Spatial Regression Models Research Articles

Federal Nutrition Assistance for Young Children: Underutilized and Unequally Accessed

The federal Child and Adult Care Food Program (CACFP) improves nutrition and reduces food insecurity among young children by helping cover the food costs for child care providers and families. This nationwide study evaluated the extent and predictors of the CACFP's utilization among licensed child care centers to identify opportunities for expanding CACFP nutrition support. Administrative data from the CACFP and child care licensing agencies in 47 states and District of Columbia were compiled and geocoded for 93,227 licensed child care centers. CACFP participation was predicted using a multivariable Bayesian spatial logistic regression model in the sample of low-income areas to target CACFP eligible child care centers. Data were collected in 2020-2021 and analyzed in 2022. Of all licensed child care centers, 36.5% participated in the CACFP, ranging from 15.2% to 65.3% across states; when restricted to low-income areas, 57.5% participated (range, 15.7%-85.7%). Income differences did not explain the large variation in CACFP participation rates across states. Having at least three CACFP sponsoring agencies per state predicted a 38% higher probability of CACFP participation (OR=1.38; 95% Credible Interval=1.08-1.78). Currently CACFP participation rates among licensed child care centers point to program underutilization and unequal access, particularly in some states and regions. Work at the federal and state levels is warranted to expand participation in the program, above all in low-income areas, so that more young children could eat healthfully with the CACFP.

Spatial association between socio-economic health service factors and sepsis mortality in Thailand.

Sepsis is a significant global health issue causing organ failure and high mortality. The number of sepsis cases has recently increased in Thailand making it crucial to comprehend the factors behind these infections. This study focuses on exploring the spatial autocorrelation between socio-economic factors and health service factors on the one hand and sepsis mortality on the other. We applied global Moran's I, local indicators of spatial association (LISA) and spatial regression to examine the relationship between these variables. Based on univariate Moran's I scatter plots, sepsis mortality in all 77 provinces in Thailand were shown to exhibit a positive spatial autocorrelation that reached a significant value (0.311). The hotspots/ high-high (HH) clusters of sepsis mortality were mostly located in the central region of the country, while the coldspots/low-low (LL) clusters were observed in the north-eastern region. Bivariate Moran's I indicated a spatial autocorrelation between various factors and sepsis mortality, while the LISA analysis revealed 7 HH clusters and 5 LL clusters associated with population density. Additionally, there were 6 HH and 4 LL clusters in areas with the lowest average temperature, 4 HH and 2 LL clusters in areas with the highest average temperature, 8 HH and 5 LL clusters associated with night-time light and 6 HH and 5 LL clusters associated with pharmacy density. The spatial regression models conducted in this study determined that the spatial error model (SEM) provided the best fit, while the parameter estimation results revealed that several factors, including population density, average lowest and highest temperature, night-time light and pharmacy density, were positively correlated with sepsis mortality. The coefficient of determination (R2) indicated that the SEM model explained 56.4% of the variation in sepsis mortality. Furthermore, based on the Akaike Information Index (AIC), the SEM model slightly outperformed the spatial lag model (SLM) with an AIC value of 518.1 compared to 520.

Connectivity-based spatial patterns and factors influencing international container multimodal hubs in China under the Belt and Road initiative

The location and layout of container multimodal hubs have received much research attention under the Belt and Road Initiative. To explore the spatial distribution pattern of international container multimodal hubs (ICMHs) and its influencing factors in China, this paper measures the domestic connectivity, international land connectivity, international rail-sea multimodal connectivity and comprehensive connectivity of ICMHs. Spatial analysis methods are used to explore the spatial patterns of ICMH connectivity, and then spatial regression models are applied to detect those key factors that influence its distribution. Results reveal that ICMH connectivity has significant spatial differences and obvious distribution trends across China and that such connectivity gradually decreases from east to west. The significant spatial agglomeration characteristic of connectivity is also highlighted. Specifically, a high connectivity is concentrated in the east and close to exit ports (land exit-ports and seaports), whereas a low connectivity is distributed in the expansive western areas. Results of the spatial regression also show that each factor has significant spatial heterogeneity with respect to connectivity distribution, especially freight turnover volume and investment in transportation. This study provides support for the layout of logistics hubs in multimodal networks and enriches the application of the connectivity method.

