The current version of the U-Surf dataset is based on the urban parameterization scheme embedded in the CESM2 for two reasons. First, CESM2 is one of the very few state-of-the-art ESMs with a physically based UCM – the Community Land Model Urban (CLMU) (Lawrence et al., 2019; Oleson and Feddema, 2020). The CLMU has been evaluated against site observations and satellite measurements across the world with consistently reasonable agreement (Demuzere et al., 2008, 2013, 2014, 2017; Fitria et al., 2019; Li et al., 2024a, b; Lin et al., 2016; Oleson et al., 2008a, b; Zhang et al., 2023a; Zhao et al., 2014, 2021) and has also demonstrated high credibility among various UCMs in the recent Urban-PLUMBER multi-model comparison project (Lipson et al., 2024). Second, the urban canopy concept that the CLMU uses is widely adopted in various UCMs embedded in weather models and regional climate models (RCMs). Therefore, a dataset developed based on this conceptual representation can be easily extended to other UCMs within climate and weather models.

The urban canopy representation used in the CLMU and many other UCMs is called an “urban-canyon” schema, where the urban landscape at a given location is conceptualized as an infinite urban canyon (Fig. 1). This canyon hypothesis assumes a geometry of an infinitely long street bordered by two building walls with identical height. An urban canyon consists of five facets: building roof, impervious (e.g., roads, parking lots, sidewalks) and pervious (e.g., lawns, street trees, parks) canyon floors, and sunlit and sun-shaded walls (Oleson et al., 2008a). This conceptual representation reduces the considerable complexity of urban surfaces into a single urban canyon and yet provides an essential base to represent key urban biogeophysical processes effectively. The UCMs using this approach therefore require sets of properties at both the facet and canopy level to represent urban landscapes and model their interactions with the lower atmosphere in climate and/or weather simulations. These properties can generally be grouped into three categories: morphological (e.g., canyon height-to-street-width ratio, roof fraction, average building height, and pervious canyon floor fraction), radiative (e.g., facet-level albedo and emissivity), and thermal (e.g., heat capacity and thermal conductivity) (Fig. 1 and Table 1). These surface properties characterize the “urban areas” and are critical for constraining their surface energy budget and, thus, the near-surface microclimate in weather and climate models. More details about the CLMU parameterization scheme and its evolution over the years can be found in Jackson et al. (2010), Lawrence et al. (2019), Li et al. (2024b), Oleson et al. (2008a, b), and Oleson and Feddema (2020).

The new urban surface parameter dataset, U-Surf, describes the global urban areas in a spatially continuous manner, providing all the required parameters in three categories (radiative, morphological, and thermal) that are compatible with the urban canyon representation in the CLMU and, potentially, in UCMs. We developed a multi-step workflow on the Google Earth Engine platform (Gorelick et al., 2017) that leverages four key categories of data: segmented land use–land cover maps, 3D building footprints, high-resolution satellite observations, and thermal properties of construction materials. Utilizing these, we first generated segmented urban imagery, which distinguishes among different facets. Then we integrated this imagery with satellite observations to derive facet-level radiative properties and fractional parameters. From there, we synthesized multiple data sources to construct the 3D urban canyon morphological attributes. Finally, we incorporated existing databases to produce thermal properties (Fig. 2).

After estimating all the required parameters as described above, we took several additional steps to ensure the accuracy, coherence, and transparency of our U-Surf data product, including masking, gap filling, and quality control. First, we only retain the grids containing all three facets – roofs and impervious and pervious canyon floors – in U-Surf because a complete urban canyon can only be formed when all three of the facets are present. This constraint helps make U-Surf more consistent with the conceptual definition of physical urban land in a UCM and is an improvement over the J2010 dataset, which used urban density classes from urban form and population estimates (LandScan), leading to large over- and under-estimations of physical urbanization depending on the region (Chakraborty et al., 2024). Second, we masked out the grids with extremely high canyon height-to-width ratios (> 12) but low building heights (< 40 m). These grids are actually very sparsely built sub-urban or rural landscapes instead of densely built areas.

