How can I further analyze high frequency data from discrete wavelet transform?
I applied a discrete wavelet transform to horizontal wind speed data to receive the below plot. I'm basically trying to use the information from the detail coefficient (the turbulent flow) for further analysis, but I'm not sure the best direction to go in. I don't have much experience with Wavelet Transform, so forgive me if there are obvious options, but the examples I've seen usually discard the higher frequency information since it's the noise of the signal. Is there anything further I can do with this discrete wavelet transform like statistic analysis or forecasting?
The path to pursue really depends on the question that you are trying to answer.
First of all, I would suggest double checking that your DWT is actually doing what you expect it to do. The plot that you shared suggests that it is successful in separating the low frequency coherent (laminar?) flow from the high frequency turbulent flow, but it would be helpful to figure out which frequencies are present in the high frequency component in order to confirm that the processing parameters (e.g. decomposition level) were properly chosen.
Once convinced that your wavelet decomposition provides you with useful information about the turbulent flow, what should you do with these high pass filtered data? I suggest computing their variance over 1 hour long intervals. This is a measure of the "energy" of the signal over the chosen interval. If you are dealing with large amounts of data this would allow you to boil down your time series into a single sample per hour. Maybe you will be able to spot diurnal variations in the turbulent flow (e.g. maybe turbulent flow is higher at dawn). If you have multiple stations it would be interesting to study if the turbulence variations share the same behavior.
Before venturing into time series forecasting, I would really take a closer look at you data and try to identify trends or nail down possible outliers.
Last but not least, I would suggest posting your question on Physics Stack Exchange (e.g. https://physics.stackexchange.com/) rather than on SO.
- Why Prometheus is not suitable for long-term storage?
- OHCL Time Series - Anomaly Detection with Multivariate Gaussian Distribution
- State-of-art for sensor's anomaly detection
- Cold run time for query
- Query timestamp gaps, grouped by a Column value
- Query parameters programatically
- How to fit regression model in R
- Cannot ingest data using ILP and RBAC
- Number of used observations for estimation when using R's 'arima' command
- Speed up nested table query
- Error when parsing dates using Lubridate. All formats failed to parse. No formats found
- Forecasting with `tslm` returning dimension error
- How to predict time series in scikit-learn?
- Forecasting with time series in python
- How to predict time series with limited data
- Advice on how to predict future time series data
- Forecasting time series data
- Is there a way to make Seaborn or Vincent interactive?
- Grafana variable dropdown with name and value with QuestDB data source
- Get the running minimum for 5 rows before on sampled data using QuestDB
- How to efficiently parallelize time series forecasting using dask?
- Filling huge/large chunks of time-series data
- fill_between plot fails on specific index value combo for a pands time-series
- google sheets app script generate 4 random number generation
- Python function for determining if selected file contains time sereis data, doesnot work,
- Validation loss and early stopping in an LSTM Model for Multivariate time series forecasting in PyTorch
- SQL to Retrieve, for all Rows, the Sum of Values of Previous Rows By Time Series
- Generating time series between two dates in PostgreSQL
- How to use the function facet_wrap() to plot time series data in ggplot2 in R
- How to use the function here::here() with the function source() to set the top level of your project folder in R