Urban growth for the period 1986 to 2020 was derived by applying a land cover classification workflow to 30 m resolution Landsat satellite imagery for the land cover AOI (Figs. 2a and 3a), including Landsat 4 Thematic Mapper (TM), Landsat 7 Enhanced Thematic Mapper Plus (ETM+), and Landsat 8 Operational Land Imager (OLI). Landsat imagery was selected June to September to avoid cloud cover during the wet season (Domínguez-Castro et al., 2018). Therefore, seasonal spectral variations in land covers are not captured. Images were pre-processed using Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr) and Google Earth Engine to create multi-image mosaics with minimal cloud cover using a medoid pixel composite (Gorelick et al., 2017; Kennedy et al., 2018). Training data were manually digitised as 500 polygons (median polygon area of 5400 m2) with reference to the 1986 image using four classes: (1) urban, (2) woodland, (3) scrub vegetation and bare ground, and (4) agriculture and grassland. Training data were masked using the normalised difference vegetation index (NDVI) vegetation loss and growth masks that are output from LandTrendr to leave areas of training data that were spectrally consistent through time (1986–2020). Landsat composites were stacked with elevation and slope layers derived from the 30 m Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) (Farr et al., 2007) since these additional variables were shown to improve land cover classification performance (Zhu et al., 2016). We used a random forest classification, which is a decision tree approach popular for land cover classifications owing to their high accuracy, broad data handling, and low sensitivity to training data noise (Rodriguez-Galiano et al., 2012; Zhu et al., 2016). The Orfeo ToolBox random forest classifier (Inglada and Christophe, 2009) (Table S1) was run 50 times for each time period using 200 trees and a random sample of training data to account for imbalance between classes (Millard and Richardson, 2015) (Table S1). The modal value was used to produce the final classification map, which was accuracy assessed using an independent stratified random sample of 200 reference points in each class created using high-resolution satellite imagery (Fig. S2). High-resolution multispectral satellite imagery was not available in the 1980s, which reduces classification confidence in training and reference data; however, a panchromatic ∼ 1 m resolution aerial orthophoto of Quito in 1977 from the Instituto Geográfico Militar (1977) was used for reference. The accuracy assessment was used to produce an error-adjusted area and confidence interval of each land cover classification (e.g. Olofsson et al., 2013, 2014).

Future urbanisation scenarios in Quito were assessed with reference to Bonilla-Bedoya et al. (2020b) and Salazar et al. (2020). Both studies used predictor variables to model future urbanisation scenarios in Quito. Salazar et al. (2020) present a scenario to the year 2050, whereas Bonilla-Bedoya et al. (2020b) define an “urbanisation probability” without a scenario end date. Nonetheless, the spatial trends in both studies are similar. Predictors used to derive urbanisation probability included biophysical (e.g. precipitation, slope, and altitude), land cover and management (e.g. protected areas), infrastructure and services (e.g. road network), socio-economic (e.g. land value), and landscape metrics (e.g. landscape patch size and shape) (Bonilla-Bedoya et al., 2020b). We used “high” (urbanisation probability: 55 %–79 %) and “very high” (urbanisation probability: 79 %–100 %) classes from Bonilla-Bedoya et al. (2020b) in this study (Fig. S3) to evaluate future land cover scenarios and the intersection of urban areas with hazards.

