Using spatially explicit machine learning to enhance assessment of the Global Gravity-based Groundwater Product for groundwater storage change in Germany

ABSTRACT Understanding spatial heterogeneity in warming rates is crucial for localized climate adaptation. This study employs a novel multi‐scale framework—integrating principal component analysis (PCA) and Multiscale Geographically Weighted Regression (MGWR)—to analyse drivers of normalized temperature trends (NTT) across the Middle East and parts of Central Asia. The MGWR model significantly outperformed global and local alternatives (adjusted R 2 = 0.98) and showed strong predictive skill on independent data ( R 2 = 0.93). Results confirm near‐universal warming (NTT > 1) but reveal pronounced spatial heterogeneity in its rate. Drivers separate into: (1) dominant, non‐stationary factors (e.g., snow cover, elevation and, specific forest types) with effects that reverse sign regionally; (2) strong, consistent moderators (atmospheric humidity, water bodies and AOD) and the dominant urban‐warming gradient, and (3) weak, stable secondary drivers (e.g., specific signals of population density and shrublands) adding fine spatial nuance. A key insight is that PCA decomposes the aggregate anthropogenic signal into independent patterns, simultaneously revealing contexts of urban‐linked warming (via heat islands) and moderation (likely via aerosols). We conclude that warming drivers are not monolithic but stem from complex, competing local processes. This necessitates a shift from uniform strategies to spatially‐explicit, driver‐specific interventions. Our framework provides a foundation for prioritizing actions in climate‐vulnerable regions.

Stunting remains a major public health challenge in Indonesia, characterized by significant regional disparities and complex multidimensional determinants. Effective intervention strategies therefore require analytical approaches that are capable of capturing spatial heterogeneity and identifying region-specific vulnerability patterns. This study applies Fuzzy Geographically Weighted Clustering (FGWC) optimized using the Flower Pollination Algorithm (FPA) to map district-level stunting vulnerability and identify priority intervention areas. The analysis covers 514 districts using 21 multidimensional indicators representing health, nutrition, housing conditions, food security, social protection, and demographic characteristics derived from the Central Statistics Agency. The integration of FGWC with FPA enhances clustering performance by incorporating spatial dependence and metaheuristic optimization, enabling the algorithm to produce more stable and geographically sensitive clusters. Cluster validity indices confirm that a four-cluster solution provides the most optimal segmentation of stunting vulnerability. The results reveal distinct regional structures, socioeconomic-driven vulnerability associated with limited asset ownership, high dependence on social assistance and large household size, multidimensional deprivation concentrated primarily in eastern Indonesia, and nutrition-related vulnerability linked to breastfeeding duration and food security. These findings demonstrate that stunting patterns in Indonesia are spatially heterogeneous and influenced by diverse structural factors. The proposed FGWC–FPA framework offers a robust geospatial optimization approach that can support more precise, evidence-based, and region-specific strategies for accelerating stunting reduction.

The ability of migrants to adapt to their new environment in a short period of time while establishing a new life is an important factor affecting the welfare of both individuals and societies. The process of adaptation can determine the quality of life of the individual, reflecting the migrant’s ability to adapt socially, economically and culturally to their new environment. Adaptation in the new life after migration is usually associated with factors such as age, post-migration support from public institutions, post-migration support from relatives, post-migration support from neighbors, and inadequate educational opportunities. In this study, migration from Iran to Van province was analyzed using the geographically weighted regression (GWR) method. The analysis focused on identifying the factors that influence individuals’ responses to the statement “I adapted to my new life in a short time after migration”. According to the analysis results, it is observed that the GWR method gives stronger results. In comparison to the ordinary least squares (OLS) model, the GWR model demonstrates clear superiority across all evaluation criteria, indicating that the GWR model provides a substantially better fit to the data by capturing spatial variability more effectively.

