land use products typically amalgamate different urban land uses into a small number of categories and are more suitable for regional-scale analyses. A bespoke land use product was therefore defined, using the UK National Land Use Database (NLUD v2006; [47]) as a framework to categorise land use information from selected map layers from the UK Ordnance Survey [44]. The NLUD infers urban land use types that are recognised internationally. The urban land use (ULU) hierarchy defined in this study was based upon NLUD categories that could be feasibly identified through processing of Ordnance survey layers (see

Table 1. Description of urban land use (ULU) Group and Class categories with mapped extent as a percentage of the study area.

ULU Group/Class	Description	% of Study Area
1. Brownfield	Land without current purpose	1.72
1.1 Brownfield	Brownfield areas, developmental land, and construction sites	1.72
2. Commercial	Areas primarily providing commercial and retail services	4.75
2.1 Commercial	Retail and professional services	4.75
3. Community Services	Government and public community welfare services	6.62
3.1 Safety and well-being	Public safety (e.g., police, fire, social support)	0.34
3.2 Cultural facilities	Services supporting cultural recreation	0.17
3.3 Health care	Health care services	1.00
3.4 Higher education	Non-compulsory adult education services	1.14
3.5 Religious facilities	Religious worship in any denomination	0.53
3.6 Schools	Compulsory non-adult education services	3.44
4. Industrial	Manufacturing, engineering, construction, and energy distribution services	4.84
4.1 Industrial	Manufacturing, warehousing, and distribution sites	4.51
4.2 Energy utilities	Generation and distribution of energy supplies	0.33
5. Non-recreational Open Space	Predominantly open-space not supporting recreation	7.72
5.1 Agriculture	Commercial farming	3.55
5.2 Cemeteries	Processing and storage of human remains	0.99
5.3 Water	Natural and purpose built water bodies and channels	1.22
5.4 Woodland	Continuous tree cover separate to other land uses	1.96
6. Public Recreation	Outdoor and indoor facilities supporting physical/social recreation	17.92
6.1 Public open space	General outdoor amenities and open spaces	11.10
6.2 Sports facilities	Land and facilities designated for sporting activities	6.31
6.3 Urban farming	Non-commercial urban farming	0.51
7. Residential	Primarily residential housing of varying dwelling density	33.30
7.1 Low-density residential	Majority of dwellings are semi-detached and detached housing	22.45
7.2 Medium-density residential	Majority of dwellings are terraced housing	7.61
7.3 High-density residential	Majority of dwellings are former buildings converted into flats or purpose-built multi-dwelling apartment housing	3.24
8. Transport	Infrastructure supporting the transport of people and goods	23.13
8.1 Car parking	Car parking areas not associated with other land uses	0.55
8.2 Limited access roads	Private roads connecting addresses to higher functioning roads	0.48
8.3 Linking roads	B roads connecting significant destinations and feeding A roads	0.35
8.4 Major roads	A roads and dual carriageways	1.50
8.5 Minor roads	Roads connecting addresses to higher functioning roads	7.11
8.6 Motorways	Motorway roads—as defined in the OS highways dataset	0.69
8.7 Railways	Land and infrastructure supporting rail and tram travel	1.94
8.8 Roadsides	Access routes between areas for non-vehicular travel	6.56
8.9 Transport terminals	Non-rail mass transit travel, e.g., bus and tram stations, airports	3.95

).

Mapping of study area ULU group and class categories for the year 2017 followed a semi-automated mapping approach, involving data integration, model prediction, and manual digitisation. The whole process is detailed in Appendix B and produced a vector ULU map product with a minimum mapping unit of <50 m² (Figure 3) and overall estimated thematic accuracy of 97%. The 2017 ULU map provided a reference model to back-date consistent land use features to the year 2000.

ULU mapping at the same resolution for the year 2000 was not possible due to limitations in OS data for this period. The legacy land-line dataset for the year 2000 [48], however, maps topographic features to the same spatial model as the 2017 OS data. Polygon-to-polygon comparison between OS datasets over time enables the identification of ULU sub-parcel features that remain consistent in shape and spatial position [49], indicating consistency in underlying land use type. An automated polygon comparison algorithm was developed in R [50] to identify consistent ULU sample features (see Appendix C). The process was successful in identifying 60.2% of candidate 2017 features as samples for further analysis, which compares favourably to estimates of 76.7–85.2% for actual consistent 2017 ULU areas over the study period. UGBI change classes were clipped to each feature to aggregate UGBI change rates for ULU classes and groups.

