Fine-scale soil mapping in Alaska, traditionally relying on fieldwork and localized simulations, remains a critical yet underdeveloped task, despite the region's ecological importance and extensive permafrost coverage. As permafrost thaw accelerates due to climate change, it threatens infrastructure stability and key ecosystem services, such as soil carbon storage. High-resolution soil maps are essential for characterizing permafrost and soil taxonomy distributions, identifying vulnerable areas, and informing adaptation strategies. In this paper, we present MiSo, a machine learning (ML) model to produce statewide fine-scale soil maps for near-surface permafrost and soil taxonomy. MiSo integrates a geospatial foundation model for visual feature extraction, implicit neural representations for continuous spatial prediction, and contrastive learning for multimodal alignment and geo-location awareness. We compare MiSo with Random Forest (RF), a traditional ML model that has been widely used in soil mapping applications. Spatial cross-validation and regional analysis across Permafrost Zones and Major Land Resource Areas (MLRAs) show that MiSo generalizes better to remote, unseen locations and achieves higher recall than RF, which is critical for monitoring permafrost thaw and related environmental processes. These findings demonstrate the potential of advanced ML approaches for fine-scale soil mapping and provide practical guidance for future soil sampling and infrastructure planning in permafrost-affected landscapes.

Predicting human mobility across multiple cities is essential for urban applications but remains challenging due to the complex and diverse spatiotemporal dynamics in human trajectories. While recent work often leverages language modeling by treating trajectories as sequences for next-location prediction, these approaches typically rely on raw movement data, process long trajectories without distinguishing between individual trips, and use a single model for all users within a city. To address these limitations, this paper presents MoVER, a transformer encoder-decoder that enriches trajectory representations with location profiles and explicit trip separators. Furthermore, we introduce a clustering-based finetuning strategy to handle user heterogeneity by tailoring models to user groups of similar travel patterns. MoVER outperforms baselines on a validation set and achieves a top-7 ranking among over 50 participating teams in the 14th SIGSPATIAL Cup competition (GISCUP 2025). The code is publicly available at: https://github.com/knowledge-computing/giscup2025-mover.

Historical geologic maps contain rich geospatial information—such as rock units, faults, folds, and bedding planes—that is critical for assessing mineral resources essential to renewable energy, electric vehicles, and national security. However, digitizing maps remains a labor-intensive and time-consuming task. We present DIGMAPPER, a modular, scalable system developed in collaboration with the United States Geological Survey (USGS) to automate the digitization of geologic maps. DIGMAPPER features a fully dockerized, workflow-orchestrated architecture that integrates state-of-the-art deep learning models for map layout analysis, feature extraction, and georeferencing. To overcome challenges such as limited training data and complex visual content, our system employs innovative techniques, including in-context learning with large language models, synthetic data generation, and transformer-based models. Evaluations on over 100 annotated maps from the DARPA-USGS dataset demonstrate high accuracy across polygon, line, and point feature extraction, and reliable georeferencing performance. Deployed at USGS, DIGMAPPER significantly accelerates the creation of analysis-ready geospatial datasets, supporting national-scale critical mineral assessments and broader geoscientific applications.

Text on historical maps contains valuable information providing georeferenced historical, political, and cultural contexts. However, text extraction from historical maps has been challenging due to the lack of (1) effective methods and (2) training data. Previous approaches use ad-hoc steps tailored to only specific map styles. Recent machine learning-based text spotters (e.g., for scene images) have the potential to solve these challenges because of their flexibility in supporting various types of text instances. However, these methods remain challenges in extracting precise image features for predicting every sub-component (boundary points and characters) in a text instance. This is critical because map text can be lengthy and highly rotated with complex backgrounds, posing difficulties in detecting relevant image features from a rough text region. This paper proposes PALETTE, an end-to-end text spotter for scanned historical maps of a wide variety. PALETTE introduces a novel hyper-local sampling module to explicitly learn localized image features around the target boundary points and characters of a text instance for detection and recognition. PALETTE also enables hyper-local positional embeddings to learn spatial interactions between boundary points and characters within and across text instances. In addition, this paper presents a novel approach to automatically generate synthetic map images, SYNTHMAP+, for training text spotters for historical maps. The experiment shows that PALETTE with SYNTHMAP+ outperforms SOTA text spotters on two new benchmark datasets of historical maps, particularly for long and angled text. We have deployed PALETTE with SYNTHMAP+ to process over 60,000 maps in the David Rumsey Historical Map collection and generated over 100 million text labels to support map searching.

Scanned historical maps in libraries and archives are valuable repositories of geographic data that often do not exist elsewhere. Despite the potential of machine learning tools like the Google Vision APIs for automatically transcribing text from these maps into machine-readable formats, they do not work well with large-sized images (e.g., high-resolution scanned documents), cannot infer the relation between the recognized text and other datasets, and are challenging to integrate with post-processing tools. This paper introduces the mapKurator system, an end-to-end system integrating machine learning models with a comprehensive data processing pipeline. mapKurator empowers automated extraction, post-processing, and linkage of text labels from large numbers of large-dimension historical map scans. The output data, comprising bounding polygons and recognized text, is in the standard GeoJSON format, making it easily modifiable within Geographic Information Systems (GIS). The proposed system allows users to quickly generate valuable data from large numbers of historical maps for in-depth analysis of the map content and, in turn, encourages map findability, accessibility, interoperability, and reusability (FAIR principles). We deployed the mapKurator system and enabled the processing of over 60,000 maps and over 100 million text/place names in the David Rumsey Historical Map collection. We also demonstrated a seamless integration of mapKurator with a collaborative web platform to enable accessing automated approaches for extracting and linking text labels from historical map scans and collective work to improve the results.

