ABSTRACT Geographic science explores the interactions among natural and human elements and their spatiotemporal patterns. Open-source geographic information systems (GIS), with their freedom, openness, and sharing, are gradually being used in geographical practice. However, a systematic review and evaluation of the existing open-source GIS remain lacking. Therefore, we systematically review and analyze 136 representative open-source GIS. Beginning with definitions, characteristics, and developmental trajectories, we outline the current technical system of mainstream open-source GIS. Based on the four key stages of geographic data collection, storage, analysis, and visualization, we categorized and descriptively analyzed the data. The results indicate that the most common license type is General Public License (GPL), with libraries and development tools dominating the software categories. Between 2001 and 2010, the number of released open-source GIS grew rapidly and was applied in multiple domains. A quantitative analysis of the literature highlights the open-source GIS, which has received the most attention. Additionally, we discuss selection preferences across multiple fields, offering insights into future development directions and the potential challenges of open-source GIS. This work provides comprehensive reference when selecting and applying open-source GIS. Open-source GIS projects and research literature are available in Figshare at https://doi.org/10.6084/m9.figshare.29929775.

ABSTRACT Grasslands are the largest terrestrial ecosystems on Earth, providing essential ecosystem services to human society through biogeochemical cycles. Despite their significance, the response of grassland nitrogen cycling to climate change remains inadequately understood on a global scale. To address this, we selected 513 field-based grassland studies that simulate climate change, from which we extracted a total of 4,632 experimental observations sourced from published studies to construct a comprehensive dataset. This dataset presents experimental data on the responses of various nitrogen cycling variables, along with some carbon cycle-related variables, to individual climate change factors, including elevated CO2 concentrations, climate warming, increased precipitation, and decreased precipitation. The database includes detailed research information, location data, vegetation information, experimental conditions, variable response ratios, and variance, providing valuable insights into the independent effects of individual climate change factors on the grassland nitrogen cycle. It also offers a data foundation for predicting nitrogen budgets under future climate change scenarios. The data is available on the Figshare platform (https://doi.org/10.6084/m9.figshare.28702616.v2).

ABSTRACT Large Language Models (LLMs) have demonstrated substantial progress in task automation and natural language understanding. However, without domain expertise in geographic information science (GIS), they continue to encounter limitations including reduced accuracy and unstable performance when processing complex spatial tasks. To address these challenges, we propose GeoJSON agents—a novel multi-agent LLM architecture specifically designed for geospatial analysis. This framework transforms natural language instructions into structured GeoJSON operations through two widely adopted LLM enhancement techniques: function calling and code generation. The architecture integrates three core components: task parsing, agent collaboration, and result integration. The planner agent systematically decomposes user-defined tasks into executable subtasks, while specialized worker agents perform spatial data processing and analysis either by invoking predefined function APIs or by dynamically generating and executing Python-based analytical code. The system produces reusable, standards-compliant GeoJSON outputs through iterative refinement. To systematically evaluate both approaches, we constructed a hierarchical benchmark comprising 70 tasks spanning basic, intermediate, and advanced complexity levels, conducting experiments with OpenAI’s GPT-4o as the core model. Results indicate that the code generation–based agent achieved 97.14% accuracy, while the function calling–based agent attained 85.71%—both significantly outperforming the best-performing general-purpose model (48.57%). Comparative analysis reveals that code generation offers superior flexibility for complex, open-ended tasks, whereas function calling provides enhanced execution stability for structured operations. This study represents the first systematic integration of GeoJSON data with a multi-agent LLM framework and provides empirical evidence comparing two mainstream enhancement methodologies in geospatial contexts, offering new perspectives for improving GeoAI system performance and reducing barriers to GIS application.

ABSTRACT For a long time, human-induced land transformation in sparsely populated areas has been a blind spot for satellite monitoring. The Tibetan Plateau (TP), with scattered human-managed land and vast wilderness, poses challenges for fine-scale monitoring due to frequent cloud cover and complex terrain. While some studies have evaluated the accuracy of land use and land cover (LULC) datasets on the TP, few have focused on a spatial scale-based, multi-category evaluation of human-managed land. Here, we used 16,800 samples within a stratified sampling design to evaluate 13 mainstream LULC datasets, focusing on their ability to capture human-managed land across spatial scales. Results show that the overall accuracy of these datasets ranges from 71.6% to 89.0%, with ESA WorldCover performing best. For built-up land, ESRI LC exhibits the least omission and less commission. For cropland, CCropland and GlobeLand30 perform well. While most datasets depict ecological land well, they tend to underestimate human activity at county and village scales. Specifically, CLCD and GISA barely reflect built-up land, and CACD captures around 50% of cropland. Our study underscores the need to enhance LULC datasets to more accurately reflect human impacts, providing a stronger foundation for evidence-based conservation policy and long-term sustainability.

