HF radar along the southern coast of Malta : validation, gap filling and seasonality

Abstract

In an attempt to address the issue of coverage gaps in HF radar data obtained by the CALYPSO HF radar network, the performance of DINEOF (Data Interpolation Empirical Orthogonal Functions) as a tool is evaluated. The key objective within this study is the fine tuning of variables for DINEOF alongside the length of data used as input. The performance of DINEOF as a function of missing data in the timeframe to be gap filled is also explored, in the hopes of finding a clear relation. The implementation details of Matlab and Python scripts to carry out the research are presented. This included a script to artificially introduce gaps to allow for easy evaluation of DINEOF’s performance, another to automate the creation of DINEOF “.init” files which store all the variables for the gap filling, and a script designed to go through all the DINEOF results, evaluate the performance, and save all the results in a singular CSV file. As a result of the methodology taken in this dissertation, it was found that DINEOF performs best when the length of data used as input is that of 72 hours or 3 days, and when the alpha parameter (a parameter controlling DINEOF filter strength) is set to a value of 0.1. The values found when fine tuning the execution of DINEOF, the length of data which provided the optimal results being 72 hours, suggests that providing additional input files might produce results that are less accurate. This indicated that using a continuous block of radar data spanning 72 hours as the input optimises DINEOF’s gap-filling accuracy. When attempting to find the point where DINEOF’s performance is hindered as a result of a lack of data in the time frame to be gap filled, no such clear relationship was found. The results obtained suggest that DINEOF has already been optimised to perform under extreme cases of data omission.