This paper presents a comprehensive survey of Vision-Language Geo-Foundation Models (VLGFMs), which are advanced models designed to process and analyze geospatial data by integrating visual and linguistic information. VLGFMs are a specialized subset of artificial intelligence models that can handle diverse geospatial data sources, such as remote sensing imagery, geographic information system data, and geo-tagged text. The paper reviews recent developments in VLGFMs, including their background, motivations, core technologies, and applications in various geospatial tasks. It also discusses challenges and future research directions in the field. VLGFMs are categorized into three types: contrastive, conversational, and generative. Contrastive VLGFMs align image and text embeddings in a shared space, enabling tasks like image-text retrieval. Conversational VLGFMs use large language models (LLMs) to generate textual responses, supporting tasks like captioning and visual question answering. Generative VLGFMs produce images based on text or image inputs, enabling tasks like text-to-image generation. The paper discusses the data pipelines used to train VLGFMs, including data collection from scratch and data enhancement using existing datasets. It also covers the architectural choices and modifications made to enhance the performance of VLGFMs, including the use of pre-trained visual encoders, LLMs, and vision-language connectors. The paper highlights the capabilities of VLGFMs, which are categorized into three hierarchical levels: perception, reasoning, and specific geospatial tasks. The paper also discusses the challenges and future directions in the development of VLGFMs, including the need for high-quality, large-scale remote sensing image-text datasets and the importance of leveraging prior knowledge in the remote sensing field. The paper concludes that VLGFMs represent a significant advancement in the application of large general models in the remote sensing domain, with the potential to address specific, real-world problems through fine-tuning networks using high-quality remote sensing image-text datasets.This paper presents a comprehensive survey of Vision-Language Geo-Foundation Models (VLGFMs), which are advanced models designed to process and analyze geospatial data by integrating visual and linguistic information. VLGFMs are a specialized subset of artificial intelligence models that can handle diverse geospatial data sources, such as remote sensing imagery, geographic information system data, and geo-tagged text. The paper reviews recent developments in VLGFMs, including their background, motivations, core technologies, and applications in various geospatial tasks. It also discusses challenges and future research directions in the field. VLGFMs are categorized into three types: contrastive, conversational, and generative. Contrastive VLGFMs align image and text embeddings in a shared space, enabling tasks like image-text retrieval. Conversational VLGFMs use large language models (LLMs) to generate textual responses, supporting tasks like captioning and visual question answering. Generative VLGFMs produce images based on text or image inputs, enabling tasks like text-to-image generation. The paper discusses the data pipelines used to train VLGFMs, including data collection from scratch and data enhancement using existing datasets. It also covers the architectural choices and modifications made to enhance the performance of VLGFMs, including the use of pre-trained visual encoders, LLMs, and vision-language connectors. The paper highlights the capabilities of VLGFMs, which are categorized into three hierarchical levels: perception, reasoning, and specific geospatial tasks. The paper also discusses the challenges and future directions in the development of VLGFMs, including the need for high-quality, large-scale remote sensing image-text datasets and the importance of leveraging prior knowledge in the remote sensing field. The paper concludes that VLGFMs represent a significant advancement in the application of large general models in the remote sensing domain, with the potential to address specific, real-world problems through fine-tuning networks using high-quality remote sensing image-text datasets.