Estimation of Factors Affecting Stunting Cases in West Java in 2021 Using Negative Binomial Spatial Regression

Stunting is a childhood growth and development disorder characterized by below-normal height. West Java, with its stunting rate of 24.5 percent, is one of the provinces included in the top 12 priority provinces in implementing the National Action Plan to Accelerate Stunting. Stunting cases are count data and their occurrence is rare. The analysis for the count data is Poisson regression with the assumption that equidispersion must be met. One way to overcome overdispersion is to use negative binomial regression. This study aimed to determine predictors/factors affecting stunting cases in West Java province in 2021 using negative binomial spatial regression. The data in this study comes from the publication of the West Java Health Service and the West Java Central Statistics Agency in 2021 with districts/cities as the object of observation. There is a spatial effect in the stunting data, so the spatial regression model is suitable. The results show that there is an overdispersion in the Poisson regression. The spatial effect test shows that there is a spatial dependence on the response variable and some predictors. The negative spatial autoregressive binomial is the best model with the lowest AIC value. The factors that have a significant effect are the percentage of infants aged less than six months who are breastfed, the percentage of food processing establishments that meet the requirements, and the percentage of infants with low birth weight.

Spatial distribution analysis and driving factors of traditional villages in Henan province: a comprehensive approach via geospatial techniques and statistical models

Traditional villages are repositories for preserving human artifacts and cultural heritage. An investigation of the spatial distribution characteristics and factors influencing traditional villages in provincial administrative regions can provide new insights regarding the protection of traditional villages and rural development. This study focused on 275 traditional villages in Henan Province. Using ArcGIS and GeoDa software, we analysed the spatial autocorrelation and heterogeneity of the nearest neighbour index, Gini coefficient, Moran’s I, and kernel density of the villages. Additionally, in conjunction with the Python sklearn library and GeoDetector, 15 indicators were selected to construct a decision tree model, spatial lag regression model, and geographic detector. Then the influence and interaction mechanisms of each indicator were analysed. The results revealed that (1) the spatial distribution of traditional villages in Henan Province was clustered and uneven, with a spatial layout comprising “3 high-density areas + 1 medium-density belt”; (2) overall, the number of traditional villages was negatively correlated with altitude, slope, rainfall, population density, proportion of the minority population, and historical-cultural intensity; and (3) the decision tree model results demonstrated that the selected 15 indicators had good predictive ability and that population density was particularly important. The spatial lag regression model results showed that the spatial distribution of traditional villages was positively correlated with distance from rivers, urbanization rate, and tourism resources, and negatively correlated with population density, per capita GRP, historical-cultural intensity, and NDVI. (4) The GeoDetector results indicated that historical-cultural intensity and population density were the two factors with the most significant explanatory power for the spatial differentiation of traditional villages in Henan Province. In terms of interactive factors, population density cap population was the strongest interactive driving force, followed by population cap historical-cultural intensity.

Spatial Association and Modeling of Infant Mortality in Thailand, 2020

Infant mortality remains a pressing public health challenge globally. Despite advancements in healthcare, glaring disparities persist, as exemplified in Thailand. This study explored spatial variations in infant mortality rates (IMRs) across Thai provinces, integrating socio-economic, demographic, and health factors. Using data from national databases, we employed univariate and bivariate Local Indicators of Spatial Association (LISA) analyses to visualize spatial disparities, and Moran's I statistic assessed global spatial autocorrelation. Spatial regression models, including Ordinary Least Squares (OLS), Spatial Lag Model (SLM), and Spatial Error Model (SEM), analyzed the associations between IMRs and determinants. Our findings revealed stark IMRs disparities, especially in provinces like Phitsanulok, Narathiwat, and Songkhla. The SEM emerged as the most fitting model, given the data's spatial autocorrelation (R-Squared = 0.46). Crucial factors such as community organization strength, nighttime light, and exclusive breastfeeding were significantly linked to IMRs. Additionally, provinces like Phra Nakhon Si Ayutthaya and Rayong underscored socio-economic challenges, emphasizing the importance of tailored interventions. This study offers valuable insights for crafting targeted strategies, underscoring the pivotal role of geospatial techniques in shaping public health policies in Thailand.