Understanding the spatiotemporal dynamics of wildfire ignitions in mainland Portugal is crucial for effective fire management and risk reduction. This study examines ignition patterns at the municipal scale, where governance structures play a central role in fire prevention and suppression. Three objectives guided the analysis: (i) mapping ignition hotspots, coldspots, and causes (negligence, intentional, and reactivation); (ii) identifying biophysical and human drivers using Multiscale Geographically Weighted Regression (MGWR); and (iii) assessing how these relationships vary across municipalities. Using national wildfire records from 1996 to 2023, we applied Moran's I to evaluate spatial autocorrelation, Getis-Ord Gi∗ statistics to detect hotspots, and MGWR with Golden Section Search bandwidth optimization to model spatial non stationarity. Results indicate that wildfire ignitions are strongly clustered, with persistent hotspots in the northwest and Lisbon regions and coldspots in southern Portugal. Negligence accounted for 47% of ignitions, followed by intentional causes (27%) and reactivations (more than 25%). MGWR revealed spatially varying relationships between ignition patterns and their drivers. Ignition density (adjusted R2=0.774) was positively associated with population density, precipitation, and shrubland cover. Negligent ignitions (R2=0.654) were primarily linked to agroforestry, pasture, and agricultural land uses, as well as terrain ruggedness. Intentional ignitions (R2=0.543) showed negative associations with rugged terrain and positive associations with illiteracy rates in specific municipalities. Reactivations (R2=0.530) were positively associated with firefighter density, reflecting operational allocation patterns rather than causal effects, and were influenced by shrubland cover. These findings highlight the importance of spatial scale and non-stationary modelling for understanding wildfire ignition processes. Municipal scale spatial analysis provides an operationally relevant framework to support targeted fire prevention strategies and evidence-based policies aimed at reducing human caused ignitions and mitigating wildfire risk under increasingly severe fire weather conditions.

ABSTRACT Quantifying the synergistic effects of urban environmental features remains a challenge in geocomputation and urban modeling. Traditional studies often treat the Green View Index (GVI) and Sky View Factor (SVF) as independent variables, overlooking their complex interaction in determining environmental quality. This study proposes a hybrid computational framework coupling semantic segmentation with spatial statistics to construct the Street Environment Composite Metric (SECM) for property valuation. Utilizing the DeepLabV3+ model, we processed massive unstructured Baidu Street View (BSV) imagery to extract high‐precision semantic features. To enhance spatial modeling granularity and mitigate the Modifiable Areal Unit Problem (MAUP), we implemented a fine‐grained segmentation algorithm for street vectors. These computed metrics were then integrated into a Multiscale Geographically Weighted Regression (MGWR) model to reveal the spatial non‐stationarity of housing price determinants in Shanghai's Xuhui District. Computational results demonstrate that the SECM provides a superior fit for capturing the non‐linear interactions between greenery and spatial openness compared to single‐variable models. Specifically, the analysis reveals a threshold effect: Sky View Factor (SVF) contributes to economic value only when supported by a sufficient ecological foundation. In low‐GVI zones, the SECM effectively identifies environmental deficits that suppress property values. This study presents a reproducible workflow for integrating deep learning‐based feature extraction with advanced spatial econometrics, offering a robust tool for fine‐scale urban environmental assessment.

Uncertainty quantification for geospatial prediction models plays a crucial role in evaluating model confidence and supporting informed decision-making. Conformal prediction (CP), as a model-agnostic framework, has recently been introduced into geospatial settings, leading to the development of geospatial conformal prediction (GeoCP). GeoCP considers spatial dependence and incorporates a geographic weighting mechanism into CP to capture spatially varying uncertainty, satisfying the localized exchangeability assumption. In many real-world scenarios, however, spatial processes could exhibit abrupt transitions, such as neighboring regions with distinct land-use types, resulting in significant differences despite close geographic proximity. To address this limitation, we propose geospatial similarity conformal prediction (GeoSIMCP), an extension of GeoCP that jointly considers both geographic distance and feature-space similarity when estimating local uncertainty. We further develop a parameter optimization framework to ensure robust model tuning. Through comprehensive simulation studies under varying spatial structures, we demonstrate the advantages of GeoSIMCP over GeoCP. Additionally, we validate the effectiveness of GeoSIMCP on two real-world prediction tasks—housing prices and PM2.5 concentration—characterized by different spatial processes. Our results highlight the potential of integrating geographic and feature similarity to enhance uncertainty quantification in spatial prediction, offering a more adaptive and reliable framework for geospatial decision-making.

Hazardous seascapes for marine turtles in the Mediterranean sea.

ABSTRACT This study employs statistical analysis to examine the impact of urban morphology on land surface temperature (LST). The findings reveal notable differences in LST across various Local Climate Zone (LCZ) types. The LST of building LCZs is significantly higher than natural LCZs. Additionally, SHapley Additive exPlanation analysis highlights that the influence of urban morphology on LST shows clear seasonal and diurnal patterns. In daytime, Floor Area Ratio is the most significant contributor to temperature variation. In contrast, at nighttime, Normalized Difference Vegetation Index and Building Height play a more dominant role. Interestingly, BH exhibits a bidirectional regulatory effect across different seasonal and diurnal cycles. Furthermore, the application of Multiscale Geographically Weighted Regression (MGWR) confirms that the relationship between urban morphology and LST is subject to spatial heterogeneity. Among the LCZs, LCZ9 shows the greatest variability in factor contribution, indicating the need for tailored interventions, while LCZ4 exhibits more homogeneous effects. This research integrates Light Gradient Boosting Machine with MGWR models, challenging the traditional static approach to studying the urban thermal environment and underscoring the importance of seasonal and spatial variations. The results provide valuable insights for urban thermal environment planning and suggest strategies for mitigating Urban Heat Island effects.

