High frequency, or HF, coastal radars collect a vast amount of data on ocean currents, winds and waves. The technology continuously measures the parameters, by receiving and interpreting electromagnetic waves scattered by the ocean surface. Formulating the methods to interpret the radar data, to obtain accurate measurements, has been the focus of many researchers since the 1970s.
Much of the existing research has been in monostatic radar theory, where the transmitter and receiver are stationed together. However, a larger, higher quality dataset can be obtained by utilising bistatic radar theory, whereby the transmitter and receiver are located at separate sites.
In this work, the focus is on bistatic radar, where the most commonly used mathematical model for monostatic radar is adapted for bistatic radar. Methods for obtaining current, wind and wave information from the model are then described and in the case of winds and waves, tested. Investigating the derived model shows that it does not always fit the real data well, due to undesirable effects of the radar. These effects can be incorporated into the model but then the existing methods used to obtain ocean information may not be applicable. Therefore, a new method for measuring ocean waves from the model is developed.
The recent advances in machine learning have been substantial, with the neural network becoming proficient at finding the link between complexly related datasets. In this work, a neural network is used to model the relationship between the developed radar model and the directional ocean spectrum. It is shown to successfully invert both monostatic and (for the first time) bistatic HF radar data and with this success, it becomes a viable option for obtaining ocean surface parameters from radar data.