Unveiling Spatial Associations between COVID-19 Severe Health Index, Racial/Ethnic Composition, and Community Factors in the United States.

There are limited efforts to incorporate different predisposing factors into prediction models that account for population racial/ethnic composition in exploring the burden of high COVID-19 Severe Health Risk Index (COVID-19 SHRI) scores. This index quantifies the risk of severe COVID-19 symptoms among a county's population depending on the presence of some chronic conditions. These conditions, as identified by the Centers for Disease Control and Prevention (CDC), include Chronic Obstructive Pulmonary Disease (COPD), heart disease, high blood pressure, diabetes, and obesity. Therefore, the objectives of this study were (1) to investigate potential population risk factors preceding the COVID-19 pandemic that are associated with the COVID-19 SHRI utilizing non-spatial regression models and (2) to evaluate the performance of spatial regression models in comparison to non-spatial regression models. The study used county-level data for 3107 United States counties, utilizing publicly available datasets. Analyses were carried out by constructing spatial and non-spatial regression models. Majority White and majority Hispanic counties showed lower COVID-19 SHRI scores when compared to majority Black counties. Counties with an older population, low income, high smoking, high reported insufficient sleep, and a high percentage of preventable hospitalizations had higher COVID-19 SHRI scores. Counties with better health access and internet coverage had lower COVID-19 SHRI scores. This study helped to identify the county-level characteristics of risk populations to help guide resource allocation efforts. Also, the study showed that the spatial regression models outperformed the non-spatial regression models. Racial/ethnic inequalities were associated with disparities in the burden of high COVID-19 SHRI scores. Therefore, addressing these factors is essential to decrease inequalities in health outcomes. This work provides the baseline typology to further explore many social, health, economic, and political factors that contribute to different health outcomes.

Investigating the Direct and Spillover Effects of Urbanization on Energy-Related Carbon Dioxide Emissions in China Using Nighttime Light Data

Cities are the main emission sources of the CO2 produced by energy use around the globe and have a great impact on the variation of climate. Although the implications of urbanization and socioeconomic elements for carbon emission have been extensively explored, previous studies have mostly focused on developed cities, and there is a lack of research into naturally related elements due to the limited data. At present, remote sensing data provide favorable conditions for the study of large-scale and long-time series. Also, the spillover mechanism of urbanization effects on the discharge of carbon has not been fully studied. Therefore, it is necessary to distinguish the types of influence that various urbanization factors have on emissions of CO2. Firstly, this study quantifies the urban CO2 emissions in China by utilizing nighttime lighting images. Then, the spatio-temporal variations and spatial dependence modes of CO2 emissions are explored for 284 cities in China from 2000–2018. Finally, the study further ascertains that multi-dimensional urbanization, socio-economic and climate variables affect the discharge of carbon using spatial regression models. The results indicate that CO2 emissions have a remarkable positive spatial autocorrelation. Urbanization significantly increases CO2 emissions, of which the land urbanization contribution towards CO2 emissions is the most important in terms of spillover effects. Specifically, the data on urbanization’s direct effects reveal that CO2 emissions will increase 0.066%when the urbanization level of a city rises 1%, while the spillover effect indicates that an 0.492% emissions increase is associated with a 1% rise of bordering cities’ average urbanization level. As for the socio-economic factors, population density suppresses CO2 emissions, while technological levels boost CO2 emissions. The natural control factors effect a remarkable impact on CO2 emissions by adjusting energy consumption. This study can provide evidence for regional joint prevention in urban energy conservation, emission reduction, and climate change mitigation.

