2015 Urban Extents from VIIRS and MODIS for the Continental U.S. Using Machine Learning Methods

The 2015 Urban Extents from VIIRS and MODIS for the Continental U.S. Using Machine Learning Methods data set models urban settlements in the Continental United States (CONUS) as of 2015. When applied to the combination of daytime spectral and nighttime lights satellite data, the machine learning methods achieved high accuracy at an intermediate-resolution of 500 meters at large spatial scales. The input data for these models were two types of satellite imagery: Visible Infrared Imaging Radiometer Suite (VIIRS) Nighttime Light (NTL) data from the Day/Night Band (DNB), and Moderate Resolution Imaging Spectroradiometer (MODIS) corrected daytime Normalized Difference Vegetation Index (NDVI). Although several machine learning methods were evaluated, including Random Forest (RF), Gradient Boosting Machine (GBM), Neural Network (NN), and the Ensemble of RF, GBM, and NN (ESB), the highest accuracy results were achieved with NN, and those results were used to delineate the urban extents in this data set.本数据集为基于机器学习方法构建的2015年美国本土城市范围数据集，用于建模2015年美国本土（Continental United States, CONUS）的城镇聚落。将该机器学习方法应用于日间光谱与夜间灯光卫星数据的融合数据集时，在大空间尺度下的500米中等分辨率上取得了较高精度。该模型的输入数据包含两类卫星影像：来自日/夜波段（Day/Night Band, DNB）的可见光红外成像辐射计套件（Visible Infrared Imaging Radiometer Suite, VIIRS）夜间灯光（Nighttime Light, NTL）数据，以及经校正的中分辨率成像光谱仪（Moderate Resolution Imaging Spectroradiometer, MODIS）日间归一化植被指数（Normalized Difference Vegetation Index, NDVI）数据。尽管本次研究评估了多种机器学习方法，包括随机森林（Random Forest, RF）、梯度提升机（Gradient Boosting Machine, GBM）、神经网络（Neural Network, NN）以及RF、GBM与NN的集成模型（Ensemble of RF, GBM and NN, ESB），但神经网络（NN）取得了最高的精度，其结果被用于划定本数据集的城市范围。