The 30 m SRTM DEM was used to extract statistics on the elevation and slope within the land cover change area of interest (AOI), which encompasses the smaller city AOI (Fig. 2a). A higher-resolution (2 and 10 m) DEM and orthoimagery were created for a smaller AOI (Fig. 2a), which bounded the administrative level 3 parishes of Quito, Cumbaya, Llano Chico, Calderon (Carapungo), Conocoto, Zambiza, and Nayon. This AOI was covered by tri-stereo Pleiades imagery, which was acquired on five separate dates (5 November 2019, 28 January 2020, 9 February 2020, 6 June 2020, and 28 July 2020) in both panchromatic (∼ 0.7 m) and multispectral (∼ 2.8 m RGB and near-infrared) modes (Table S2). Tri-stereo acquisitions produce elevation models with lower uncertainties compared to bi-stereo acquisitions due to greater point cloud densities afforded by the extra viewing angle (Zhou et al., 2015). All imagery was delivered with radiometric processing to surface reflectance and processed using rational polynomial coefficients (RPCs) without ground control points (GCPs) (e.g. Airbus Defence and Space, 2012; Zhou et al., 2015). Agisoft Metashape v.1.6.5 was used to process the imagery to create a digital surface model (DSM), digital terrain model (DTM), and orthorectified imagery. Briefly, (1) the panchromatic and multispectral imagery was aligned in one bundle to produce a sparse point cloud; (2) the sparse cloud was filtered to remove outliers using Metashape's gradual selection tools; and (3) a dense point cloud was constructed using the panchromatic imagery, which was used to create a 2 m resolution DEM and (4) orthorectify the satellite imagery. Metashape's ground classification (maximum angle: 15∘; maximum distance: 0.5 m; cell size: 50 m) was applied to the dense cloud and used to create the DTM. An additional DSM was output at 10 m resolution to reduce data gaps for deriving a topographic wetness index (TWI) (Sect. 3.3).

Since the Pleiades DEM was processed without GCPs, we assessed the accuracy using Ice, Cloud and land Elevation Satellite (ICESAT-2) altimetry data. ICESAT-2 data have an expected vertical accuracy that is lower than the error expected from a Pleiades DEM created without ground control points (> 3–5 m) (Passalacqua et al., 2015; Markus et al., 2017) and were therefore used as an independent validation check. We extracted high confidence returns from the Advanced Topographic Laser Altimeter System (ATLAS) instrument ATL03 Global Geolocated Photon Height data acquired from 6 December 2018 to 3 June 2020 that intersected with the Pleiades data (Neumann et al., 2019, 2020). Photons were filtered to exclude slopes steeper than 20∘ and aggregated into 5 m grid cell mean values. Cells containing ≥ 2 photons with an elevation range ≤ 1 m were carried forward for the validation (n=11922). We coregistered the Pleiades DEM and gridded ICESAT-2 data following the x, y, z shift correction of Nuth and Kääb (2011), and the differences in elevation values were compared. The mean vertical difference between the ICESAT-2 and Pleiades data was 0.38 m (1 standard deviation: 1.32 m) with a normalised median absolute deviation of 0.84 m.

Information on natural hazards affecting Quito were collated from published sources and Ecuador's Open Government data. We used a global landslide susceptibility model that was validated against local and global landslide inventories, with an emphasis on rainfall-triggered events (Kirschbaum et al., 2016; Stanley and Kirschbaum, 2017). Landslide susceptibility was ranked on a scale of 1 (low) to 5 (high), and the model combined data on slope, faults, geology, forest loss, and road networks, aggregated to ∼ 1 km grid cells (Stanley and Kirschbaum, 2017). Open Government records of “accidents” 2006–2017 were used to identify the geographic distribution of mass movement events (n=1321), which were compared to the global landslide susceptibility model (Fig. S4) (Ministry of Territory, 2020). We masked Class 5 (high) of the landslide susceptibility model out of the future urbanisation scenario of Bonilla-Bedoya et al. (2020b) to create a restricted scenario of urban growth, which reflects DMQ's vision to remove high risk areas from future land occupation. We also excluded development on the slopes of Pichincha volcano (as unrealistically inaccessible given steep slopes) and included an area of development spanning the metropolitan district boundary in the south (Fig. S3). We refer to the original scenario of future urbanisation and the modified scenario as F-U and M-U respectively. Information on volcanic hazards was obtained from the Geophysical Institute of the National Polytechnic School (IG-EPN) through the National Information System (SNI) (SNI, 2020). Spatial variation in earthquake hazard across Quito was not explored in this study due to the coarse resolution (∼ 10 km) of available hazard information (Fig. S1). However, the high regional seismic hazard (Alvarado et al., 2014; Pagani et al., 2018) motivates our city-wide analysis of greenspace.