2.5. Mapping UGBI Change

UGBI change between 2000 and 2017 was analysed for: (a) city extent, (b) individual ULU classes and class groups, (c) 100m grid cells to visualise neighbourhood trends in UGBI change (see earlier study [51] for demonstration of this grid-based approach). To account for classification error in the UGBI change detection layer, the error adjustment method described by [52] was used to estimate net UGBI change for each analysis (see Appendix D). Total error net change rates with upper (upper UGBI gain—lower UGBI loss) and lower (lower UGBI gain—upper UGBI loss) bounds of change confidence levels were calculated according to total change class composition for each analysis area. UGBI stasis was therefore determined for the respective analysis component where upper net UGBI change ≥ 0 ≥ lower net UGBI change. Net area change estimates were used to back-date estimates of UGBI levels for the year 2000. Non-stasis UGBI change trends per ULU class were used to linearly extrapolate UGBI levels approximately 17 years into the future (i.e., 2034) by applying change rates to UGBI proportions recorded for current 2017 ULU class parcels.

3. Results

3.1. Overall UGBI Change

Total UGBI cover for the study area in 2000 was estimated at 50.2% (±2.6%; 95% CI), in comparison to 44.7% in 2017. This change converts to 5.5% net UGBI loss (±2.6%; 95% CI) of the total study area, or 10.9% net UGBI loss (low estimate = 6.0%, high estimate = 15.3%; 95% CI) as a percentage of the estimated UGBI in 2000. Approximate UGBI cover per resident in 2000 was 128.1 m² compared to 99.8 m² in 2017—a 22.1% reduction in existing UGBI per resident. However, despite the overall trend of UGBI loss, UGBI change varies across the study area. For example, 6.4% (±1.4%; 95% CI) of the study area recorded UGBI gain, in comparison to UGBI loss for 11.9% (±1.2%; 95% CI).

At the analysis cell level, net gains are recorded for 25.7% of cells in comparison to net losses recorded for 55% of cells. Figure 4 visualises this dynamism in UGBI gain and loss: existing built infrastructure has been removed and replaced with UGBI (gain), however, car-parking facilities now replace UGBI (loss). Overall, 42.7% of analysis cells showed relatively minor UGBI change (±5%; 95% CI), whilst the maximum recorded UGBI change was 77.9% and 92.2% for gain and loss cells, respectively (Figure 5). Patterns in analysis cell UGBI change exhibit a high degree of spatial autocorrelation as evidenced by the Moran’s I test (I = 0.55, p < 0.001).

3.2. Urban Land Use

Overall trends for ULU classes are largely negative; 17 out of 29 ULU classes experience statistically significant loss of UGBI (Figure 6). In comparison, Railways is the only class with net gain in UGBI, with stasis recorded for all other classes. Rates of UGBI change (as a percentage of estimated year 2000 sample UGBI cover) vary considerably between classes (Figure 7). Large losses in UGBI are apparent for Car parking (74.2%) and Major roads (41.7%). Declines in UGBI in Low-, Medium-, and High-density residential classes are 11.9%, 28.3%, and 5.7%, respectively. This indicates considerable loss in UGBI from 2000–2017 for residential areas (e.g., gardens), particularly for Medium-density residential areas characterised by terraced housing.

Change in UGBI is dynamic across ULU classes, with areas of loss and gain recorded for all classes (Figure 8). Overall UGBI change trends are determined by the balance between loss and gain UGBI cover, referred to here as the loss area dominance. Whilst the Woodland ULU class exhibits a lower loss area dominance than Railways, the overall UGBI gain trend is not significant, as the confidence intervals are neither wholly positive nor negative (Figure 5). ULU classes which have experienced overall losses in UGBI exhibit large variation in UGBI gain areas (as a percentage of total sample area) between 1.9% (Car parking) and 10.4% (Urban farming). ULU classes with stasis in UGBI exhibit ranges of 2.5–14.6% and 2.1–11.2% for gain and loss area coverage, respectively. The degree of overall UGBI change rates’ aggregate dynamism between gains and losses in UGBI thus varies between ULU classes.