The last step is to gap-fill the missing values caused by synthesizing multiple datasets with different spatial coverage. For example, the emissivity product from ASTER GEDv3 has missing pixels in certain regions due to cloud coverage. These missing values were gap-filled using a simple approach. We combined two classification data: the Koppen climate zones (Beck et al., 2018) and the 33 urban regions defined in J2010, both at 1 km resolution. The average parameter values for each combined class were then used to fill the missing values of the corresponding parameters. Note that only a small proportion of grids need to be gap-filled, accounting for less than 3.5 % of the total among all parameters. To keep the aforementioned data source, processing, and gap-filling information accessible and to make it easier for users to track changes in future version releases, each parameter comes with an additional quality control (QC) band using a four-digit code (Table 2). The first and second digits differentiate between algorithms and single- or multi-source data, respectively, while the last two digits indicate whether the parameter was directly derived or gap filled. These QC codes are consistent across the entire dataset and will be updated accordingly in later versions.

Validating urban surface parameters on the global scale is extremely challenging, primarily due to the lack of globally consistent measurement networks. This challenge is exacerbated by the scarcity of long-term urban observational sites, especially in diverse urban environments. The inherent variability within urban areas further complicates validation efforts as data from one site may not represent the broader urban landscape. U-Surf is composed of the extraction of satellite measurements, satellite-derived products (i.e., land cover data and building footprints), and our own derived parameters. The satellite measurements and derived products have already been validated and quality-controlled by their development teams, as summarized in Table 3. U-Surf parameters derived based on these input data sources are therefore subject to their inherent uncertainties and uncertainty propagation during data synthesis and processing. To systematically evaluate the accuracy and uncertainty of U-Surf parameters, we first conducted a thematic validation based on the derived morphological parameters at 1 km resolution against the 3D World Settlement Footprint (WSF-3D, Esch et al., 2022) observational site data and Urban-PLUMBER site metadata. We then further employed Monte Carlo simulations to quantify the final uncertainties of U-Surf parameters arising from input data errors or uncertainties and their propagation (see Sect. 3.4 for a detailed discussion).

U-Surf demonstrates significant improvements over the default CLMU parameters. As U-Surf directly provides spatially continuous urban surface parameters without relying on any density class or land use classification, here, just for ease of comparison and illustrative purpose, we separated raw U-Surf pixels into the four urban density classes (TBD, HD, MD, and LD), following their locations as defined by J2010, and plotted the distributions of the urban surface parameters in both U-Surf and J2010 data at these locations (Fig. 3). The location data defined by J2010 and OF2020 at 1 km resolution can be accessed at 10.6084/m9.figshare.28169324.v1 (Cheng, 2025a). The overall distribution of U-Surf raw data is also shown in the figure. As most of the thermal properties in U-Surf are adapted from J2010, the discussion here will be mainly focused on radiative and morphological parameters, and the comparison of thermal properties can be found in Fig. S3.

Retrieved from direct remote sensing measurements, the radiative properties exhibit physically more reasonable ranges compared to J2010 data. As described above, urban surface properties in J2010 were estimated on the basis of building materials sampled in a predefined region and were then generalized to the entire region. This clearly leads to unreasonable values, such as abnormally low emissivity and high albedo, across entire regions for certain countries. The former issue is true not only for the CLMU but also for urban emissivity constraints in regional models like WRF (Chakraborty et al., 2021). For instance, the minimum roof emissivity in J2010 is as low as 0.04 in regions like Mongolia, Kazakhstan, France, and Germany (Figs. 3 and S5), and the roof albedo can be as high as 0.61 for Chili, Argentina, Mongolia, and Kazakhstan (Figs. 3 and S9). These values were derived from specific low-emissivity and high-albedo materials (e.g., zinc and galvanized steel coating), which might be possible in individual buildings but are highly unlikely for all urban areas in a large region (Chakraborty et al., 2021). Broadcasting to an entire region from sampled material estimates results in an oversimplified representation of urban surfaces. In contrast, the “effective” emissivity retrieved from ASTER GEDv3 (Hulley et al., 2015) in U-Surf is generally higher and more narrowly concentrated, typically between 0.95 and 1.0 across urban facets, with exceptions in a few specific areas. This pattern also aligns with urban canyon characteristics, where the effective emissivity of an urban canyon is slightly higher than the weighted-average values from all individual components due to the “canyon-trapping” effects (i.e., increased absorption from reflections between facets) (Harman et al., 2004; Oke et al., 2017). Likewise, we can observe a narrower spread of roof albedo values concentrated between 0.1 and 0.3 across countries, which align with the aggregated values from the commonly used urban roof materials such as tiles (0.10–0.35), shingles (0.05–0.25), and slate (0.08–0.18) (Oke et al., 2017). Our results confirm that the blue- or clear-sky albedo (total albedo for shortwave radiation) calculated in U-Surf, an interpolation between white- and black-sky albedo (Liang et al., 1999), represents the real-world conditions more accurately.