The 10 m Pleiades DEM was hydrologically corrected by breaching sinks (Lidberg et al., 2017), using the breach depressions least cost tool of Whitebox 1.4.0. The breached DEM was used to derive a TWI, which was intersected with flood events in the Open Government database (n=1274) to assess whether high TWI values correspond to greater incidences of flood events and therefore was indicative of potential flood hazard (Jalayer et al., 2014; Kelleher and McPhillips, 2020). 1TWI=ln⁡atan⁡β, where a represents the specific catchment area, and tan⁡β represents the local DEM slope. Therefore, the TWI describes the tendency for a cell to accumulate and evacuate water (Beven and Kirkby, 1979; Manfreda et al., 2011; Mattivi et al., 2019).

We assumed a positional uncertainty radius of 20 m in the flood event records based on the observed positional spread of recorded traffic collisions at road junctions in the same database (Fig. S5). The maximum TWI value within a 20 m radius of the recorded point was extracted and compared to the TWI for a random sample of 10 000 points to test whether there was a statistically significant difference in the TWI at locations of flood events (e.g. Kelleher and McPhillips, 2020). Notably, this method does not account for the subsurface drainage network present in an urban setting and therefore represents an assumption that this subsurface drainage network is overwhelmed during the flood event such that all flow passes over the DEM (Kelleher and McPhillips, 2020).

Orthorectified multispectral Pleiades imagery was pan-sharpened in ArcGIS Pro 2.6.0 using the Gram–Schmidt algorithm and Pleiades sensor band weights to create a four-band (red, green, blue, and near-infrared (NIR)) 0.5 m resolution multispectral image. Quito's vegetated greenspace distribution was mapped using the NDVI applied to the NIR and red bands of the pan-sharpened Pleiades satellite imagery (Fig. 3b). 2NDVI=(NIR-Red)(NIR+Red) Negative NDVI values correspond to areas lacking vegetation, whereas increasingly positive values represent healthy vegetation (Tucker et al., 1981; Pettorelli et al., 2005). In some cases, shadowed areas, for example due to buildings, display similar NDVI values to vegetation (Leblon et al., 1996; Yamazaki et al., 2009). We therefore used 100 randomly sampled patches (200×200 m) to evaluate the NDVI classification with reference to the pan-sharpened Pleiades orthoimage. Incorrect classifications had a small overall impact, accounting for 0.4 % of the evaluated NDVI area (Table S3) with a mean patch size of 13 ± 16 m2. Bright blue roofs also displayed a high NDVI value and were masked out using a simple “blueness” index of values ≤ 0.2, which was derived through manual inspection of blue roofs. 3Blueness=2×Blue-Red-Green

Whilst global coverage and daily observation are possible with the paired constellation, Pleiades imagery is not routinely acquired nor open access. Therefore, we also compared Pleiades NDVI values with those from an open-access Sentinel-2 image acquired on 6 February 2020 with the aim of testing their consistency, noting that whilst the spectral bands overlap, the bandwidth of Pleiades is greater (Pleiades: red 590–710 nm, NIR 740–940 nm; Sentinel-2: red 649–680 nm, NIR 780–886 nm).

