Geospatial AI in Earth Observation, Remote Sensing, and GIScience

Geospatial artificial intelligence (Geo-AI) methods have revolutionarily impacted earth observation and remote sensing. With the advancement of deep-learning algorithms and the availability of high-resolution geospatial data, significant strides have been made in geospatial artificial intelligence. Additionally, complex analysis combined with Geo-AI has significantly advanced spatial analysis, providing a better platform for capturing the dynamic processes of spatial changes.

Multimodal methods enrich geographic information science by integrating data from multiple angles, spectra, platforms, and scales, allowing for the effective combination and utilization of complementary information. Nonetheless, multimodal artificial intelligence structures require substantial computational demands. Consequently, there is an emphasis on streamlining the algorithm structure and pivoting towards data-driven machine learning to simulate reality within a relatively simple framework.

This Special Issue spotlights the latest developments in Geo-AI method applications in earth observation and remote sensing. This collection serves as a conduit for scholars in the affiliated disciplines to present their distinct perspectives and most recent accomplishments.

This Special Issue highlights six seminal papers. While they do not encompass every facet of “Geospatial artificial intelligence“, they do echo the most recent innovations and developments in this domain. These articles can be broadly categorized into the following topics.

1. Geo-AI Applied to Disaster Monitoring and Analysis

Drought, landslides, and other natural disasters have profound global impacts. Satellite remote sensing and GIS technology have become instrumental in monitoring and analyzing natural disasters and developing efficient response strategies. This Special Issue contains six articles on the aforementioned fields.

Drought detection based on MODIS, Sentinel-2, and other data may overlook meteorological emergencies. On the other hand, geostationary meteorological satellites have a wide observation range and high sampling frequency, which can enable rapid monitoring. In ‘Validation Analysis of Drought Monitoring Based on FY-4 Satellite’, Han Luo et al. [1] proposed to use hourly FY-4A satellite data to calculate the daily temperature vegetation drought index (TVDI) and used the daily maximum composite (MVC) method to reduce interference from future clouds, atmospheric conditions, and anomalies. The authors selected several representative drought events for validation and compared the results with those obtained from earth observation satellite data (Landsat-8 and MODIS). The results indicate that geostationary meteorological satellites can provide higher frequency drought monitoring capabilities, ensuring the timeliness and accuracy of drought monitoring and introducing innovative data sources for multi-source remote sensing data. Currently, deep-learning algorithms used for landslide detection have high computational complexity and memory requirements, and lightweight algorithms can significantly alleviate these problems. In ‘A Lightweight and Partitioned CNN Algorithm for Multi Landslide Detection in Remote Sensing Images’, Peijun Mo et al. [2] proposed a lightweight LP-YOLO algorithm based on YOLOv5 and applied it to automatic landslide monitoring. This algorithm can mine and aggregate remote information in two directions while consuming less computational costs. The algorithm used the SCYLLA-IoU (SIoU) boundary box regression loss function to replace the complete IoU (CIoU) loss function and proposed a new feature fusion structure. The experimental results showed that the model proposed by the author has better monitoring performance compared to the YOLOv5 algorithm, providing technical guidance for achieving reliable and real-time automatic landslide detection.

The systematic evaluation of geological disasters and the comprehensive risk assessment of extreme climate conditions can provide important references for effectively implementing disaster prevention and reduction work. In the study titled ‘Analysis and Evaluation of Extreme Rainfall Trends and Geological Hazards Risk in the Lower Jinshajiang River’, Xiaojia Bi et al. [3] selected the lower reaches of the Jinsha River as the research area for geological hazard risk assessment. The authors used the time series and trends of extreme precipitation events in the region over the past 60 years, identified nine key indicators for landslide susceptibility evaluation, and conducted a risk assessment on geological disasters such as landslides and debris flows. The geological hazard zoning under extreme rainfall trends was derived using the GIS spatial analysis function. After their analysis, the authors concluded that terrain is the main controlling factor of geological disasters and that human engineering activities and rainfall are the main triggering factors for geological disasters. The research results of this article lay a foundation for informed planning and strategies geared toward geological disaster prevention and control planning in the Yangtze River Basin.

2. Geo-AI Applied to Atmospheric Density Measurement

Understanding the thermal layer density is of utmost importance. Yet, simulating it comes with a high degree of uncertainty. Among many methods for measuring atmospheric density, satellite accelerometer data stand out and have been extensively utilized.

In the study titled ‘Atmospheric Density Inversion Based on Swarm-C Satellite Accelerometer’, Yin et al. [4] used Swarm-C accelerometer data to invert orbital atmospheric density. While the authors built upon the calibration method employed with GRACE-A satellite accelerometer data, they introduced a pivotal modification: a linear temperature correction on this basis. Subsequently, they established a satellite surface shape model. The simplified conical shadow model replaced the traditional cylindrical shadow model for optical radiation pressure modeling. The authors retrieved atmospheric density by calculating the atmospheric drag coefficient of the Swarm-C satellite and conducted tests during geomagnetic storms. The comparison with existing research data demonstrated the necessity of temperature correction and the effectiveness of the proposed method.

3. Geo-AI Application in Hydropower Functional Zoning

This Special Issue also includes a paper on applying geospatial artificial intelligence in hydropower functional zoning. Hydropower development is key to achieving energy conservation and emission reduction strategies, so studying ecological function zoning from the perspective of hydropower functions has relatively high value. In ‘Hydropower Functional Zoning with Crowdsourced Geospatial Data: A Case Study in Sichuan Province’, Li Ju et al. [5] proposed a new framework for hydropower functional zoning based on geographic spatial data crowdsourcing, focused on data collection and utilized crowdsourcing geospatial data and spatial clustering methods. The authors combined the Open Street Map (OSM) with paper reports and geological database data, applied the theoretical framework of water ecological zoning, and combined the spatial clustering results of the hydropower development evaluation to carry out hydropower functional zoning. This paper selected Sichuan Province, which is rich in hydropower resources, as the research area and provided new ideas for developing regional hydropower projects and sustainable management of river basins.

4. Geo-AI Applied to the Study of Human Mobility Patterns

This Special Issue delves into the application of Geo-AI to examine human mobility patterns. In the context of a large-scale human virus epidemic, understanding these patterns becomes crucial to devise strategies for establishing measures to hinder the spread of the virus. In the study titled ‘Geographic Variations in Human Mobility Patterns during the First Six Months of the COVID-19 Pandemic in California’, Kenan Li et al. [6] used a dynamic time-warped self-organizing mapping (DTW-SOM) clustering algorithm to classify human mobility patterns (HMPs) in California census block groups. The author used DTW-SOM to identify four-time patterns of changes in five human mobility indicators and conducted variance analysis and linear discriminant analysis on HMP using 17 social, economic, and demographic variables. This paper focused on the early COVID-19 pandemic in California, and the results indicated that HMP can help predict the effectiveness of non-drug interventions in hindering the spread of the pandemic virus.

Our reviewers and authors have made rigorous efforts to guarantee the high quality of this Special Issue. As a result, numerous papers underwent extensive revisions. While this process entailed extensive and strenuous work, we are confident our efforts and dedication were well justified. We would like to pay special tribute and appreciation to all authors of the articles and all professional and diligent reviewers.