Comparing overall UGBI change rates reveals similar values for a number of classes, therefore, in terms of explaining varying rates of UGBI change, some redundancy in class categorisation may be apparent. To statistically test whether differences in the distribution of estimated UGBI change rates exist, distributions of class UGBI change rates were created from exclusive random sample sub-sets. As ULU no-change sample areas vary considerably in size, thus having variable influence upon overall estimates of class UGBI change, equal size pixel groupings representing the ULU minimum mapping unit area (45 m² = 20 pixels) were used as analysis units. The number of groups selected per class (n = 219) was determined from the number of units contained within the smallest ULU class sample pixel area (Linking roads; n = 4402 pixels/20 ≈ 219 sub-sets). UGBI change (as a percentage of existing UGBI) was then calculated for each sub-set.

As the Kruskal–Wallis test (χ² = 2492, p < 0.001) provided strong evidence of inter-class differences in the distribution of UGBI change rates, a pairwise Wilcoxon–Mann–Whitney U test with Bonferroni correction (Base package, R Statistical Programming language; [50]) was used to test for differences between ULU classes. In all, 321 out of a total of 406 (79.1%) of class pairings displayed significant differences in estimated UGBI change rates, with the majority of non-significant differences (70 out of 85) recorded between classes of different ULU groups and the remaining non-significant differences among similar land uses within ULU groups (Appendix E). Insignificant pairings recorded for Community services (n = 5), Non-recreational open space (n = 2), and Transport (n = 8) ULU groups evidence similar development patterns and redundancy between some sub-ULU group classes. In contrast, Commercial and Industrial ULU classes represent similar land uses for private enterprise but exhibit significant differences in UGBI changes rates. Significant differences between the majority of ULU sub-group classes indicate that the current ULU class categorisation scheme provides an approximation of varying UGBI change rates within most ULU groups.

3.3. Extrapolation of Change Rates

Linear extrapolation of change rates to levels of UGBI in 2017 provides a simple method to indicate future levels of UGBI across the study area. Following this method, the majority of loss in UGBI is expected to occur within the Roadsides, Medium-density residential, and Low-density residential classes (Figure 9). Whilst rates of net UGBI change for the Low-density residential class are relatively low (−11.9%), this class contains over 20% of all 2017 UGBI. As such, assuming land use areas remain relatively static approximately 17 years into the future, current trends indicate 45% of total UGBI loss will occur within this class (Figure 9). In contrast, expected losses for Roadsides and Medium-density residential are also relatively high, with 13.5% and 19.8% total losses, respectively. All other classes record below 4% in total share of predicted loss. Relatively higher rates of UGBI loss for Car parking and the Major roads class have lower implications for future UGBI levels due to respective study area coverage of just 0.6% and 1.5% for these classes.

The implications may be examined on a spatial level at both the analysis cell and administrative ward level (Figure 10). As evidenced, high estimates of potential UGBI loss are prevalent within sub-urban residential areas south of the city centre, which contain large areas of Low-density residential areas. In contrast, when considering the study area as a whole, future UGBI cover estimates depend upon the calculation method used. When considering statistically significant UGBI change for ULU classes, UGBI cover in the 2030s is estimated to decrease by 3.1% (±1.0%; 95% C.I.). In comparison, using study area baseline change estimates for all UGBI resources, future UGBI cover is expected to decrease by a total of 4.9% (+1.9%/−2.2%; 95% C.I.). The difference between the two central estimates (1.8%) provides a basic indication of the level of UGBI change due to land use conversion. Raw neighbourhood level estimates for UGBI decline do not consider UGBI change from this process, nor additional influencing factors, and therefore currently provide an indication only of the magnitude of future UGBI decline.