Our land cover classifications showed that the urban area of Quito expanded ∼ 192 km2 over the study period, more than doubling from 160 ± 50 km2 in 1986 to 352 ± 47 km2 in 2020 (Fig. 4, Table S5). Urban expansion was primarily aligned along-valley (north–south) and eastward (Fig. 5a) into areas of previously scrub vegetation/bare and agricultural/grassland classes. The future urbanisation scenario of Bonilla-Bedoya et al. (2020b) covered an urban area of 1232 km2 (F-U), whereas the M-U scenario covered 705 km2 (Fig. 4a), which was still double the observed 2020 urban area. Future urbanisation in the modelled scenarios was predominantly eastward, where lower-density urbanisation interspersed with the scrub vegetation/bare ground class was already apparent in 2020 (Fig. 5). The area of woodland and agriculture/grassland classes also increased in 1986–2020. A notable example of afforestation (4.8 km2) was the park Metropolitano del Sur, which is located on the southeast of the city limit (Fig. 5a).

The median elevation of Quito's urban extent in 2020 (2780 m) was similar to 1986 (2810 m); however, the city covered a broader elevation range in 2020, tending towards lower elevations (Fig. 6a), which was also apparent for the F-U and M-U scenarios. The urban class displayed the smallest spread of values for topographic slope (Fig. 6b). Here, the median slope of the urban class was ∼ 5∘ in 1986 and 2020; however, this increased to 11 and 7∘ in the F-U and M-U scenarios respectively, in addition to a broader spread of slope values. Woodland featured the highest median slope of all land cover classes (∼ 28∘) and a comparable median elevation to the urban class (∼ 2700–2800 m).

Landslides are one of the most common natural hazards in Quito (DMQ, 2018). We found good spatial association between observations of landslide events in Ecuador's Open Government database (2006–2017) and a landslide susceptibility model (Stanley and Kirschbaum, 2017) (Fig. S4). Of 1321 recorded events, 82 % (n=1089) fell within landslide susceptibility categories 3–5, of which 44 % (n=576) were in the highest category (5). A total of 10 events were observed in the lowest category (1). We observed a small change in the landslide susceptibility of the urban class for 1986–2020. Here, the urban area in the highest landslide susceptibility categories (4 and 5) increased by 2 percentage points for 1986–2020 (Fig. 6c). The largest change was observed in the agriculture/grassland class, which featured a 9 percentage point increase in category 5 (high) landslide susceptibility. Woodland mostly occurred within the highest landslide susceptibility category 5 (87 %) (Fig. 6c). Regarding future urbanisation, the M-U scenario restricted future urbanisation in landslide susceptibility category 5; therefore, the observed percentage of urban area in category 5 (6 %) was notably lower than in the F-U scenario (47 %), which did not enforce any restrictions.

Flood events in Quito that were recorded in Ecuador's Open Government database were evaluated alongside a TWI derived from the 10 m resolution Pleiades DEM, noting that this does not account for subsurface drainage. Median TWI values for all flood events (n=1274), clustered flood events where two or more events were located within 40 m of each other (n=125), and a random sample (n=10000) were 13.3, 14.4, and 12.1 respectively (Fig. S6). Clustered flood events, which displayed the highest TWI, could correspond to areas of nuisance flooding since multiple events are located in close proximity (Kelleher and McPhillips, 2020). Two-sample independent Welch t tests (one-tailed) showed that the difference in TWI values between all flood events and clustered floods events was statistically significantly different from the random sample (p<0.05). Therefore, the mean TWI value was observed to be larger in areas of flood locations compared to the random sample.

Quito includes multiple types of greenspace that provide ecological, social, and disaster risk reduction benefits (Figs. 1a, 7). Within our AOI, 18.6 km2 of potential DRR greenspace was identified, which covered 6 % of the urban zone (Fig. 8). DMQ-designated greenspace had an area of 36.9 km2, of which 2.5 km2 (7 %) intersected with potential DRR greenspace. Similarly, DMQ-designated safe spaces covered 17.3 km2, of which 1.7 km2 (10 %) intersected with potential DRR greenspace. Comparing DRR greenspaces with hazard information revealed that 62 % of DRR greenspace intersected with areas of high TWI values (≥ 14.4 (median value for clustered flood events; Sect. 4.2)), 10 % intersected with areas of high (category 5) landslide susceptibility, and 6 % intersected with both hazards (Fig. 8b).