Socioeconomic and environmental determinants of foot and mouth disease incidence: an ecological, cross-sectional study across Iran using spatial modeling

Foot-and-mouth disease (FMD) is a highly contagious animal disease caused by a ribonucleic acid (RNA) virus, with significant economic costs and uneven distribution across Asia, Africa, and South America. While spatial analysis and modeling of FMD are still in their early stages, this research aimed to identify socio-environmental determinants of FMD incidence in Iran at the provincial level by studying 135 outbreaks reported between March 21, 2017, and March 21, 2018. We obtained 46 potential socio-environmental determinants and selected four variables, including percentage of population, precipitation in January, percentage of sheep, and percentage of goats, to be used in spatial regression models to estimate variation in spatial heterogeneity. In our analysis, we employed global models, namely ordinary least squares (OLS), spatial error model (SEM), and spatial lag model (SLM), as well as local models, including geographically weighted regression (GWR) and multiscale geographically weighted regression (MGWR). The MGWR model yielded the highest adjusted {R}^{2} of 90%, outperforming the other local and global models. Using local models to map the effects of environmental determinants (such as the percentage of sheep and precipitation) on the spatial variability of FMD incidence provides decision-makers with helpful information for targeted interventions. Our findings advocate for multiscale and multidisciplinary policies to reduce FMD incidence.

District-level heterogeneity in overweight or obesity among women of reproductive age: A geo-spatial analysis in India.

Globally by 2030, 38% of the world's population would be overweight, and another 20% would be obese. This has led to rising concerns regarding how swiftly and substantially the world is moving towards this epidemic of "globesity". India too is facing an increased burden of overweight and obese population. The changing dietary patterns are significantly associated with the increasing prevalence of overweight/obesity and related complications, especially among women. Hence, the present study aims to observe the spatial patterns of overweight or obesity among women in reproductive age group in India and factors associated with it. The study analyzed data from a cross-sectional nationwide household survey, i.e. National Family Health Survey (NFHS-4), 2015-16. The primary outcome variable of this study was overweight/obesity among reproductive-age women, which was measured through the body mass index (BMI) of the women. Bivariate and multivariate logistic regression analysis was used to analyze the data. Additionally, for spatial analysis in terms of overweight/obesity among women in India, univariate and bivariate Moran's I index measurements were used along with the usage of spatial regression models. The value of spatial-autocorrelation for overweight or obese was 0.64, which depicts the moderately high prevalence of the overweight/obesity coverage over districts of India. The overall prevalence overweight/obesity among women in India is around 25% and higher proportion of women from urban areas (37.8%), and non-poor (33.4%) economic group reported to be overweight or obese. From spatial lag model, the lag coefficient was found to be 0.28, implying that a change in the prevalence of overweight/obesity among women in a certain district may statistically lag the prevalence of overweight/obesity by 28% in the neighbouring districts. There were significantly high clustering of overweight/obese women and non-poor wealth quintiles in 132 districts, mainly from states of Punjab, Haryana, Gujarat, Maharashtra, Kerala, Tamil Nadu, Karnataka and Andhra Pradesh. Additionally, there was high-high clustering of overweight/obese women and those who ever had caesarean in 82 districts, mostly from Kerala, Tamil Nadu, Andhra Pradesh and Karnataka. The spatial patterns on the prevalence of overweight and obesity in India show that the women belonging to the southern states' districts are more overweight or obese in comparison to other states. The determinants like older age, higher education, urban residence, higher economic status are the key factors contributing to the prevalence of overweight or obesity among women in the reproductive age group. The study concludes and recommends an urgent need of interventions catering to urban women belonging to higher socio-economic status, to reduce the risks of health consequences due to overweight and obesity.

