Recently, numerous machine learning methods have been effectively applied to uncover spatial relationships between health risk factors and health outcomes. However, traditional machine learning methods often fail to address confounding bias, which arises when a common factor simultaneously influences both the treatment and the outcome – a challenge frequently encountered in observational studies. Deep learning-based causal inference models seek to mitigate confounding bias by learning balanced representations of covariates between treated and control groups, thereby reducing the dependence of treatment assignment on covariates. This enables accurate estimation of causal effects on health outcomes. Moreover, distinct geospatial patterns of risk exposure and health outcomes are common in many chronic diseases. Therefore, developing a spatially-aware causal inference model is essential for guiding geospatial health interventions. Here, we propose SpatialCausal, a spatially-aware deep learning-based causal inference model that explicitly integrates spatial, non-spatial, and unmeasured confounders, enabling accurate estimation of spatially-aware causal effects. We demonstrate the effectiveness of our approach through an application to Out-of-Hospital Cardiac Arrest survival outcome prediction. Our method surpasses state-of-the-art approaches and exhibits robust adaptability to various geospatial disease scenarios, making it a valuable tool for spatially-aware causal effect estimation in health geography.

Understanding the relationship between risk factors, geospatial patterns, and disease outcomes is essential in health geography research. These relationships can inform the implementation of healthcare and public health strategies to improve health outcomes. To accurately uncover such complex relationships, it is necessary to have a predictive model capable of integrating both health variables and spatial information to forecast health outcomes, along with a tool to interpret and reveal the patterns identified by this model. We developed a Spatial Counterfactual Explainable Deep Learning model (SpaCE), comprising a spatially explicit health outcome predictor and a prototype-guided counterfactual explanation. The SpaCE model unifies geospatial and health variables to improve predictions and generates hypothetical examples with minimal changes but opposite outcomes. Using these counterfactuals, SpaCE assesses the impact of each variable in different spatial contexts. We evaluated the model for predicting cardiac arrest survival outcomes. With a 0.682 AUCROC score, the SpaCE exceeds baseline models by 10.2%. Further analysis also reveals that the geospatial context significantly affects how various risk factors affect the survival outcomes of patients. Overall, the SpaCE model significantly improves predictive accuracy and explainability. It provides targeted interventions at both individual and geographic levels, and the cardiac arrest case study shows its high adaptability to various disease scenarios.

With the widespread development of multiple unmanned aerial vehicle (multi-UAV) systems, flocking motion has become a common but essential application in UAV swarms. However, most existing approaches are rule-based and only valid for specific scenarios, which are limited in adaptability and scalability. In this paper, we present a novel framework termed attention-biased deep deterministic policy gradient (ABDDPG), which combines the multi-agent deep deterministic policy gradient (MADDPG) algorithm and attention-biased Transformer (ABTransformer). First, we encode the comprehensive environmental features observed by each UAV as input. Then, we introduce ABTransformer into the actor-critic network architecture. This allows the model to fully utilize each drone’s positional relationship with surrounding objects (including obstacles and other drones), endowing it with the ability to allocate attention reasonably. Furthermore, we utilize LoRA for pretraining and fine-tuning to further improve performance and training efficiency. This attention-biased reinforcement learning model enables each UAV in the swarm to learn and allocate local attention autonomously, guaranteeing internal communication stability and flocking motion task completion. The experimental results demonstrate that ABDDPG outperforms previous methods in terms of arrival rate, training speed, and inference efficiency, and is robust to different scenarios.

Image geolocalization is a fundamental yet challenging task, aiming at inferring the geolocation on Earth where an image is taken. State-of-the-art methods employ either grid-based classification or gallery-based image-location retrieval, whose spatial generalizability significantly suffers if the spatial distribution of test images does not align with the choices of grids and galleries. Recently emerging generative approaches, while getting rid of grids and galleries, use raw geographical coordinates and suffer quality losses due to their lack of multi-scale information. To address these limitations, we propose a multi-scale latent diffusion model called LocDiff for image geolocalization. We developed a novel positional encoding-decoding framework called Spherical Harmonics Dirac Delta (SHDD) Representations, which encodes points on a spherical surface (e.g., geolocations on Earth) into a Hilbert space of Spherical Harmonics coefficients and decodes points (geolocations) by mode-seeking on spherical probability distributions. We also propose a novel SirenNet-based architecture (CS-UNet) to learn an image-based conditional backward process in the latent SHDD space by minimizing a latent KL-divergence loss. To the best of our knowledge, LocDiff is the first image geolocalization model that performs latent diffusion in a multi-scale location encoding space and generates geolocations under the guidance of images. Experimental results show that LocDiff can outperform all state-of-the-art grid-based, retrieval-based, and diffusion-based baselines across 5 challenging global-scale image geolocalization datasets, and demonstrates significantly stronger generalizability to unseen geolocations.

