Artificial Neural Network for Precise Satellite Altimetry Sea Levels Estimations: Testing Using Simulated Data

EasyChair Preprint 6032

Abstract

This paper reports the finding of deep learning technique based on artificial neural network to improve the precision of altimetric sea levels over coastal oceans. It is well-known that waveform retracking protocol is necessary to optimise the estimated geophysical parameters. Most of waveform retracking algorithms are specialised for a specific waveform shapes (e.g. multi-peak, ocean-like and quasi-specular). In attempting to produce precise sea levels from multiple retracking algorithms, one should be concerned about the issue of the existence of relative offset among retrackers, which create ‘a jump’ in the sea level profiles, thus resulting in imprecise sea level estimation. In this study, the neural network technique is explored based 10,000 simulated data, which considered various physical shapes of altimetric waveforms. The data are created using Monte Carlo simulation. The experiment consists of two sets of varying parameters (i.e. number of hidden layer, algorithms in hidden and output layers, and training algorithm). The results indicate that neural network Sets 2 and 3 with 10 and 9 hidden layers are the best parameters for offset reduction. This is indicated by the lowest root mean square error (0.7 cm) and standard deviation of difference (0.2 cm).

Keyphrases: Coastal Altimetry, Coastal Sea Level, Waveform retracking, deep learning, neural network

@booklet{EasyChair:6032,
  author    = {Muhammad Haikal Fayyadh Munadi and Nurul Hazrina Idris},
  title     = {Artificial Neural Network for Precise Satellite Altimetry Sea Levels Estimations: Testing Using Simulated Data},
  howpublished = {EasyChair Preprint 6032},
  year      = {EasyChair, 2021}}