The synthetic dataset used to train, validate, and test the FSIF-PINN full-spectrum solar-induced chlorophyll fluorescence (SIF) reconstruction algorithm.
收藏资源简介:
This dataset contains the synthetic spectral data used to train, validate, and test the FSIF-PINN full-spectrum solar-induced chlorophyll fluorescence (SIF) reconstruction algorithm. FSIF-PINN is designed to reconstruct the full SIF emission spectrum over 650–800 nm by combining a flexible deep-learning framework with physically constrained radiative transfer consistency. The dataset was generated by combining the ANGERS leaf spectral database, the SCOPE canopy radiative transfer model, and tower-based downwelling irradiance measurements. It includes simulated full-spectrum SIF spectra, canopy reflectance spectra, and corresponding upwelling radiance spectra over the 650–800 nm spectral range. These synthetic spectra were produced to support supervised training, validation, and testing of FSIF-PINN, and can also be used for benchmarking and comparison with other full-spectrum SIF reconstruction methods. Each TFRecord file contains two features: data and labels. The data feature has a length of 1 + 990 × 3 = 2971 for each sample. The first element is an identifier, while the remaining 2970 elements contain three spectral variables: downwelling irradiance, upwelling radiance, and noise-equivalent radiance, each sampled at 990 spectral bands. The labels feature has a length of 990 × 2 = 1980 for each sample, where the first 990 elements correspond to the reference SIF spectrum and the last 990 elements correspond to the reference reflectance spectrum. File structure / data format Each TFRecord sample contains: data: float32, shape = (2971,) element 0: sample identifier elements 1:991: downwelling irradiance spectrum elements 991:1981: upwelling radiance spectrum elements 1981:2971: noise-equivalent radiance spectrum labels: float32, shape = (1980,) elements 0:990: reference full-spectrum SIF spectrum elements 990:1980: reference reflectance spectrum All spectra are sampled on the same 990-band wavelength grid over 650–800 nm.



