Machine Learning For Economic Development Mapping From Open Geospatial Data.pdf

The paper proposes a machine learning-based approach to estimate and map economic development using multi-source open geospatial data. This includes remote sensing imagery and OpenStreetMap road networks. The method involves extracting knowledge-based features from various data sources, constructing multi-view graphs based on spatial adjacency and feature similarity, and employing a multi-view graph neural network (MVGNN) model trained in a self-supervised manner. Handcrafted features and learned graph representations are combined to estimate regional economic development indicators using random forest models.

The study focuses on China's county-level GDP as an example, demonstrating the effectiveness of this approach through extensive experiments. The results show that combining knowledge-based and learning-based features significantly outperforms baseline methods. This method aims to provide timely and accurate socioeconomic variables from accessible geospatial data, supporting smart governance and policy-making.

The paper addresses two key questions: which open geospatial data sources are effective for economic development estimation and how to maximize their use for improved prediction accuracy. The proposed method leverages nighttime light imagery, multispectral remote sensing imagery, and OpenStreetMap road networks to estimate GDP at the county level in China, confirming the contributions of these open data sources.

The paper is organized into sections covering related work, study area and data description, detailed methodology, experimental results and analysis, and conclusions with discussions. Related works highlight the demand for up-to-date socioeconomic information for smart governance and business intelligence, traditionally acquired through labor-intensive surveys. The paper explores leveraging remote sensing imagery and emerging geospatial big data to infer socioeconomic attributes more efficiently.