The morphological parameters in our dataset also provide more reasonable estimates of both mean values and variability. The four morphological parameters follow similar trends to J2010 in their variations across urban density types. For example, the roof fractions (pervious fractions) are generally higher (lower) in TBD locations identified in J2010 and decrease (increase) as the built density decreases (i.e., HD, MD, and LD). However, U-Surf captures much larger variabilities in these parameters compared to J2010, reflecting a more diverse urban morphology. This is again because of J2010's approach of applying uniform parameter values based on selected representative buildings in a region. This approach not only fails to represent the granular spatial variability in a region but also easily skews the estimates. For example, J2010 reported an unrealistically high roof fraction of 0.8 for the MD class over Brazil, whereas U-Surf presents a more realistic roof fraction predominantly ranging between 0.03 and 0.14, with a median value of 0.07 for this region, which aligns more closely with observations. Note that the median values of the four morphological parameters in U-Surf raw data (black boxes in Fig. 3) are generally lower (or higher in pervious fraction) than the TBD, HD, MD, and LD categories (blue boxes). This is because U-Surf raw data cover more pixels than the sum of locations identified as TBD, HD, MD, and LD in J2010, most of which are sparsely built landscapes. In fact, the less densely built urban areas dominate the global urban landscapes. The four density classes TBD, HD, MD, and LD in J2010, for example, account for 0.022 %, 5.85 %, 23.76 %, and 70.37 % of all urban grids, respectively.

The H/W values in U-Surf are concentrated within ranges of, respectively, 0.6–1.4, 0.2–0.8, and 0.1–0.4 for the TBD, HD, and MD locations identified in J2010. These values are close to real-world observations which typically vary between 0.5 and 2 at the neighborhood scale (Vardoulakis et al., 2003). Note that high H/W values are very rare at a 1 km resolution in real cases. Only very densely built central metropolitan areas (such as the lower Manhattan area in New York City, US) exhibit ratios exceeding 1. These occasions, however, usually only constitute a small proportion. This explains why the overall raw U-Surf H/W values are mostly concentrated between 0.06 and 0.5. We note that, in very rare cases, there are some very high roof fraction numbers (≥ 0.9) in U-Surf, which, nevertheless, are not located in central metropolitan areas. These outliers are places with relatively large roof cover but very small urban impervious areas (such as near the edge of a sub-urban area or in small-sized dispersed town areas).

In this section, we present selected radiative and morphological parameters as illustrative examples to demonstrate the improvement of the new urban surface dataset in terms of spatial heterogeneity, granularity, accuracy, and broader applicability from the global to city scale. The global maps of the complete list of parameters can be found in the Supplement (Figs. S5–S26).

U-Surf exhibits significant advancements in capturing spatial heterogeneity and granularity, which is a crucial improvement over traditional categorical urban classifications, such as density classes used in the CLMU and LCZs used in state-of-the-art mesoscale models. The dataset's fine-scale resolution reveals detailed variations in both urban radiative and morphological parameters (Figs. 4 and S5–S17). For instance, in J2010, the emissivity of pervious canyon floors is uniformly set at 0.95 globally to represent a typical value for vegetation using a bulk parameterization scheme (Oleson et al., 2010), and roof albedo is limited to 11 distinct values (Fig. 4a and b). These discrete values lead to oversimplifications that fail to represent the critical variations in urban areas, potentially affecting the accuracy of cross-sample variability in urban climate states. Conversely, U-Surf data show clear variability both within and across regions (Fig. 4c and d). In general, albedo exhibits greater variability than emissivity across different facets (Figs. S5d–S12d). The albedo of impervious canyon floor is comparable to that of the pervious one, while roof and wall albedo tend to be higher, especially in the city center, with densely built tall buildings (Figs. S9d–S12d). In New York City, for example, the mean albedo values are 0.13 for both impervious and pervious canyon floors, while roof albedo averages 0.16, and wall albedo is even higher at 0.22. This pattern is consistent with the fact that commonly used road pavement materials, such as asphalt and concrete, exhibit similarly low albedo values when compared to urban vegetated surfaces like parks and lawns (Oke et al., 2017). Moreover, roofing materials in metropolitan areas often feature higher reflectivity to reduce heat absorption by buildings (Jia et al., 2024), further contributing to the observed differences in albedo. These variations reflect not only the differences in the materials used but also adaptation strategies to local climate conditions, thereby providing more insights into local climate-sensitive urban design practices.