ABSTRACT Quantifying river runoff is essential for water resource planning, flood control, and ecological protection globally. However, in the high-mountain Third Pole (TP) region, with extensive cryosphere coverage and harsh climate, runoff observations are extremely scarce, especially for transboundary rivers. Complex runoff processes, strongly influenced by snow and glacier melt, challenge hydrological modeling. Currently, long-term, high-resolution runoff datasets are still lacking in TP. Here, we employ an observation-constrained physically based cryosphere-hydrological model to simulate daily runoff during 1981‒2020 over seven transboundary TP rivers. This model is calibrated and validated with daily runoff observations near seven basin outlets, as well as available daily observations from sub-basins and hourly observations. Results show that the model can effectively reproduce the daily streamflow dynamics during both calibration and validation periods in seven TP basins, with Kling-Gupta-Efficiency, Nash-Sutcliffe-Efficiency and correlation coefficients all exceeding 0.67, 0.63 and 0.82, respectively, and absolute-percentage bias within 15.2%. Additionally, validations at sub-basins and hourly-scale validations yield satisfactory results. Furthermore, the snow and glacier modules are validated using reliable remote-sensing-derived data, showing good consistency. Consequently, we generate a long-term 3-km, daily gridded runoff dataset and streamflow records for 11 TP cross-sections, which provide essential support for regional sustainable development and ecological conservation.

ABSTRACT The global proliferation of photovoltaic power plants (PV) has greatly contributed to alleviating the energy crisis and addressing global warming caused by carbon emissions. Accurately assessing the ecological impacts of PV power plants is crucial for balancing ecological protection and the sustainable development of the PV industry. This study used the Google Earth Engine (GEE) platform to construct an object-oriented random forest model, obtaining annual spatial distribution data for PV in China from 2000 to 2023, and quantifying the ecological effects of PV installations at the regional level. The results revealed a phased expansion of PV power plants since 2000, including a germination phase, start-up phase, expansion phase, and boom phase. As of 2023, the total PV area in China was 7,716.13 km2, with the northwest region having the largest installed PV capacity, accounting for 35.76% of the total PV area in China. Most PV installations were located in grassland, cropland, and barren land, which together accounted for more than 85% of the total. Compared to cropland, where about half of the samples improved, PV installations on barren land and grassland showed around 90% and 85% improvement respectively, with smaller NDVI declines and stronger recovery. Furthermore, large-scale PV construction enhanced ecological functions such as shading, cooling, wind-proofing, and sand fixation, thus contributing more effectively to promoting local vegetation growth. This study provides data support and a scientific basis for optimizing the distribution of PV power plants and promoting the development of “PV+” compound projects.

ABSTRACT Remotely sensed air temperature data from NASA POWER are widely used in regions with scarce climatic observations, particularly for agricultural applications such as calculating crop water requirements. This study employed a suite of machine learning (ML) algorithms to correct biases in NASA POWER air temperature outputs, including multiple support vector regression (SVR) variants—Linear SVR, Quadratic SVR, Cubic SVR, Fine Gaussian SVR, Medium Gaussian SVR, Coarse Gaussian SVR—and ensemble decision tree models: bagged trees (BGT) and boosted trees (BT). The objective of this study was to assess the ability of different ML algorithms to reduce biases in NASA POWER air temperature data, with the broader goal of identifying the most suitable ML method for air temperature bias correction in Nigeria. For this analysis, we used daily air temperature records from seven meteorological stations across diverse regions of Nigeria. The performance of NASA POWER minimum and maximum air temperature datasets was evaluated using standard error metrics. Subsequent application of ML algorithms significantly improved data accuracy: the normalized root mean square error (NRMSE) of the corrected outputs was mostly below 10%, indicating excellent predictive performance when ML was integrated. Among the SVR variants tested, Fine Gaussian SVR consistently yielded the best prediction results. This finding suggests that Fine Gaussian SVR is a robust tool for enhancing the reliability of air temperature data—critical for improving the accuracy of crop water requirement calculations in regions where in-situ air temperature observations are limited.

