In this paper, we conducted a scientometric analysis based on the Night-Time Light (NTL) remote sensing related literature datasets retrieved from Science Citation Index Expanded and Social Science Citation Index in Web of Science core collection database. Using the methods of bibliometric and Social Network Analysis (SNA), we drew several conclusions: (1) NTL related studies have become a research hotspot, especially after 2011 when the second generation of NTL satellites, the Suomi National Polar-orbiting Partnership (S-NPP) Satellite with the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor was on board. In the same period, the open-access policy of the long historical dataset of the first generation satellite Defense Meteorological Satellite Program’s Operational Linescan System (DMSP/OLS) started. (2) Most related studies are conducted by authors from USA and China, and the USA takes the lead in the field. We identified the biggest research communities constructed by co-authorships and the related important authors and topics by SNA. (3) By the visualization and analysis of the topic evolution using the co-word and co-cited reference networks, we can clearly see that: the research topics change from hardware oriented studies to more real-world applications; and from the first generation of the satellite DMSP/OLS to the second generation of satellite S-NPP. Although the Day Night Band (DNB) of the S-NPP exhibits higher spatial and radiometric resolution and better calibration conditions than the first generation DMSP/OLS, the longer historical datasets in DMSP/OLS are still important in long-term and large-scale human activity analysis. (4) In line with the intuitive knowledge, the NTL remote sensing related studies display stronger connections (such as interpretive frame, context, and academic purpose) to the social sciences than the general remote sensing discipline. The citation trajectories are visualized based on the dual-maps, thus the research preferences for combining the environmental, ecological, economic, and political science disciplines are clearly exhibited. Overall, the picture of the NTL remote sensing research is presented from the scientist-level, topic-level, and discipline-level interactions. Based on these analyses, we also discuss the possible trends in the future work, such as combining NTL studies with social science research and social media data.
Geospatial code generation is emerging as a key direction in the integration of artificial intelligence and geoscientific analysis. However, there remains a lack of standardized tools for automatic evaluation in this domain. To address this gap, we propose AutoGEEval, the first multimodal, unit-level automated evaluation framework for geospatial code generation tasks on the Google Earth Engine (GEE) platform powered by large language models (LLMs). Built upon the GEE Python API, AutoGEEval establishes a benchmark suite (AutoGEEval-Bench) comprising 1325 test cases that span 26 GEE data types. The framework integrates both question generation and answer verification components to enable an end-to-end automated evaluation pipeline—from function invocation to execution validation. AutoGEEval supports multidimensional quantitative analysis of model outputs in terms of accuracy, resource consumption, execution efficiency, and error types. We evaluate 18 state-of-the-art LLMs—including general-purpose, reasoning-augmented, code-centric, and geoscience-specialized models—revealing their performance characteristics and potential optimization pathways in GEE code generation. This work provides a unified protocol and foundational resource for the development and assessment of geospatial code generation models, advancing the frontier of automated natural language to domain-specific code translation.
Geospatial foundation models encode geographic environments into compact representations, but their interpretation across spatial scales and time remains challenging. We investigate how frozen AlphaEarth Foundations (AEF) representations can generate descriptions of land-surface characteristics, urban functions, and socioeconomic conditions. We develop AEFEcon-GPT, which aligns multiscale AEF representations with a language model through a learned projection and two-stage semantic alignment. Structured labels from independent geographic datasets supervise regional types, attributes, and spatial–temporal relations; auxiliary attributes are not required during inference. The framework supports static description and spatial comparison at pixel, area-of-interest, and 1 km grid scales, and temporal interpretation at the grid scale. Experiments cover 36 Chinese cities during 2017–2024, with 167,088 supervision instances and 3,282 evaluation samples. Static interpretation achieves 0.620 exact-grade accuracy, 0.858 adjacent-grade accuracy, and 0.934 attribute coverage. Attribute-level accuracies are 0.608 for temporal change directions and 0.625 for spatial comparison relations. Representation diagnostics and ablations show that factual descriptions depend on both information encoded in AEF representations and structured semantic supervision. The framework enables verifiable geographic interpretation of geospatial foundation representations across scales and time.
Pixel-level slum mapping remains constrained by limited cross-city generalisation, the lack of continuous estimates of intra-pixel slum coverage, and weak comparability across geographic contexts. AlphaEarth Foundations (AEF), which provides globally consistent 64-dimensional annual surface embeddings at 10 m resolution, offers an analysis-ready representation for scalable and lightweight slum monitoring. However, its applicability to slum mapping remains unclear because slum conditions are only indirectly observable from Earth observation data and are jointly shaped by built-form and socioeconomic processes. We systematically evaluate AEF for binary slum classification and sub-pixel slum cover fraction estimation across 12 cities and 69 city–year pairs from 2017 to 2024, using high-resolution GRAM-derived pseudo-masks as supervisory labels. The evaluation spans four training strategies, two protocols (random split and 3 × 3 spatial block cross-validation), six auxiliary feature configurations, and five baseline models, complemented by representation-level analyses (PCA, SHAP) and full-AOI mapping. Five findings emerge. (1) Same-city cross-year training is optimal under both protocols (median spatial F1 = 0.616, R2 = 0.466); temporal expansion outperforms cross-city transfer, indicating city-scale representational drift. (2) Regression R2 is driven primarily by zero/non-zero boundary discrimination: positive-pixel R2 is consistently negative across all cities, revealing limited capacity to model intra-pixel variation in sub-pixel slum cover fraction at 10 m. (3) PC36 is consistently top-ranked across tasks; classification saturates at k = 32 while regression remains unsaturated at k = 64. (4) POI features yield the largest density gain (ΔR2 = +0.064). (5) For six cities meeting dual-task usability thresholds, full-AOI inference across 2017–2024 preserves slum cluster structure (mean SSIM = 0.926). The study delineates the capabilities and complementarity needs of foundation-model embeddings for slum monitoring.
The rapid advancement of large language models (LLMs) offers new avenues for overcoming the limitations of current data-driven GIS analysis methods, giving rise to the emerging field of large language model-driven GIS analysis (LLM4GIS). This study presents a systematic review of 673 relevant research articles published between 2022 and July 2026. By tracing the five-stage evolutionary trajectory of GIS analysis methods and elaborating the framework for LLM4GIS, we conduct a quantitative bibliometric analysis that reveals the field’s rapid growth from sporadic exploration to explosive expansion, along with the shifting landscape of its dominant technical trajectories. Building on these analyses, we develop a taxonomy spanning four dimensions—technical approaches, datasets, applications, and evaluation—and systematically analyze the advances and evaluation methodologies across five core application areas: geospatial knowledge question answering, knowledge extraction, spatiotemporal reasoning, geospatial code generation, and analysis modeling. Key challenges and future research directions are further discussed along five dimensions: multimodal data integration, high-quality data development, domain-specific model development, integration of world models with real-world geospatial environments, and interpretability and trustworthiness. This review provides a systematic reference for advancing the bidirectional synergy between LLM4GIS and GIS4LLM and for fostering a new generation of spatiotemporal intelligence ecosystems.