Oral Session 1 | Monday, October 3, 12:00–12:20 | Abstract 497
Simulation and correction of LISST-VSF multiple scattering errors
LISST-VSF is a commercially available instrument providing state-of-the-art in situ measurements of the volume scattering function (VSF) at 0.01-150 degrees. Accurate VSF measurements are important for several reasons, such as understanding underwater particle-light interactions and improving satellite and in situ observation capabilities. However, significant errors in the measured VSF are observed in turbid waters. We have simulated the LISST-VSF using a Monte Carlo algorithm, validated through comparison with laboratory measurements. The results demonstrate that the observed errors originate from multiple scattering. When computing VSFs from LISST-VSF measurements, one assumes that all scattered photons are only scattered once. This assumption breaks down when the water becomes turbid, causing elevated VSFs and distorted phase functions. The measured VSF can be orders of magnitude larger than the true VSF, and the error is shown to be dependent on both optical depth and phase function. By employing a neural network, we can approach a general correction function for the LISST-VSF using a large set of training data. While experimental data is limited, training data can be generated using the Monte Carlo algorithm. Training the network on VSFs simulated with large variations in phase function and scattering coefficient enables accurate predictions of the true VSF from VSF measurements. Experimental results show that the error in the VSF can be reduced from 384% to 11% for very turbid water. Multiple scattering in turbid waters is likely a widespread problem for scattering instrumentation. Hence, methods presented here can be applicable to other VSF instruments and past measurements.
Håvard Ugulen, University of Bergen
Håkon Sandven, University of Bergen
Camilla Sætre, University of Bergen
Børge Hamre, University of Bergen
Arne Kristoffersen, University of Bergen