Model Evaluation dataset for work titled " A Deep Convolutional Neural Network for Retrieving Tropospheric Temperature and Moisture Profiles from Refractivity over Tropical Oceans: Framework Development and Characterization"

This dataset contains about 27,400 independent radiosonde profiles used exclusively for evaluating the convolutional neural network (CNN) retrieval framework described in the companion manuscript titled " A Deep Convolutional Neural Network for Retrieving Tropospheric Temperature and Moisture Profiles from Refractivity over Tropical Oceans: Framework Development and Characterization". None of these profiles were used in model training. Each file in the datasets comprise of co-located measured and CNN-retrieved vertical profiles of temperature (T), water vapor pressure (Pw), and relative humidity (RH), interpolated to a uniform 10 m vertical grid from 100 m to 20 km above mean sea level. The refractivity (N) profile used as model inputs are also included, alongside the reference and the predicted dry refractivity and partial pressure of dry air. Profiles are drawn from the following publicly available sources: DYNAMO 2011–12 field campaign data (NCAR/EOL): RV Baruna Jaya III (https://doi.org/10.26023/GT5S-GSCP-3Q05), RV Mirai (https://doi.org/10.26023/ZFQ2-TG96-TA0P), RV Roger Revelle (https://doi.org/10.5065/D6TH8K3Z), and RV Sagar Kanya (https://doi.org/10.26023/2JNV-7YQ5-610T). TOGA-COARE 1992–93 data (NCAR/EOL): Kapingamarangi (https://doi.org/10.5065/D6TQ5ZTV), Kavieng (https://doi.org/10.5065/D62805XG), Manus (https://doi.org/10.5065/D66W98CD), Nauru (https://doi.org/10.5065/D6K64GB2), RV Kexue (https://doi.org/10.5065/D67S7M21), RV Shiyan 3 (https://doi.org/10.5065/D6416VBM), RV Xiangyanhong 5 (https://doi.org/10.5065/D6GB22BS), RV Moana Wave (https://doi.org/10.5065/D6VH5M3J), RV Hakuho Maru (https://doi.org/10.5065/D6513WHF), and RV Vickers (https://doi.org/10.5065/D6319T5S). note these observations predate the CNN training period by approximately three decades GUAN high-resolution data: University of Wyoming Atmospheric Science Radiosonde Archive (https://weather.uwyo.edu/upperair/sounding.shtml). Each profile in this dataset is entirely independent of the approximately 20,800 profiles used for CNN model training, which are described separately in the manuscript. The training data are not included in this deposit. Details of the file contents are described in the README.docx file