一种融合文本中地理位置和土地利用/覆被信息的野生动物活动细粒度定位方法

文本蕴含大量地理位置描述信息,有效融合地理关联信息以实现文本的精细定位是地理信息服务的难点。本文提出一种融合土地利用/覆被信息的描述地理位置的细粒度定位方法：在文本描述地理关联信息（地理位置实体、土地利用/覆被实体与空间关系）抽取、土地利用/覆被精细分类与地理位置粗粒度匹配定位的基础上,使用自然语言空间关系近似转换模型,确定地理位置的细粒度定位范围;基于土地利用/覆被实体及其周边精细分类信息,在该范围内检索匹配,确定地理位置的细粒度定位坐标。本文以野生亚洲象活动/肇事监测文本为例开展实验,并用匹配率与位置精度评价定位质量,结果表明：本文方法显著提升了地理位置的细粒度定位质量,实验精确匹配率（81.51%）、位置误差距离的均值（65.97 m）及其≤50 m的比例（70.50%）均优于国内主流在线地理编码与地名检索服务结合空间关系或其单独使用结果。该方法有助于完善地理位置定位方法体系、提升地理信息空间化质量,并可服务于野生动物活动/肇事监测预警等精细定位任务。

A Novel Method for Fine-Grained Geolocation of Wildlife Activities by Integrating Geographical Information and Land Use/Cover in Texts

Text data contain rich geographic information. How to mine and spatialize the geographic information embedded in text data through linking the geographic location text with its spatial location in the real world is fundamental for utilizing geographic information. However, as the semantic granularity of geographic location in texts is too raw to be directly used in most cases. It becomes a major challenge for geographic knowledge services to achieve fine-grained geolocation of texts by effectively integrating geographical information with other related features such as land use/cover. The existing geolocation methods, including geocoding, place name retrieval, and fuzzy area modeling, have been widely used to decode non-urban geographic location texts without considering land use/cover information. These methods usually failed to precisely extract geolocation in texts on wildlife activities. In this study, we proposed a fine-grained geolocation method through the inclusion of land use/cover information in texts on wildlife activities. This method employed a natural language spatial relationship approximation conversion model to determine a fine-grained geolocation domain by integrating geographically relevant information (including geographic location entities, land use/cover entities, and spatial relationships), fine classification of land use/cover, and coarse-grained matching geolocation. The coordinates of fine-grained geolocation were determined by iteratively searching and matching within the fine-grained geolocation domain by combining the natural language form of land use/cover entities and fine land use/cover classification map. Our experiments were conducted using texts information of the wild Asian elephants' activities/accidents occurred in southern Yunnan Province of China. The quality of geolocation in the experiments was evaluated using matching level and location accuracy. The results show that the method proposed here can soundly mine fine-grained geolocation of texts on wild Asian elephants' activities/accidents. By mining and analyzing the texts on Asian elephants' activities/accidents in an area with frequent human-elephant conflict in 2020, fine-grained geolocation of the examined Asian elephants' activities/accidents was accurately extracted. Compared with the domestic mainstream online geocoding and place name retrieval services with or without considering spatial relationships, the proposed method greatly improved the quality of fine-grained geolocation. The exact matching ratio of experimental location points reached to 81.51%, and the mean value of the location error distance between location points and real points was 65.97 m, with a proportion of the location error distance below 50 m of 70.50%. The significant outperformance of this method in mining and spatializing geographic information of texts on Asian elephants' activities/accidents sheds new light on wildlife monitoring and early warning, and human-wildlife conflict emergency management based on the fine-grained geolocation derived from multi-media texts on wildlife activities.