Land Use from Nontraditional Analytics

OBJECTIVE: This announcement seeks proposals to develop a flexible and adaptable artificial intelligence (AI) system capable of classifying fine-grained subcategories of urban land use via the fusion of imagery, remote sensing, and non-imagery geospatial data layers. DESCRIPTION: For decades, geographers have used land use maps to quantify and depict spatial heterogeneities in land use and land cover (LULC) for purposes of land management and urban planning. These efforts have traditionally relied upon satellite imagery and remote sensing techniques to categorize LULC across a landscape1. Machine learning methods have enabled the classification of broad LULC categories in urban environments (e.g., commercial versus industrial2)], but finer grained features of urban landscapes such as the industry of a manufacturing plant or the business sector of an office complex can be optically cryptic and difficult to distinguish using satellite imagery and remote sensing techniques alone. A wide range of non-imagery sources of geospatial information (e.g., measures of transportation flow and connectivity; pedestrian pattern-of-life analytics; bike tracks; census data; OpenStreetMap data; etc.) offer complementary sources of information that can be fused with imagery to classify finer grained subcategories of urban land use3. However, the heterogeneous nature of these non-traditional datasets presents challenges that complicate analytic development (e.g., spatiotemporal variability in data density and coverage). The purpose of this effort is to develop and validate an adaptable AI system that can process and integrate multivariate, heterogeneous non-imagery data sources with satellite imagery to classify fine-grained subcategories of urban land use with greater accuracy than that which would be afforded by imagery and remote sensing methods alone. NGA will ONLY be accepting DIRECT to Phase II proposals for this topic. PHASE I: Provide documentation to substantiate that the scientific and technical merit and feasibility described in the Phase II section of the topic has been met and describes the potential commercial applications. Documentation should include all relevant information including, but not limited to: technical reports, test data, prototype designs/models, and performance goals/results. PHASE II: Develop a prototype that fuses satellite imagery and remote sensing data layers with other non-traditional non-imagery sources of geospatial information to classify land use at the neighborhood scale in three distinct areas of interest outside of the continental United States. This effort strives to integrate geospatial datasets derived from imagery and non-imagery sources to derive land use signatures that surpass those derived from imagery alone. Although the offeror may select the land use classes germane to the particular use case under investigation, the offeror is encouraged to include the following in the analysis: medical, education, government, military, commercial, industrial, residential, recreational, and undeveloped. In the proposal, the offeror should include a description of the proposed methodology, experimental plan, and validation strategy. Although deep learning approaches are encouraged, these techniques are not required to meet the objectives of this effort. Nevertheless, it is important that the methodology be scalable and adaptable to alternative use cases and inputs. The offeror should also clearly identify any proposed non-imagery data layers that will be included in the analysis, the ground truth that will be used to validate the proposed approach, and the proposed spatial areas to be investigated. PHASE III DUAL USE APPLICATIONS: Follow-on activities are expected to be aggressively pursued by the offeror. Follow-on work is intended to transition to a secure compartmented information facility for further development in Phase III. REFERENCES: 1. Liverman D, Moran EF, Rindfuss RR, Stern PC, eds. People and Pixels: Linking Remote Sensing and Social Science (1998). Natl Acad Press, Washington, DC. 2. Huang, Bo, Bei Zhao, and Yimeng Song. Urban Land-Use Mapping Using a Deep Convolutional Neural Network with High Spatial Resolution Multispectral Remote Sensing Imagery." Remote Sensing of Environment 214 (2018): 73 86. 3. Grippa T, Georganos S, Zarougui S, Bognounou P, Diboulo E, Forget Y, Lennert M, Vanhuysse S, Mboga N, Wolff E. Mapping Urban Land Use at Street Block Level Using OpenStreetMap, Remote Sensing Data, and Spatial Metrics. ISPRS International Journal of Geo-Information, 2018; 7(7): 246. https://doi.org/10.3390/ijgi7070246. KEYWORDS: land use; land cover; artificial intelligence; machine learning; deep learning; convolutional neural network; geospatial information; pattern-of-life; transportation; social network analysis; network theory; network embedding