4. Discussion

4.1. UGBI Change Trends

An approximate 3% decline in greenspace was measured in the city of Manchester’s urban core between 1991 and 2006 [6]. This earlier study was not representative of the city as a whole but indicates that UGBI degradation recorded here is part of an ongoing process of decline. Efforts to re-build Manchester’s post-industrial economy began in the 1980s with regeneration of the city centre and have continued apace with substantial re-development since the year 2000 [53]. Notable developments between 2000 and 2017, such as Manchester Sportcity [54] and the New Islington district close to central Manchester [55], are representative of overall economic and population growth during this period. UGBI degradation has occurred due to densification in built infrastructure [7], through land use conversion and infill development. The analysis cell (100 m) maps developed in this study therefore provide a beneficial visual guide at the neighbourhood scale where this change has occurred in the past and therefore may impact residents further in the future.

For the ULU Transport group, losses or gains in vegetation canopy will impact both Roadsides (where vegetation is likely to be planted) and adjoining road areas (where tree canopy may overhang). UGBI losses in Transport road classes are not unexpected given that countrywide urban street tree losses have been documented in the UK national press [56]. However, to the best of the authors’ knowledge, this process has not been previously quantified. Rates of UGBI decline are high for Linking, Minor, and Major roads but do not represent a significant loss of UGBI when considering total UGBI coverage within the study area. If the rates of UGBI losses in Roadsides are to remain consistent over time, then this presents a concern given that such resources are accessible to pedestrians and provide additional ecosystem services, including particulate capture and noise buffering [57].

UGBI decline recorded for Residential ULU group classes coincide with temporal declines in garden green infrastructure and pervious surface area identified in other studies [18,58,59]. For Medium-density residential areas, estimated year 2000 UGBI levels (27.3%) were already relatively low in comparison to Low-density (46.3%) and High-density (37.7%) residential classes; existing low levels of ecosystem service provision may therefore have degraded significantly further over the study period in these ULU class areas [58,60]. UGBI losses in low-density housing are less severe, but if the change rates recorded here remain consistent over time, they represent a serious concern for future ecosystem service management, given that this class covers over a fifth of the study area. Evidence from other studies indicates that population growth may influence conversion of single dwellings into multi-occupancy units, where garden paving occurs to provide car parking space for tenants [58,61]. In addition, population pressures on households may also influence decisions to extend existing housing units or sub-divide existing garden areas for new housing, in turn pressurising existing UGBI in residential areas [62,63].

The socio-economic status of home-owners may also influence management of private UGBI. For example, refs. [21,64] found differing levels of UGBI decline associated with the general affluence of districts, with higher levels of disposable income to invest in garden development. Certainly, changing lifestyles may also explain an apparent rise in “plastic lawns” in recent years, which simulate the appearance and texture of grass and are easier to manage but as a manmade surface may increase surface runoff and add unwanted plastic debris to the general environment [65]. In comparison, UGBI change rates for the High-density residential class (e.g., apartment blocks or flats) are significantly lower than those of other residential classes, as these areas are not subject to the “tyranny of small decisions” from private garden owners [66,67]. General declines in residential classes mirror UGBI declines in public (Community Services: Schools, Further education, Health care, Safety and well-being) and commercial sector (Commercial and Industrial groups) management, indicating various development pressures to do more with less land [68,69].

Whilst residential UGBI has declined, recent public surveys find that urban residents generally have positive attitudes towards urban green–blue space [70,71,72]. There is further scope in local and national policies to guide urban residents to adopt garden management practices that enhance nature conservation and climate resilience [20]. Greening efforts by environmental non-governmental organisations in Manchester [73,74] have undoubtedly contributed to the UGBI gains measured in this investigation, as evidenced by the expansion of tree canopies and tree lines (e.g., street trees, edge of woodland, or tree clusters) in many areas of the city. Indeed, a surprising finding from this investigation is the amount of dynamism between UGBI gain and loss within the study area. Net gains were recorded for Railways, whilst overall UGBI stasis was recorded for all ULU Non-recreational Open Space group classes, in addition to Public open space, and Sports facilities, indicating that consistent open space land uses over the study period witnessed limited grey development.