ABSTRACT Geospatial code generation is crucial in integrating AI with geo-scientific analysis, but standardized evaluation tools are lacking. This study presents AutoGEEval++, an enhanced framework for evaluating large language models (LLMs) that generate geospatial code on the Google Earth Engine (GEE) platform. Built on the GEE Python API, AutoGEEval++ includes a benchmark dataset—AutoGEEval++-Bench—comprising 6,365 test cases across 26 GEE data types and three task categories: unit test, combination test, and theme test. The framework offers a fully automated evaluation pipeline, from code generation to execution-based validation, using multi-dimensional metrics such as accuracy, resource consumption, runtime efficiency, and error types. It also supports boundary testing and error pattern analysis. We assess 24 leading LLMs (as of June 2025) spanning general-purpose, reasoning-enhanced, code-centric, and geoscience-specific models. Experimental results highlight distinct performance, stability, and error patterns, demonstrating the framework’s scalability for vertical-domain code generation. This study establishes the first standardized evaluation protocol and resource suite for GEE-based LLM code generation, providing a unified benchmark and a methodology for evaluating the transition from natural language to domain-specific code, advancing geospatial AI research.

ABSTRACT Early traffic event detection is essential for transportation networks’ quick response and accurate performance recovery. Social media data (e.g., Twitter data) can be a valuable source for detecting and describing traffic events, such as accidents, congestion, and road closures. Integrating spatial and temporal features inherent in Twitter can enhance models by revealing non-trivial information, patterns, and knowledge, resulting in improved performance. Recent research in Deep Learning (DL) has revealed the strength of learning features directly from data to extract potential hidden features that efficiently infer human activities and interactions and detect the relationships to generate fine-grained information. This research explores the efficiency of integrating Twitter data’s spatial, temporal, and semantic features for traffic event detection using DL. A novel framework employing a transformer-based multi-feature fusion approach is proposed, designed to detect traffic incidents comprehensively via Twitter data. The framework classifies tweets based on multiple dimensions: (1) semantic content is numerically represented and categorized traffic incidents, non-traffic, or traffic conditions and information; (2) spatial characteristics are analyzed through hot-spot analysis techniques, classifying locations into hot or cold spots; and (3) temporal attributes (date and time) are visualized and analyzed through heat maps reflecting incident densities. The performance of the models was then evaluated based on various fusion scenarios combining spatial, temporal, and semantic data using performance metrics such as F-score and accuracy. The results showed that the scenario of transformer-based multi-feature fusion of spatial, temporal, and semantic data for traffic event detection yielded better results. The model achieved a 7.93% accuracy improvement when distinguishing between the two classes, “Traffic Incident” and “Non-Traffic”, and a 6.09% increase in accuracy when classifying across three categories: “Traffic Incident,” “Non-Traffic,” and “Traffic Conditions and information.” These findings highlight the effectiveness of using a multifeatured Twitter dataset for improved detection accuracy.

ABSTRACT Wetlands provide numerous ecosystems services and benefits that are essential for human society and the environment. However, wetlands have suffered significant loss and degradation globally over the past few centuries due to human disturbances and climate change. It is thus critical to monitor wetlands comprehensively and manage effectively. Meanwhile, comprehensive monitoring is challenging due to difficulties in collecting various wetland data (e.g., in situ hydrological and ecological data, remote sensing images), data analysis using diverse models (e.g., physically based and data-driven), and data visualization. Digital twins, which integrate data collection, analysis, visualization, and sharing into a comprehensive platform, are promising for addressing these challenges. While the concepts and technologies of digital twins have been frequently explored for cities and farms, they have been discussed far less for wetlands. This study attempts to explore the concept of wetland digital twins, identify technologies needed, and discuss associated challenges and opportunities. Though technologies from digital twins of cities and farms are transferable, it is essential to recognize the unique challenges of wetlands, such as their remote locations, limited accessibility, and the need to minimize human interventions. This study aims to bring insights to wetland policymakers and practitioners, promoting digital twins for more effective managements.