The association between population, socio-economic classification (Instituto Geográfico Militar, 2019), and greenspace accessibility was investigated for greenspaces ≥ 2000 m2. The number of people living within close proximity to designated greenspace was higher than for DRR greenspace (Fig. 9a). For example, 2.3 million (98 %) of Quito's population were within 800 m of a designated greenspace, compared to 2.1 million for the DRR greenspace (88 %). Distance to the nearest greenspace was greater for “low” and “medium-low” socio-economic classifications compared to “high” and “medium-high” (Fig. 9b). Here, the difference in median values was greatest for designated greenspace (466 m), compared to our classification of DRR greenspace (80 m). The amount of designated greenspace per person was smaller for lower socio-economic classifications, with a median of 3 m2 per person for the “low” classification compared to 8 m2 for “high”. However, the amount of DRR greenspace was greatest for lower socio-economic classifications, with a median of 24 m2 per person for “low” compared to 4 m2 for “high” (Fig. 9c). This reflects lower population densities on the city margins (Fig. 9d) and the persistence of agricultural land and undeveloped ground in these areas following urbanisation.

Quito's historical urban expansion is largely aligned north–south, whereas future urban expansion is focussed to the north and east (Fig. 5). Our study captures a period of land occupations starting in the 1980s including the settlement of Atucucho (Figs. 2b, 5a), which formed informally in 1988 (Testori, 2016). This occupation is visible in our land cover classification (Fig. 5a). The formation date is labelled as 2003 in Open Government data (Fig. 2b), which likely reflects its origins as an informal settlement that was potentially not included in official maps until 2003. In this case, satellite imagery can capture the urban sprawl of a city, including occupations that may not be apparent in historical maps. However, image classification methods usually only capture 2D sprawl and not vertical high-rise developments or redevelopments that are important for measuring exposure to natural hazards (e.g. Amey et al., 2021). Quito's past and projected urban growth has been studied by several authors in recent years (e.g. Bonilla-Bedoya et al., 2020b; Salazar et al., 2020; Valencia et al., 2020). Cross-comparisons are complicated by the use of different study areas since Quito's urban area now exceeds the designated metropolitan district boundary, which has prompted investigations to create a new district area (Salazar et al., 2021). By comparing our urban classification (year 2020) to that of Bonilla-Bedoya et al. (2020b) (year 2016) within the same area of interest, we find urban areas of 213 and 210 km2 respectively, which indicates classification consistency using EO data despite different methodological approaches.

We observed that expansion of Quito and future projections tend towards lower elevations (Fig. 6a) and steeper slopes (Fig. 6b), the latter of which is associated with encroachment into areas of high landslide susceptibility (Fig. 6c, d). Limited urban expansion to the west of Quito on the steep slopes of Pichincha volcano suggests that a programme of protection to avoid encroachment is working (Vidal et al., 2015). However, several of these areas or their vicinities are inhabited because of previous land invasion dynamics that affected the peripheral green belt. They can be characterised from a spatial and socio-economic approach as a homogeneous space, in which the less economically favoured classes experience greater possibilities of isolation from other social groups (Bonilla-Bedoya et al., 2020a). Further limiting eastward urban growth reduces the ashfall and lahar hazard in the event of an eruption (Fig. 2c) and the hazard posed by landslides (Fig. 2d). Additionally, the predominantly woodland slopes east of Quito (Fig. 5a) featured the highest landslide susceptibility scores (87 % of woodland is in class 5 (high) (Fig. 6c)) and are therefore a valuable target for protection against urbanisation. Our observed decreasing elevation trend of Quito's urban area (Fig. 6a) reflects north–south and eastward expansion into lower-lying flatter areas such that at a city scale, Quito's landslide susceptibility did not notably increase from 1986 to 2020 (Fig. 6c). These areas are also the location of projected future expansion (Bonilla-Bedoya et al., 2020b; Salazar et al., 2020; Valencia et al., 2020), predominantly through conversion of scrub vegetation and bare ground (Fig. 5a). Notable ravines exist in these areas; therefore, risk-informed planning to reduce encroachment on steep slopes, which was reflected in our M-U future urban scenario, is desirable to minimise landslide risk to future developments. These areas are also likely to be most susceptible to multi-hazards such as rainfall-triggered lahar remobilisation or landslides, as well as flood- and earthquake-triggered landslides (Gill and Malamud, 2017). Similarly, the filling of ravines from the 17th century onwards restricts the drainage capacity during intensive rainfall and increases flood risk (Aragundi et al., 2016); therefore, incorporating additional DRR greenspaces here to attenuate run-off and store water could be beneficial.