A spatially adaptive empirical Bayes framework with dynamic dispersion parameters for enhanced crash frequency prediction across rural highway networks.

Identifying environmental factors related to motorcyclist crash rates: variable selection using spatial Random Forest with network distance and barriers.

Selenium (Se) is an essential trace element for humans, and agricultural soils are a major source of dietary Se. Therefore, identifying the key environmental drivers of Se in farmland is crucial for evaluating the resource base for Se-rich agriculture and improving human health. Although soil Se distribution and its controlling factors have been widely investigated, quantitative assessments of soil Se in small-scale farmland systems under humid monsoon conditions remain limited. Sampling sites were designed to represent different geological types, soil types, and topography, and 314 farmland topsoil (0–20 cm) samples were collected. Total Se was determined after complete HNO3–HClO4 wet digestion and quantified by HG–AFS (AFS–830), with certified reference materials showing recoveries of 95.3–101.2%. The spatial patterns were mapped using ordinary kriging. Geographically weighted regression (GWR) and Geodetector were used to explore the impact of environmental factors (geological type, precipitation, etc.) on soil Se from both local and overall perspectives. The findings reveal a mean total soil Se of 1.76 mg/kg (95% CI: 1.540–1.974), and 91.40% (n = 287) of soil samples were classified as Se-rich (0.4–3 mg/kg). Organic matter (OM), elevation, slope, and the topographic wetness index (TWI) exhibited non-stationary spatial relationships with Se. The spatial variation trend of precipitation corresponds with the local R2 values between Se and elevation, indicating that precipitation may strengthen the association between elevation and Se distribution. Geological type and rainfall were identified as key driving factors affecting soil Se content within the study area, particularly through their interactions with OM. Overall, the synergistic effects of geological type, precipitation, and OM are responsible for the accumulation of Se in the agricultural soils of Xin’an Town.

In high-density urban cores, pedestrian safety perception is shaped not only by street physical environments but also by pronounced spatial heterogeneity. However, existing studies often rely on global regression or small-sample surveys, making it difficult to simultaneously reveal city-scale regularities and localized mechanisms. Taking Futian District, Shenzhen, as a case study, this study develops an integrated analytical framework that combines street-view imagery, machine learning, and multiscale geographically weighted regression (MGWR) to measure pedestrian safety perception at the city scale and to unpack its spatial mechanisms. The results show that model explanatory power improves markedly after accounting for spatial non-stationarity, indicating strong context dependence in the formation of pedestrian safety perception. MGWR further reveals clear multiscale differentiation across streetscape visual elements: greenery-related elements (e.g., tree and plant) exhibit near-global and consistently positive effects, whereas traffic exposure and interface-related elements (e.g., car, road, and wall) operate more locally, with both the direction and magnitude of their effects varying substantially with neighborhood structure and traffic contexts. These findings suggest that the impacts of individual street elements on pedestrian safety perception are not universally transferable and should be interpreted within a spatial-scale and contextual framework. By integrating machine learning-based prediction with MGWR-based spatial interpretation, this study enables both efficient city-scale measurement and multiscale mechanism identification of pedestrian safety perception, providing empirical support for safety perception-oriented street planning and fine-grained urban design.

Urban sky gardens—elevated green spaces on buildings, encompassing rooftop gardens and podium gardens—are critical to the improvement of urban ecosystem services and functions. Understanding the spatial patterns and the influencing factors of sky gardens is essential for the precise allocation of elevated spaces in urban development. Taking the four central urban districts of Lanzhou in China as the study region, a GIS database of 508 sky gardens was established by identifying high-definition image maps and on-site investigations. The spatial patterns and influencing factors, such as building height, ground-level green area, and population density, were analyzed. The development of sky gardens was also compared in Lanzhou and Guangzhou, China. The distribution of sky gardens in Lanzhou exhibited spatial heterogeneity. Most sky gardens were distributed along the Yellow River. Chengguan District had more sky gardens than Xigu District. In terms of structural characteristics, 82% of sky gardens were rooftop gardens, 73% were located in residential buildings, and 63% were attached to mid- and low-rise buildings. Most sky gardens were one floor, characterized by no public accessibility, a location in high-density plots, and low vegetation coverage. Sky garden area was negatively correlated with building height, ground-level green area, and green plot ratio in sky gardens. There were positive associations between sky garden area and higher plot ratio, building density, and population density based on Multiscale Geographically Weighted Regression. Due to the proper climate conditions and economy, Guangzhou had more sky gardens than Lanzhou. Our study suggests that the utilization of rooftops and podiums is relatively low, and the development of sky gardens exhibits spatial clustering. A suite of optimizing strategies should be implemented to enhance the accessibility and usability of sky gardens.