At the global scale, U-Surf also reveals high-level distinct spatial patterns that correspond to the varying stages of urban development across regions (Fig. 5). In the Global North, particularly in Europe and the United States, urban areas typically exhibit higher building density (roof fraction × urban percentage), greater average building height, and higher average canyon height-to-width ratio. These characteristics are indicative of more developed urban forms and well-established infrastructure, often driven by the need to accommodate growing populations in limited spaces. For instance, metropolitan centers (e.g., Manhattan, New York City, USA; Quartiers 1–4, Paris, France) in these areas frequently exceed 30 %–40 % roof coverage, with average building heights surpassing 30 m. In contrast, the Global South (e.g., Latin America, Africa, and Central Asia) generally exhibit lower values for these parameters. For example, building density values in these regions are, respectively, 38.59 %, 46.46 %, and 88.71 % lower than in the United States. Similarly, their median building heights are, respectively, 11.94 %, 31.65 %, and 12.75 % lower than in Europe. Consequently, their median canyon height-to-width ratios are, respectively, 29.88 %, 37.18 %, and 23.99 % lower than those in Europe. However, this trend is rapidly changing in emerging economies, including India and Brazil, where cities are experiencing swift urban growth. For instance, rapidly urbanizing places such as Delhi, India, and Sao Paulo, Brazil, have demonstrated tall and densely built environments, where Delhi has a roof fraction of 31.02 % and an average building height of 12.63 m, while Sao Paulo has a roof fraction of 49.42 % and an average building height of 13.87 m, all of which exceed the 75th percentile in the global distribution (Fig. 3c). Additionally, regions such as East Asia exhibit urbanization patterns that are more akin to those in North America and Europe, characterized by high roof fractions (e.g., Fig. S4a) and significant vertical development. For example, many cities in eastern China have exhibited city-wide average roof fractions above 14 % and average building heights exceeding 13 m, reflecting rapid industrialization and economic growth that have rapidly transformed the urban landscape over the past few decades (Cai et al., 2022). These observations further demonstrate the fidelity of U-Surf in revealing globally comparable yet regionally nuanced urbanization representations, which are essential for understanding geographical disparities and advancing region-specific sustainable urban development.

At a more localized level, U-Surf uncovers intriguing patterns within countries and even individual cities, offering insights into the complex interactions between urban morphology and local climate conditions. For instance, in the southwestern United States (e.g., California, Arizona), in northern African countries like Egypt and Tunisia, and in northeastern China, U-Surf captures lower pervious canyon floor emissivity (below 0.93) and higher roof albedo (above 0.25) (Fig. 4c and d), reflecting the potential impact of arid conditions and the use of high-albedo materials for heat adaptation in hot climates. Furthermore, U-Surf highlights regional differences in surface morphological properties (Fig. 5), which play crucial roles in determining local urban climates. Building heights are notably higher along the coasts and in southern regions of the Contiguous United States (CONUS), with cities like New York, Chicago, and Miami showing exceptionally high values (> 100 m) and correspondingly high canyon height-to-width ratios (> 2) in city cores. These cities also exhibit high roof fractions, showing more clustered building patterns in city centers, with density decreasing outwards (Fig. S4b). In densely populated developing countries like India and China, high roof fractions exceeding 40 % are observed, particularly in regions such as the Indo-Gangetic Plain and the area spanning from the Bohai Economic Rim to the Yangtze River Basin (Fig. S4a). In underdeveloped regions of South America and Africa (e.g., Bolivia, Chad) with widely dispersed urban areas, buildings are more sparsely distributed, typically concentrated within fewer metropolitan areas. It is interesting to note that the high-resolution U-Surf data even capture the very densely populated informal settlements (such as the slum areas in Delhi) where buildings are tightly packed and often overcrowded (characterized by high roof fraction and population density) (Fig. S4c and d). This illustrates the potential use of U-Surf as a valuable tool to better inform us of socioeconomic disparities in environmental and climate hazards within cities, which is currently difficult to do using process-based models (Chakraborty et al., 2023; Zhao et al., 2021).

