Dynamic Mode Decomposition-based short-term prediction of UT1-UTC and LOD using Atmospheric Angular Momentum time series.

We present preliminary results of ultra-short-term prediction (10-day forecast horizon) of UT1-UTC and LOD. Forecast procedure is based on Dynamic Mode Decomposition (DMD) and uses IERS EOP 14 C04 as a reference as well as Atmospheric Angular Momentum (AAM) as auxiliary data (AAM come from GFZ Potsdam and ETH Zurich).

Two main prediction experiments (using different types of input data) were conducted: the ideal and the operational case. The ideal one was based on final (historical) data (both C04 and AAM); predictions performed within one year time span starting on a random MJD with 7 day step between subsequent 10 day forecasts, each yearly time span includes 51 10-day predictions. The operational case was based on operational data, covers the period of the 2nd Earth Orientation Parameters Prediction Comparison Campaign, each variant of this case includes 69 10-day predictions.

Within each experiment and for each considered EOP we prepared some additional analysis. In case of LOD, we conducted predictions using 2 types of input LOD time series: directly on published IERS EOP 14 C04 time series and computed as a derivative of IERS EOP 14 C04 UT1-UTC time series. The final UT1-UTC predictions vary depending on the method of determining the constant of integration restoring the proper scale of UT1-UTC.

In the ideal case the mean absolute prediction errors for UT1-UTC vary from 0.009 ms – 0.036 ms for the 1st day and 0.224 ms – 0.292 ms for the 10th day of prediction, whilst those values vary from 0.016 ms – 0.028 ms and 0.045 ms – 0.063 ms for LOD prediction. Corresponding values in the operational case are within the range of 0.058 ms – 0.065 ms and 0.438 ms – 0.463 ms for UT1-UTC, whilst for LOD these values are 0.032 ms – 0.040 ms and 0.093 ms – 0.099 ms.

In operational settings of UT1-UTC prediction, we can observe that our results are slightly worse than the accuracy of IERS predictions (Bulletin A), while comparable to the accuracy of the forecast of methods from 2nd EOPPCC. The results demonstrate that the proposed techniques can efficiently forecast UT1-UTC and LOD. Nevertheless, a deeper analysis is needed on efficient incorporation of Effective Angular Momentum Functions information to improve presented prediction procedure.