While risk-informed urbanisation can mitigate some hazards such as landslides, an intensive earthquake hazard exists in Quito (Fig. S1) such that urban risk reduction requires building resilience at community to city-wide levels (Alvarado et al., 2014; Valcárcel et al., 2017). A key element of resilience is the access to “safe spaces” following an earthquake event in which communities can avoid damaged buildings and infrastructure and receive emergency aid (Sphere Association, 2018). These spaces are increasingly viewed within a broader network of benefits to society and ecosystems (e.g. Fig. 1a) and are framed within Eco-DRR strategies (UNDRR, 2020). We therefore evaluated greenspaces in Quito that could offer DRR capabilities by both considering existing designated greenspaces and assessing other non-designated greenspaces.

Our study was designed to identify the basic requirements for sites that could be designated or developed as DRR greenspace using an earth-observation-based methodology that could be adapted and applied to other cities. This is timely since greenspace is becoming increasingly desirable to improve environment quality and contribute to addressing climate breakdown, and greenspace within Eco-DRR strategies can simultaneously mitigate against multiple hazards (Onuma and Tsuge, 2018; McVittie et al., 2018; Sudmeier-Rieux et al., 2021). Our DRR greenspace primarily addresses the basic requirements of people-space and amenable topography for medium- to long-term accommodation requirements, such as following a major earthquake. Examples are shown in Fig. 12 for areas in central Quito and on the periphery. Regarding urban risk, green space in Quito has been thought of from the perspective of threat. For example, interventions have been developed on the slopes of Pichincha from a logic of risk mitigation (Vidal et al., 2015). Recently, after the 2016 Ecuador earthquake, green and open spaces were incorporated throughout the city as safe points in case of evacuation (Rebotier, 2016) (Fig. 10a).

We found that 7 % (2.5 km2) of the DMQ-designated greenspace was identified as potential DRR greenspace. Similarly, 10 % (1.7 km2) of the DMQ-designated safe spaces intersected with our classified DRR greenspace (Figs. 8, 10a). The total area of potential DRR greenspace within Quito was 18.6 km2; therefore, a large potential exists to incorporate new greenspaces into a DRR framework, especially in the south and east of the city, which are locations of projected future expansion and where urban expansion and population densities are lower (Figs. 5b, 9d). New designation of greenspaces could address some of the imbalance between greenspace access since 98 % (∼ 2.3 million) of Quito's population was within 800 m of a designated greenspace, compared to 2.1 million for the DRR greenspace (88 %) (Fig. 9a). Lower socio-economic classifications had a greater distance to travel to the nearest designated greenspace and a lower greenspace area per person overall (Fig. 9b, c), which was also observed by Cuvi et al. (2021), noting that informal developments have less access to larger designated parks. We found a median designated greenspace of 3 m2 per person for the “low” socio-economic classification. However, the availability of potential DRR greenspace to these same communities (median of 24 m2) shows that additional designations could help address the imbalance. This is also aligned with Quito's Vision 2040 document to increase greenspace in urban areas to ∼ 9 m2 per person (DMQ, 2018). Critical to addressing these inequalities is to ensure that all formal and informal settlements are reflected in socio-economic statistics and included in official maps.