4.2. Limitations of the Framework and Future Research Directions

As indicated in similar remote sensing studies, change in urban landcover is typically non-linear over time [22]. Models predicting land cover change require consideration of a wide range of climatological, environmental, physical, socio-economic, demographic, and policy-level factors [30,75]. The simple extrapolation method used in this study provides a limited indication of UGBI cover into the future. To improve the analysis here, future research should take advantage of developments in accessible remote sensing imagery to repeat the process at various time points and examine the linearity of UGBI change under varying political and socio-economic conditions [76]. For example, this could be used in Manchester to examine the impact of various green strategies from the local government since 2015 [35,77] or even further into the future, consider the impacts on UGBI resources of recently adopted (February 2024) nationwide (in England) Biodiversity Net Gain legislation [78].

The information gained from this research may serve to validate the effectiveness of different UGBI protection policies and thus evidence whether direct intervention mechanisms, such as financial incentives for garden de-paving [79] or more restrictive urban planning regimes [80], could benefit conservation of UGBI and urban ecosystem services. In addition, as the GIS data used for urban land use mapping have remained consistent since 2017, the approach could be repeated to coincide with UGBI change detection exercises to consider interactions between the processes of land use and land cover development. UGBI change, resulting from conversions between distinct land use classes, could further inform research into cellular automata and machine learning approaches to improve prediction of urban environmental conditions from various development and policy scenarios [75].

Another limitation of the methodology is the use of the binary classification system that considers various vegetation and water features, with differing levels of ecosystem service benefits, as a uniform block of land cover. Whilst change detection in multiple land cover classes remains a challenge in remote sensing applications [29], future research should attempt to consider additional stratification of the UGBI class to examine changes in distinct UGBI components, such as trees, grasses, wetlands, and water, to measure change in both the quantity and quality of UGBI across the urban area [32,33]. The current framework could be used to achieve this, as incorporation of the error adjustment method enables consideration of higher rates of misclassification, which are likely to arise when incorporating additional classes into change detection processes [45]. Alternative classification approaches (e.g., direct change classification, neural networks) could also be explored as a method to ensure the retention, or even improvement of, accuracy levels in any change detection layer [29,81].

Information on the change in quality of UGBI, incorporated with improvements to the temporal resolution of land cover to land use change analysis, and improvements in change modelling, could enhance the level of projected information on future ecosystem services available to urban planning stakeholders. Given the risks posed by climate change, further research to improve methods and outcomes for UGBI change detection will become increasingly vital to support policy decisions to protect urban resident health and well-being.

5. Conclusions

This study provided a novel examination of urban green and blue infrastructure (UGBI) change within 29 urban land use classes for the case study city of Manchester, UK between the years 2000 and 2017. The main findings for Manchester include:

• 11% of existing UGBI in 2000 was lost by 2017.
• Dynamic change in UGBI, with 6.4% of the study area recording gains in UGBI compared to 11.9% of the study area recording losses in UGBI.
• All urban land use classes (n = 29) record areas of UGBI gain and loss; however, overall rates, considering the balance between gains and losses, are negative for the majority (58%) of classes.
• Projecting rates of change into the future indicates that nearly two-thirds (64%) of future UGBI loss could occur within existing Low- and Medium-density residential areas in the city.

Overall, Manchester has experienced a decline in the extent of UGBI, whilst its resident population increased in the early part of the 21st century. Local access to associated ecosystem services for the average resident over this time period is likely to have been impacted. Whilst the extrapolation methods here can only provide an indication of future UGBI provision, the overall decline in UGBI across the city is expected to persist as the population and economy of the city continue to develop (as they have done since the end of the study). This is a concern given that the ecosystem services provided by local UGBI support resident well-being and regulate the impacts of extreme weather events, which are likely to increase in severity and frequency in future with climate change.

The findings here have implications more widely across the UK and further afield in providing proxy indications of environmental change in cities that share similar socio-economic and structural characteristics. The framework offers a suitable approach adopting readily available methods and data, which may be adapted in other urban areas. To the best of our knowledge, no other study has attempted to measure this process for such a wide range of land use classes across an entire an urban area. The UGBI to land use change rates contribute novel information for further research, to improve land cover/land use change prediction, and support scenario-based policy development for effective environmental land use management.