The spatial distribution of immigrants and associated patterns of residential segregation and integration can manifest not only at the metropolitan scale but also at finer micro-spatial resolutions, reflecting the interaction between path dependence and structural reconfiguration. This article examines the micro-spatial residential patterns of Chinese immigrants in Seoul under institutional and market constraints. Using a Spatial Durbin Model and Multiscale Geographically Weighted Regression, it shows that from 2011 to 2025, immigrant settlements shifted from a monocentric pattern to a polycentric, functionally differentiated, and networked structure. While overall spatial embeddedness is high and segregation remains low, traditional cores such as Guro–Daerim persist. Selective clustering is shaped by path-dependent migrant networks, urban redevelopment policies, and intra-group differentiation, while infrastructure homogenization renders transportation accessibility a background condition. The findings support segmented assimilation theory in high-density East Asian cities and underscore the importance of incorporating immigrant needs into urban policy to promote inclusive integration.

Multi-scale drivers of spontaneous plant diversity in urban residential green spaces across dispersal modes: Implications for ecological planning.

ABSTRACT Multiscale geographically weighted regression (MGWR) is a widely used spatial coefficient of variation model that assumes nearby sample points exhibit similar characteristics. However, geographical processes may exist cross‐border differences in space due to boundary effects, leading MGWR to suffer from precision loss and local coefficient distortion. To address this problem, we propose an MGWR that considers boundary effects (MGWRBE). MGWRBE first uses a priori knowledge to identify boundaries where changes non‐stationarily in a particular spatial process occur. These boundaries are used to construct a boundary‐aware matrix for each independent variable. MGWRBE then reformulates the spatial weight function of MGWR using these matrices, thereby incorporating boundary effects from different spatial processes into the modeling. MGWRBE was evaluated based on both simulated and real datasets. Results showed that compared to MGWR, MGWRBE achieved higher fitting accuracy and more reliable local parameter estimates.

Linking Climate Features to Human Life Expectancy in the United States: Implications for Integrated Climate and Health Policies.

Multiscale geographically weighted regression (MGWR) allows explanatory variables to exhibit spatially nonstationary relationships across different spatial scales. Traditional MGWR, however, assumes that the spatial nonstationarity of each variable is influenced by a single spatial scale. This study relaxes that assumption, proposing that spatial nonstationarity is not only variable-specific but that even a single variable can be shaped by multiple spatial scales simultaneously. The proposed extension, termed double multiscale geographically weighted regression (M2GWR), incorporates a linear multiscale kernel function into the MGWR. An efficient coefficient estimation technique for M2GWR is developed by leveraging advancements in algorithms and hardware. Additionally, two pseudo-bandwidth parameters are introduced as diagnostics for spatial nonstationarity, allowing for the detection and interpretation of variable-specific multiscale effects. Simulation experiments demonstrate that M2GWR consistently outperforms MGWR in coefficient estimation accuracy (mean absolute error [MAE]) and model fit (corrected Akaike’s information criterion [AICc]). The pseudo-bandwidth parameters effectively differentiate multiscale patterns among variables and, to some extent, capture the combined multiscale influences acting on individual variables. Computational results also reveal substantial improvements in efficiency. Real-world applications further validate M2GWR’s effectiveness, showing gains in R 2, adjusted R 2, AICc, local condition numbers, and reduced spatial autocorrelation in residuals. Overall, our findings position M2GWR as a robust and efficient framework for modeling complex spatially nonstationary relationships.

In Southwest China’s multi-ethnic mountainous regions, fragmented terrain has preserved numerous traditional villages. Yet the progression of urbanization and tourism has eroded cultural heritage in these villages, rendering the preservation-development equilibrium an urgent challenge. Crucially, the specific factors affecting cultural inheritance in local villages require further investigation. To address this gap, this study applied Cultural Ecosystem Theory to evaluate the Cultural Inheritance Level (CIL) of 43 villages in Leishan County, Guizhou. The Multiscale Geographically Weighted Regression (MGWR) model was employed to identify driving factors and quantify their spatially varying impacts. The findings revealed significant regional spatial differentiation in the CIL. Notably, areas with rugged terrain were more affected by positive factors—cultural heritage protection policies. The pressures of mass tourism are negatively correlated with CIL, with amplified effects in tourism-developed regions. This study delivers a CIL assessment framework and targeted policy recommendations for cultural heritage protection within this regional context.