Instantaneous Photosynthetically Available Radiation (IPAR) Prediction Models Based on Neural Network for Ocean Waters

IPAR just below the ocean surface and its subsurface profile controls photosynthesis, heat flux, as well as the change in cellular pigmentation of phytoplankton. Surface IPAR depends on solar position and atmospheric conditions whereas its attenuation below the surface depends on inherent optical properties of ocean waters. The numerical simulations of subsurface light field using radiative transfer techniques are very time consuming and computationally expensive. Some semi analytical models have been developed to predict subsurface light field, which tends to overpredict in clear waters and underestimate in turbid waters. Moreover, there are not any standalone models which can compute surface IPAR and its profile below the surface considering both atmospheric and oceanic conditions simultaneously as inputs. In this work, we carried out a comprehensive study of IPAR influential factors using a large volume of data (8000 datasets each for open ocean waters and coastal waters) derived from radiative transfer simulations for coupled atmosphere and ocean system using the successive order of scattering technique (SOS-CAOS). The simulation datasets were used to train and validate two IPAR prediction models based on Neural network one for open ocean and the other for coastal waters. We modeled the subsurface IPAR profile using the product of a polynomial and an exponential function with two free parameters, which are independent of depth and only depend on atmospheric and ocean inherent optical properties. Neural networks predict surface IPAR and the two parameters with coefficient of determination value greater than 0.98. The diffuse attenuation coefficient of IPAR at 10m depth is calculated from model outputs with root mean square error of 0.018 and mean percentage error of less than 1 percent, which is better than the performance of present available model on the same dataset. The two models will benefit the research community by providing an efficient and robust algorithm in retrieving IPAR from satellite sensor measurements, which directly aligns with the research goals of NASAs Plankton, Aerosol, Cloud, and Ocean Ecosystem (PACE) mission. Furthermore, incorporation of these models in satellite based primary production model will help in the accurate estimation of net primary productivity in global scale.