Recent advances in incorporating physical knowledge into deep neural networks can estimate previously unknown governing partial differential equations (PDEs) in a data-driven way. They have shown promising results in spatiotemporal predictive learning. However, these methods typically assume universal governing PDEs across space, which is impractical for modeling complex spatiotemporal phenomena with high spatial variability (e.g., climate). Also, they cannot effectively model the evolution of potential errors in estimating the physical dynamics over time. This paper introduces a physics-guided neural network, SVPNet, which learns effective physical representations by estimating the error evolution in physics states for correction and modeling spatially varying physical dynamics to predict the next state. Experiments carried out in four scenarios, including benchmarks and real-world datasets, show that SVPNet outperforms state-of-the-art methods in spatiotemporal prediction tasks for natural processes and significantly improves prediction when training data are limited. Ablation studies also highlight that SVPNet is powerful in capturing physical dynamics in complex physical systems.

Large network logs, recording multivariate time series generated from heterogeneous devices and sensors in a network, can reveal important information about abnormal activities, such as network intrusions and packet losses. Existing machine learning methods for anomaly detection on multiple multivariate time series typically assume that 1) infrequent behaviors beyond some inference threshold are anomalous for unsupervised models or 2) require a large set of labeled normal and abnormal sequences for supervised models. However, in practice, the reported abnormal events might be available but incomplete and sparse (i.e., much fewer than normal cases). This paper presents a novel semi-supervised approach, SNetAD, that takes advantage of the incomplete and imbalanced labels to effectively learn separable feature embeddings of network activities representing normal and abnormal events. Specifically, SNetAD first generates network representations by capturing relationships across time points and between network devices. Then SNetAD encourages the embeddings to form two clusters using contrastive center loss and improves the separability of the learned clusters using labeled and unlabeled samples in a semi-supervised manner. The experiments demonstrate that SNetAD significantly outperforms state-of-the-art approaches for abnormal event prediction on a large real-world network log.

Many scientific prediction problems have spatiotemporal data- and modeling-related challenges in handling complex variations in space and time using only sparse and unevenly distributed observations. This paper presents a novel deep learning architecture, Deep learning predictions for LocATion-dependent Time-sEries data (DeepLATTE), that explicitly incorporates theories of spatial statistics into neural networks to addresses these challenges. In addition to a feature selection module and a spatiotemporal learning module, DeepLATTE contains an autocorrelation-guided semi-supervised learning strategy to enforce both local autocorrelation patterns and global autocorrelation trends of the predictions in the learned spatiotemporal embedding space to be consistent with the observed data, overcoming the limitation of sparse and unevenly distributed observations. During the training process, both supervised and semi-supervised losses guide the updates of the entire network to: 1) prevent overfitting, 2) refine feature selection, 3) learn useful spatiotemporal representations, and 4) improve overall prediction. We conduct a demonstration of DeepLATTE using publicly available data for an important public health topic, air quality prediction, in a well-studied, complex physical environment - Los Angeles. The experiment demonstrates that the proposed approach provides accurate fine-spatial-scale air quality predictions and reveals the critical environmental factors affecting the results.

Forecasting spatially correlated time series data is challenging because of the linear and non-linear dependencies in the temporal and spatial dimensions. Air quality forecasting is one canonical example of such tasks. Existing work, e.g., auto-regressive integrated moving average (ARIMA) and artificial neural network (ANN), either fails to model the non-linear temporal dependency or cannot effectively consider spatial relationships between multiple spatial time series data. In this paper, we present an approach for forecasting short-term PM2.5 concentrations using a deep learning model, the geo-context based diffusion convolutional recurrent neural network, GC-DCRNN. The model describes the spatial relationship by constructing a graph based on the similarity of the built environment between the locations of air quality sensors. The similarity is computed using the surrounding "important" geographic features regarding their impacts to air quality for each location (e.g., the area size of parks within a 1000-meter buffer, the number of factories within a 500-meter buffer). Also, the model captures the temporal dependency leveraging the sequence to sequence encoder-decoder architecture. We evaluate our model on two real-world air quality datasets and observe consistent improvement of 5%-10% over baseline approaches.

With the widespread installation of location-enabled devices on public transportation, public vehicles are generating massive amounts of trajectory data in real time. However, using these trajectory data for meaningful analysis requires careful considerations in storing, managing, processing, and visualizing the data. Using the location data of the Los Angeles Metro bus system, along with publicly available bus schedule data, we conduct a data processing and analyses study to measure the performance of the public transportation system in Los Angeles utilizing a number of metrics including travel-time reliability, on-time performance, bus bunching, and travel-time estimation. We demonstrate the visualization of the data analysis results through an interactive web-based application. The developed algorithms and system provide powerful tools to detect issues and improve the efficiency of public transportation systems.

Air quality models are important for studying the impact of air pollutant on health conditions at a fine spatiotemporal scale. Existing work typically relies on area-specific, expert-selected attributes of pollution emissions (e,g., transportation) and dispersion (e.g., meteorology) for building the model for each combination of study areas, pollutant types, and spatiotemporal scales. In this paper, we present a data mining approach that utilizes publicly available OpenStreetMap (OSM) data to automatically generate an air quality model for the concentrations of fine particulate matter less than 2.5 μm in aerodynamic diameter at various temporal scales. Our experiment shows that our (domain-) expert-free model could generate accurate PM2.5 concentration predictions, which can be used to improve air quality models that traditionally rely on expert-selected input. Our approach also quantifies the impact on air quality from a variety of geographic features (i.e., how various types of geographic features such as parking lots and commercial buildings affect air quality and from what distance) representing mobile, stationary and area natural and anthropogenic air pollution sources. This approach is particularly important for enabling the construction of context-specific spatiotemporal models of air pollution, allowing investigations of the impact of air pollution exposures on sensitive populations such as children with asthma at scale.