Spatial representation learning (SRL) aims at learning general-purpose neural network representations from various types of spatial data (e.g., points, polylines, polygons, networks, images, etc.) in their native formats. Learning good spatial representations is a fundamental problem for various downstream applications such as species distribution modeling, weather forecasting, trajectory generation, geographic question answering, etc. Even though SRL has become the foundation of almost all geospatial artificial intelligence (GeoAI) research, we have not yet seen significant efforts to develop an extensive deep learning framework and benchmark to support SRL model development and evaluation. To fill this gap, we propose TorchSpatial, a learning framework and benchmark for location (point) encoding,which is one of the most fundamental data types of spatial representation learning. TorchSpatial contains three key components: 1) a unified location encoding framework that consolidates 15 commonly recognized location encoders, ensuring scalability and reproducibility of the implementations; 2) the LocBench benchmark tasks encompassing 7 geo-aware image classification and 10 geo-aware imageregression datasets; 3) a comprehensive suite of evaluation metrics to quantify geo-aware models’ overall performance as well as their geographic bias, with a novel Geo-Bias Score metric. Finally, we provide a detailed analysis and insights into the model performance and geographic bias of different location encoders. We believe TorchSpatial will foster future advancement of spatial representationlearning and spatial fairness in GeoAI research. The TorchSpatial model framework and LocBench benchmark are available at https://github.com/seai-lab/TorchSpatial, and the Geo-Bias Score evaluation framework is available at https://github.com/seai-lab/PyGBS.

Geolocating precise locations from images presents a challenging problem in computer vision and information retrieval. Traditional methods typically employ either classification-dividing the Earth's surface into grid cells and classifying images accordingly, or retrieval-identifying locations by matching images with a database of image-location pairs. However, classification-based approaches are limited by the cell size and cannot yield precise predictions, while retrieval-based systems usually suffer from poor search quality and inadequate coverage of the global landscape at varied scale and aggregation levels. To overcome these drawbacks, we present Img2Loc, a novel system that redefines image geolocalization as a text generation task. This is achieved using cutting-edge large multi-modality models (LMMs) like GPT-4V or LLaVA with retrieval augmented generation. Img2Loc first employs CLIP-based representations to generate an image-based coordinate query database. It then uniquely combines query results with images itself, forming elaborate prompts customized for LMMs. When tested on benchmark datasets such as Im2GPS3k and YFCC4k, Img2Loc not only surpasses the performance of previous state-of-the-art models but does so without any model training. A video demonstration of the system can be accessed via this link https://drive.google.com/file/d/16A6A-mc7AyUoKHRH3_WBRToRC13sn7tU/view?usp=sharing

In this paper, we consider the real-time pricing problem for a small scale local power supplier (LPS) in a smart energy community. The LPS supplies power to the residential users (RUs) in a local area and sells the remaining power to the main grid. Since the selling price to the main grid is relative low, LPS intends to sell more power to the RUs with an appropriate price. The LPS determines the price based on the proposed pricing scheme to maximize its revenue. The price is informed to RUs through the communication infrastructure. According to the announced price of LPS, each RU schedules its power consumption to maximize its utility. We model the interactions between the local power supplier and all users as a one-leader multi-followers Stackelberg game, where the LPS acts as the leader and RUs act as the followers. To address this problem, a distributed algorithm based on information exchange between the LPS and RUs is proposed. Simulation results show that the distributed algorithm converges to the Stackelberg equilibrium.