The high-resolution U-Surf data enable intra- and inter-city comparisons in global-scale urban climate modeling in an unprecedented way. To illustrate this point, we identified two cities with similar background climates: Chicago, IL, USA, and Seoul, South Korea (Fig. 6), both of which are classified as Dfa under the Köppen-Geiger climate classification (humid continental climate with hot summers) (Beck et al., 2018). Because of the coarse-resolution urban surface input in J2010, these two cities share the exact same roof-specific parameters of the MD class. However, U-Surf reveals distinct contrasts in their radiative and morphological properties. Chicago, which has a history of applying heat mitigation strategies such as cool roofs (Mackey et al., 2012; Zhao et al., 2014), demonstrates higher roof emissivity and albedo, with average values of 0.972 and 0.175, respectively, compared to Seoul's 0.955 and 0.114. With regard to intra-city variations, Chicago's urban form is characterized by a higher concentration of buildings, with an average roof fraction of 0.284, in the northern part of the city. High-rise buildings or skyscrapers are predominantly clustered around Lake Michigan and the Chicago Loop area. On the contrary, Seoul exhibits a more dispersed urban structure, with buildings being spread more widely across the city. Such detailed representation facilitates comprehensive attribution and sensitivity analyses, permitting the examination of how individual parameters, such as emissivity, albedo, and building height, can alter the city microclimate and further influence the role of cities in local to global climate change scenarios (Krayenhoff et al., 2018, 2021; Zhao et al., 2017) and potentially informs more actionable climate adaptation and mitigation strategies.

U-Surf demonstrates a remarkable ability to capture the spatial heterogeneity and textural details of global urban landscapes across scales. To demonstrate this point, we aggregated the 1 km U-Surf data to coarser resolutions of 0.125° and nominal 1° (a typical resolution that ESMs are run at) to compare with J2010 side by side. More detailed information about the aggregation process can be found in Sect. S1, Table S1, and Fig. S27. For illustrative purposes, only the comparisons of H/W are shown here. Our results demonstrate that U-Surf represents the detailed urban form considerably well, even at much coarser resolutions. The spatial variability and urban texture are well preserved at global (Fig. 7d), national (Fig. 7e), and city (Fig. 7f) scales. This further demonstrates the adaptability and application of U-Surf in multi-scale, cross-scale urban modeling studies, with potential usage in regionally refined models (RRMs) such as E3SM-RRM (Tang et al., 2023a), as well as in the variable-resolution models (Huang et al., 2016) like Multi-Scale Infrastructure for Chemistry and Aerosols (MUSICA; Pfister et al., 2020), where seamless transitions between different spatial scales are crucial for comprehensive and coherent analysis.

For the derived morphological parameters, we conducted a thematic validation based on two recently available, observation-based datasets, Urban-PLUMBER and WSF-3D. WSF-3D is a high-resolution (∼ 90 m at the Equator) global dataset that provides detailed three-dimensional information on building fraction, height, and volume derived from satellite imagery, offering crucial insights into urban structures and their spatial distribution across the globe (Esch et al., 2022). We compared the roof fraction and height at 1 km resolution across WSF-3D's 17 validation sites. The Urban-PLUMBER project primarily aims to enhance the understanding of the accuracy of current urban climate models and has also produced a harmonized dataset of quality-controlled and gap-filled observations from 21 urban flux tower sites across different climate zones and urban built environments (Lipson et al., 2022). We compared all four morphological parameters across these sites by using neighboring pixels around the flux towers to evaluate against the site-specific information.

The roof fraction showed strong agreement across the reference sites in both WSF-3D and Urban-PLUMBER, with low mean absolute errors (MAEs) of 0.076 and 0.081 (Fig. 8a). Similarly, the pervious fraction also aligned well at most Urban-PLUMBER sites, with a mean MAE of 0.124 (Fig. 9a). Some discrepancies were observed in building height (MAEs of 5.918 and 7.446 m, Fig. 8c and d) and canyon height-to-width ratio (MAEs of 0.387, Fig. 9c). These discrepancies are primarily attributed to the disparity between the neighborhood-scale values captured by flux towers, typically representing areas within several hundreds of meters, and the 1 km resolution averaged values. Detailed values at each individual site can be found in Tables S3 and S4.