Although we found high accessibility of greenspace within 800 m of populations, the capacity to serve surrounding populations for emergency refuge was 1.7 % considering the recommended space allocation of 45 m2 per person (Fig. 9a) (Sphere Association, 2018). Incorporating all additional spaces that are DRR suitable could increase this to 8 % or 40 % using a minimum living allocation of 3 m2 per person (Sphere Association, 2018). A network analysis producing the ranked top 10 DRR greenspaces (Fig. 11) showed that eight intersected with currently designated greenspaces or safe spaces, and two did not. These two spaces could be investigated for negotiating formal access to these spaces for use in an emergency, such as the golf course forming Site 5 (Fig. 11e).

We focus on greenspace as an emergency refuge; however, these spaces can also contribute to mitigating hazards both through physical processes such as water retention or slope stabilisation (Phillips and Marden, 2005; Maragno et al., 2018; Sandholz et al., 2018) and also through their existence in places that would be hazardous if urbanised. We found that of the potential DRR greenspace identified in Quito, 62 % intersected with TWI values indicative of potential flooding (Sect. 4.2), 10 % with areas of high landslide susceptibility, and 6 % with both hazards (Fig. 8, red circles). Therefore, there is potential to mitigate future risk by maintaining greenspace and therefore avoiding development in potentially hazardous areas, as well as incorporating additional DRR greenspaces that are not exposed to hazards for use as refuges.

Our study has provided a city-wide assessment of Quito's historical and future growth projections, as well as the potential role of greenspace in reducing disaster risk. The first-pass analysis of greenspace suitable for DRR could be used for local community-scale evaluation and stakeholder engagement to deliver improved resilience for the city. Subsequently, the methodology could be expanded to define a continuum of greenspace suitability for DRR by incorporating other important factors including site-specific suitability trade-offs such as land value, ownership, and access to water, electricity, and hospitals (Anhorn and Khazai, 2015; Hosseini et al., 2016). Similarly, we focussed on greenspaces since these spaces are most likely to be accessible, and they provide multiple benefits; however, concreted grey spaces such as commercial car parks could also serve a role in providing safe spaces for DRR, particularly if a disaster event occurred during work hours. Methodological developments could include multi-temporal and potentially higher-resolution datasets, for example landslide susceptibility information that reflects changing land cover and therefore an evolving hazard (Emberson et al., 2020). For example, a dynamic landslide susceptibility map could consider a potentially increased landslide hazard due to road cuttings in areas undergoing urban development (Froude and Petley, 2018) and the dynamic nature of landslide hazard in response to precipitation events (Kirschbaum and Stanley, 2018). Additionally, our investigation of flood events alongside a TWI would benefit from a better understanding of the capacity and distribution of the subsurface drainage network within Quito, particularly where natural drainage channels are blocked (e.g. Aragundi et al., 2016). Nonetheless, our assumptions that all flood water would flow on the surface represents a worst-case scenario during a flood event in which the artificial drainage network is at capacity.

The use of EO-based datasets broadens the applicability of our methods to other cities. Whilst other sources of multispectral satellite imagery (e.g. 3 m resolution PlanetScope or 10 m resolution Sentinel-2) could still delineate the types of greenspaces relevant to DRR (e.g. Fig. 8 inset), we relied on a high-resolution Pleiades DEM to provide topographic relief information on the greenspace DRR suitability. Global 30 m resolution DEMs could likely substitute this in some cases, though they are potentially less suitable in densely built urban environments where flat open greenspaces are interspaced with tall buildings and trees for example (Fig. 12a), which cannot be distinguished in 30 m elevation models. Here, elevation and slope values derived from 30 m resolution DEM represent an average of features (for example buildings, cars, and trees) within the 30 m cell. Therefore, the topography of greenspaces is resolved in less detail.