ANALYZING THE RETRIEVAL ACCURACY OF OPTICALLY ACTIVE WATER COMPONENTS FROM SATELLITE DATA UNDER VARYING IMAGE RESOLUTIONS

Abstract. Water quality monitoring has a key role in maintaining a sustainable ecosystem and environmental health. To ensure consistent monitoring, remote sensing provides regular data acquisition with varying spatial resolutions. However, more accurate, and effective solutions can be achieved by integrating remote sensing data with in-situ measurements. This study investigates the integration of in-situ measurements with satellite data, which have different spectral and spatial resolutions, using linear and exponential regression models for four optically active components in the Gulf of Izmit. In this context, Sentinel-2 (S2) and PlanetScope SuperDove (PS) multispectral images, which were acquired on the same date, were used for the comparative analysis of the accurate mapping of chlorophyll-a (Chl-a), turbidity, Secchi disk depth (SDD) and total suspended matter (TSM) water quality parameters combined with simultaneously collected in-situ measurements. The models were evaluated using validation data, along with visual comparison, to assess their accuracy. The results indicate that, overall, exponential models provide more accurate results than linear models, except for the SDD parameter. Furthermore, models created with S2 data demonstrate better performance in retrieving water quality parameters for Chl-a, turbidity, and TSM, with R2 values of 0.71, 0.84, and 0.91, respectively. The linear model created with PS data stands out in the accurately mapping of SDD parameter. Nevertheless, the spatial distribution of these parameters using both satellite dataset exhibits a similar pattern throughout the gulf, which is under threat from significant terrestrial pollution sources, particularly in the eastern part.

Key Factors Influencing Vehicle Availability Associated with the User Waiting Time of Demand Responsive Transport (DRT) for People with Disabilities

Demand Responsive Transport (DRT) services for the safe and comfortable travel of people with disabilities have long waiting times and significant variations in time and space. Due to these characteristics, user dissatisfaction with DRT services continues to increase. This study aimed to identify factors that affect vehicle availability, which is closely related to waiting time, and to prepare measures to solve problems in waiting time. The data used for the analysis was composed of spatial units, so Moran's I factor was used to test for spatial dependence. After confirming the presence of spatial dependence, spatial regression models optimized for spatial data analysis were estimated. As a result of predicting the model using five independent variables, four variables, Medical Infrastructure Concentration Index (MICI), Disabled Population Concentration Index (DPCI), Depot Size, and Demand Variability Index (DVI), were found to be statistically significant at a 95% confidence level, except for congestion level. The analytical results indicate that the frequency of vehicle operation is low in areas with a low population of the disabled, little medical infrastructure related to the disabled, and irregular DRT demand, which may affect increasing user waiting time. Therefore, supply-oriented policies to increase the number of available vehicles, such as adding more depots or deploying more vehicles, will contribute significantly to reducing user waiting time in these areas.

Ecology and space in the COVID-19 epidemic diffusion: a multifactorial analysis of Italy’s provinces

ABSTRACT This study aimed to identify the ecological and spatial correlates of the COVID-19 secondary transmission mechanisms. The dataset comprised cross-province daily COVID-19 cases recorded in Italy from 25 February to 26 May 2020. Several physical and socio-economic variables and a broad set of spatial matrices were employed to explain epidemic spreading. First, the serial transmission interval was identified through autoregressive models. Second, the impact of the ecological and spatial variables on the epidemic’s secondary transmission was analysed by means of ordinary least squares and spatial regression models. The study found that, against widespread belief, the epidemic diffusion was higher in the more affluent and developed territorial units. As for the spatial component, a secondary transmission spatial spillover emerged. Contiguity and distance but also mobility fluxes proved to be significant in this transmission. This study’s contribution is methodological and empirical. As for methodology, it shows that spatial models can be used not only to identify epidemic localisation but also to predict epidemic transmission, provided the problem of the multidirectional nature of spatial dependence is adequately dealt with. Moreover, this study shows that no satisfactory understanding of the epidemic would be achieved without considering both ecological and spatial determinants. As for the empirical aspect, it was found that the main contribution to epidemic spreading came from the anthropologic context rather than the physical one. In particular, direct and indirect indicators of mobility, transport and social interaction emerged as crucial for the epidemic transmission.