As discussed briefly in Sect. 2.4, U-Surf's parameters are inherently influenced by the uncertainties embedded in the synthesized data sources and uncertainty propagation during calculations. To systematically evaluate the uncertainties in final U-Surf parameters, we first documented the available validation approaches, as conducted by the development teams, and associated uncertainties for all input data sources in Table 3. Based on these numbers, we then employed the Monte Carlo simulation approach to quantify the final uncertainties in all our derived urban surface parameters in U-Surf (see Sect. S2).

Specifically, the three datasets used to differentiate between roofs and impervious and pervious canyon floors demonstrate high global classification accuracy. The 10 m resolution ESA land cover (Zanaga et al., 2022) was validated using the updated Copernicus Global Land Service-Land Cover Validation (CGLS-100) dataset. The global overall accuracy across all land cover types is 76.7 ± 0.5 %. The confidence intervals for specific land cover types are 3.3 % for built-up surfaces and an average of 1.2 % for pervious canyon (the average value of tree cover, grassland, shrubland, bare soil). The MS-BFP data (Microsoft, 2022) were evaluated using building polygon labels from Bing Maps, including Maxar and Airbus data. The precision of semantic segmentation (i.e., building pixel detection) showed regional variations, with the lowest false-positive rate of 0.1 % in Mexico and the highest false-positive rate of 2.98 % in Indonesia. The EA-BFP data (Shi et al., 2024) were validated in sampled Chinese cities with manual annotation, compared against OSM building data and regional roof vectors (Zhang et al., 2022). The data have an overall average accuracy of 89.63 % and an F1 score of 82.55 %. The primary data source of building height (Che et al., 2024) underwent rigorous validation against various reference height datasets and selected cities from Google Earth Pro. The validated results showed R2 values ranging from 0.66 (Europe) to 0.96 (South America) and root mean squared errors (RMSEs) from 1.9 m (South America) to 14.6 m (Japan, North and South Korea) across different subregions. The supplementary dataset (Li et al., 2022) was also validated and compared against WSF-3D, yielding a global RMSE of 2.56 m, with the lowest RMSE of 1.35 m in Sub-Saharan Africa and the highest RMSE of 4.94 m in China.

All remote sensing products and algorithms used to derive radiative properties were validated against ground measurements with high credibility. ASTER GEDv3 (Hulley et al., 2015) was compared with MODIS Collection 4 and 5 emissivity and was validated against lab measurements at four large sand dune fields, yielding a relatively low average RMSE of 0.077. The broadband emissivity regression algorithm (Eq. 1) was validated against the ASTER spectral library covering the wavelength ranging from 2 to 15 µm, yielding an R2 of 0.913 and RMSE of 0.011 (Malakar et al., 2018; Ogawa et al., 2008). The 10 m land blue-sky albedo (Lin et al., 2022), retrieved from Sentinel-2 surface reflectance, was validated against local flux tower measurements, achieving an overall R2 of 0.94 and RMSE of 0.03 across five land cover types. The RMSE ranges from around 0.0154 for urban areas (see Sect. S2 for detailed calculations) to 0.032 for grassland. In addition, the narrow-to-broadband algorithm (Bonafoni and Sekertekin, 2020) demonstrated an R2 of 0.77 and RMSE of 0.023 when compared against the ground measurements at six surface radiation budget network (SURFRAD) stations. It also showed an R2 of 0.98 and RMSE of 0.021 when compared against albedo meter measurements at 18 Perugia sites (Bonafoni and Sekertekin, 2020).

The primary source of uncertainty in the AC adoption rate (Li et al., 2024b) stems from the linear model that correlates AC adoption rate with the number of AC units per household. The linear model with saturation effect has an R2 of 0.9 (p<0.001), RMSE of 11.5, and MAE of 8.5 (both in units of %).

Using these documented uncertainties, we conducted Monte Carlo simulations with 1000 trails of randomly perturbed input parameters based on 10 000 randomly selected samples across 10 countries (Table S2) to quantify the uncertainty of error propagation through our data synthesis and processing (Sect. S2). The resulting 95 % confidence intervals for all parameters across all sampled regions and global averages are presented in Table 4. These intervals provide the expected error and/or uncertainty ranges for our final estimates. Overall, the uncertainties propagated through our data synthesis and processing align closely with those in the input data and remain relatively small – partly due to spatial upscale from finer resolutions to 1 km – which confirms the robustness of our methodology.