Data and code for photovoltaic area estimation under a fixed review budget with variance-constrained sampling
Description
This dataset provides derived numerical research data and analysis code for photovoltaic footprint area estimation under a fixed review budget. The target inventory comprises 150 aerial and satellite image frames and 25,280 unique native review windows. Primary evaluation uses 48 newly evaluated aerial frames with 19,072 windows in California and Queens. The files include released-human-annotation area summaries, explicitly identified frozen computer vision model outputs for three fixed seeds, 780 sampling configurations, 60,000 saved review draws at budget 400, oracle diagnostics, sensitivity results, 2,016 recorded CPU measurements and a portable numerical reproduction program. Reference areas are derived from the original public human annotations; predictions and risk scores are machine learning outputs. Original imagery, annotations and model weights remain available through the source repositories cited in the package. FILE_INDEX.csv describes each file, and DATA_DICTIONARY.csv defines every column.
Files
Steps to reproduce
PV_Area_Data_v1.zip: Numerical research data, analysis code, column definitions, source attribution, individual file descriptions and checksums for the study. Extract the archive and start with PV_Area_Data/README.txt. README.txt: Dataset overview, study populations, units, reproduction instructions, source attribution and the mapping from data files to manuscript figures/tables.