The most efficient way to save energy in cellular networks is to switch ON/OFF base stations (BSs) dynamically according to the distribution of user equipment (UE) at real time. When a BS is switched ON/OFF, there is a switching energy cost incurred, which is a significant amount and cannot be ignored. By considering this switching cost, we formulate the energy saving problem of BSs in cellular networks as the minimum energy cost problem (MECP). The objective of MECP is to choose the BSs to be active during a period of time and determine the levels of transmission power of the active BSs according to the UEs that are served by the BSs, such that the total energy cost of the BSs is minimized. We propose a scheme to solve the MECP in two steps. In the first step, we aim to minimize the energy cost of all BSs in a time unit independently, without considering the switching ON/OFF BSs across adjacent time units. In the second step, we consider the switching cost of state transitions of BSs by introducing a state transition graph a BS over an entire time period, and transform the MECP into a minimum energy cost flow problem. A minimum cost flow algorithm is developed to solve this problem. Simulation results show that our proposed scheme can achieve significant energy cost reduction of the cellular network, compared with the existing methods.

In this paper, we investigate the residential load scheduling problem within a smart energy community, which is powered by a primary utility along with a small scale local power supplier. As a premise, unit prices set by these two suppliers are different and both are time-varying. Therefore, users are motivated to control their household appliances' operation time and calculate appropriate portions of power purchased from these two suppliers to achieve bill curtailments. The capacity constraint of local power supplier, arising from the renewable energy source and the limited storage capability, also should not be violated. We formulate a residential load scheduling problem to address this situation. Distributed scheme based on information exchange among users is proposed, without over revealing individual user's load profile. Then we propose a distributed algorithm to solve this scheduling problem. Simulation results show that the proposed approach can reduce energy cost of the community and cut down electricity payments of users, and the peak-to-average ratio in load demand is also decreased.

Geographic Information Systems (GIS) users now have multiple options for using elevation data ranging from submeter to kilometers, owing to the rapid development and extensive use of geographic information technologies, such as light detection and ranging (lidar). However, such data are often provided ‘as is,’ creating a need for error propagation and validation of digital elevation data, which is the motivation for this research. We start by seeking the ‘signature’ of elevation distributions. Zipf's law and hypsometry both analyze the distribution of data by aggregation and ranking, the former for discrete data and the latter for continuous data. The objectives of this study are (1) to adopt Zipf's law for discrete nominal or ordinal data and apply it to continuous interval or ratio data; (2) to propose a uniform, parametric model of the elevation distribution of inland water basins for characterizing the overall topographic landscape; (3) to open discussion on another possible statistical method to generalize physical phenomena for geographers and other researchers; and (4) to explore new approaches for error propagation and validate digital elevation data from lidar and other sources. Combining Zipf's law and hypsometry, this article proposes a quadratic polynomial fitting of the log(value)–log(frequency) plot to study the elevation distribution of inland water basins, thus providing a holistic description of the topographic landscape of an inland water basin. Based on several experimental designs, we conclude that the method is scale independent, and it can be applied to different hierarchical levels of water basins. The vertical resolution of elevation is more sensitive than the horizontal resolution. However, the method cannot be applied to arbitrary regions or basins with outflow to the ocean, and a value shift is suggested for using the method in near sea level inland basins. This method introduces additional statistical regularity based on empirical observations of elevation data of inland water basins. It extends Zipf's law from the nominal/ordinal scale to the interval/ratio scale and extends hypsometry from a nonparametric histogram to a parametric quadratic polynomial. Future research that will specifically tackle issues in applications, such as lidar data validation and archeological site prediction models, is also discussed.

We present a geometric and graphic approach to studying spatial patterns of urban hierarchy in the US. The multiplicatively weighted Voronoi diagram is found to be effective for visualizing theoretical regions delineated by socio‐economic variables. The population landscape of the continental US demonstrates overall and stepwise patterns reflecting population, neighborhood and distance, with overwhelming influence from huge metropolitan areas. Stepwise exploration and cluster analysis of the spatial pattern reveal an urban hierarchy. Attributes and arrangement are the two important factors of urban hierarchy, with attribute having a stronger local influence and arrangement having a stronger global influence. The study also presents a variation of Zipf's law to visualize the rank‐size distribution from tabular and statistical space to map space.