Spatial Dependence of Poverty Rate and Socio-Economic Indicators in Nigeria: An Empirical Analysis

This study investigates the spatial dependence between the poverty rate and various socio-economic indicators in Nigeria. The analysis is based on a dataset comprising unique geographic identifiers and the poverty rate along with other relevant variables. Descriptive statistics reveal that the poverty rate exhibits moderate variability with an average of 4.1240. The correlation analysis shows significant relationships between the poverty rate and household size as well as income level, indicating that larger households and higher incomes are associated with higher and lower poverty rates, respectively. Spatial regression models, including Spatial Autoregressive (SAR), Spatial Error (SEM), Spatial Durbin (SDM), and Spatial Autoregressive Conditional (SAC) models, are employed to explore the spatial dependence. Results indicate the presence of spatial clustering and positive autocorrelation in the poverty rate, as indicated by the Moran's I index with a value of 0.3579 (p-value = 0.0012). However, tests for spatial heteroscedasticity do not reveal significant departures from the assumption of constant error variance. The findings suggest that spatial factors play a crucial role in explaining the poverty rate in Nigeria. The positive spatial autocorrelation indicates the presence of localized poverty clusters, emphasizing the importance of considering spatial effects in policy formulation and targeted interventions. The significant relationships between the poverty rate and household size and income level underscore the need for comprehensive strategies to address these socio-economic indicators for poverty reduction.

Agricultural intensity interacts with landscape arrangement in driving ecosystem services

Agricultural intensification has enhanced productivity but also led to enormous ecosystem service and biodiversity losses. Strategic spatial landscape design could counteract this trend, but, the scientific understanding of how ecosystem services respond to agricultural practices on one hand and land use composition and configuration on the other is not complete. This study aims to methodically explore how the effect of landscape layout settings on ecosystem services depends on the intensity of agricultural practices in their surroundings. Using the Netherlands as a case study, we used spatial regression models to analyze how agricultural management intensity affects the relationship between spatial composition and configuration metrics and ecosystem service indicators. We found that the effect of large shares of agricultural land use on species richness, pollination and landscape appreciation was increasingly negative with amplified intensity of agricultural practices. With higher agricultural intensity in the surroundings, the positive effects of well-connected natural vegetation on species richness were impaired. In contrast, the negative effects of high-intensity agriculture on pollination service were be buffered well through high shares of natural grassland vegetation. Water-quality related indicators were less affected by variation in spatial metrics and agricultural intensity. The main interactions between agricultural intensity and the spatial metrics were robust at varying scales. Our analysis suggests that both low- and high-intensity agriculture can have a place in future sustainable agricultural systems, provided they are integrated in the appropriate spatial layout. Explicitly addressing farming practices in connection to local spatial settings can improve both landscape planning and ecosystem service modelling.

Spillover effects of internal audit effectiveness among commercial banks

This study explores the multifaceted landscape of factors influencing the effectiveness of internal audit within the context of Vietnamese commercial banks. While a plethora of literature exists on the determinants of internal audit effectiveness, an unexplored dimension is the examination of spillover effects amongst companies operating within the same industry. This research addresses this gap by investigating not only the factors that impact the effectiveness of internal audit in Vietnamese commercial banks but also the inter-bank spillover effects of such effectiveness. Utilizing data obtained from 24 commercial banks listed on the Vietnam Stock Exchange and Vietnam Upcom (The Unlisted Public Company Market), we employ a Bayesian spatial regression model (a robust analytical tool) to conduct comprehensive analyses of the interrelationships and co-movements of internal audit effectiveness across these banks. Our findings offer empirical support for the positive effects of key determinants, namely, the "Competence of the IA Department," "IA Independence," and "Management Training Ground," on the effectiveness of internal audit in Vietnamese commercial banks. Notably, our results also validate the significant positive inter-bank spillover effect of internal audit effectiveness. This underscores the importance for banks to collectively prioritize measures aimed at enhancing the quality of internal audit, including the establishment of a joint internal audit association and the organization of collaborative training initiatives for internal audit personnel.

Unveiling the link between rainfall, temperature, and childhood undernutrition in Bangladesh using spatial analysis

Child undernutrition is still a major public health concern in Bangladesh in spite of significant decline in the last few years. Climate change may impact the prevalence of undernutrition and its geographical variability through food security and recurring outbreaks of disease, as well as impede efforts to reduce the undernutrition burden. This study aims to evaluate rainfall and temperature associated with childhood malnutrition in Bangladesh. A spatial mixed effect logistic regression model was used to estimate the association between undernutrition (measured by the composite index of anthropometric failure) and residential area-level rainfall and temperatures, controlling for covariates and spatial effects of residential locations using national data from the 2017 to 2018 Bangladesh Demographic and Health Survey. Our findings indicate that an increase in rainfall was associated with increased odds of undernutrition [adjusted odds ratio (AOR) 1.15, 95% confidence interval (CI) 1.07–1.24], whereas an increase in temperature was associated with decreased odds of undernutrition (AOR 0.86, 95% CI 0.80–0.93). We also found statistically significant structured and unstructured spatial variations in undernutrition, indicating locational effects. Investing in infrastructure and education could be potential strategies for mitigating the negative effects of climate change. High-risk regions in terms of climatic change and malnutrition could be given priority for intervention implementation.

Estimation of non-technical loss rates by regions

Identifying vulnerable regions to non-technical losses allows more assertive combat against them. In this context, this paper presents a spatiotemporal methodology composed of two modules, spatial and temporal, to assist distribution companies in action planning to decrease the rates of non-technical losses by region. The spatial module contains a neighborhood structure based on the similarity among small regions named “neighborhood by the similarity of attributes”, which improves the characterization of non-technical losses actions performed by end-consumers. That neighborhood structure is incorporated as an input parameter into a hierarchical spatial autoregressive regression model to represent the relationships between inhabitants. On the other hand, the temporal module uses a linear mixed-effects model to consider future values that are subject to the actions of consumers or distribution companies. The proposed methodology is applied to a medium-sized city with approximately 200,000 inhabitants, considering the inspections carried out by a Brazilian distribution utility. The proposal identified the future non-technical loss state in all the city’s regions with values greater than 69% of the success rate in identifying NTL to residential, commercial, and industrial consumer classes.

Putting MARS into space. Non‐linearities and spatial effects in hedonic models

AbstractMultivariate Adaptive Regression Spline (MARS) is a simple and powerful non‐parametric machine learning algorithm that automatizes the selection of non‐linear terms in regression models. In this study, we propose using MARS in a spatial regression framework to account for potential non‐linearities and spatial effects in spatial regression models. Using a relatively large data set of 17,000 dwellings in St. Petersburg (Russia), we examine how this algorithm works. The empirical evidence shows that most explanatory variables in the spatial regression model—including the spatial lag of the dependent variable—have a non‐linear impact on the asking prices of dwellings.

Deep trip generation with graph neural networks for bike sharing system expansion

Bike sharing is emerging globally as an active, convenient, and sustainable mode of transportation. To plan successful bike-sharing systems (BSSs), many cities start from a small-scale pilot and gradually expand the system to cover more areas. For station-based BSSs, this means planning new stations based on existing ones over time, which requires prediction of the number of trips generated by these new stations across the whole system. Previous studies typically rely on relatively simple regression or machine learning models, which are limited in capturing complex spatial relationships. Despite the growing literature in deep learning methods for travel demand prediction, they are mostly developed for short-term prediction based on time series data, assuming no structural changes to the system. In this study, we focus on the trip generation problem for BSS expansion, and propose a graph neural network (GNN) approach to predicting the station-level demand based on multi-source urban built environment data. Specifically, it constructs multiple localized graphs centered on each target station and uses attention mechanisms to learn the correlation weights between stations. We further illustrate that the proposed approach can be regarded as a generalized spatial regression model, indicating the commonalities between spatial regression and GNNs. The model is evaluated based on realistic experiments using multi-year BSS data from New York City, and the results validate the superior performance of our approach compared to existing methods. We also demonstrate the interpretability of the model for uncovering the effects of built environment features and spatial interactions between stations, which can provide strategic guidance for BSS station location selection and capacity planning.