Qdrant/NOAA-Buoy

NOAA Buoy meterological data

NOAA Buoy Data was downloaded, processed, and cleaned for tasks pertaining to tabular data. The data consists of meteorological measurements. There are two datasets

1. From 1980 through 2022 (denoted with "years" in file names)
2. From Jan 2023 through end of Sept 2023 (denoted with "2023" in file names)

The original intended use is for anomaly detection in tabular data.

Dataset Details

Dataset Description

This dataset contains weather buoy data to be used in a tabular embedding scenarios. Buoy 42002 was chosen because it had many years of historical data and was still actively collecting information

Here is the buoy's page and its historical data page:

• https://www.ndbc.noaa.gov/station_page.php?station=42002
• https://www.ndbc.noaa.gov/station_history.php?station=42002

Only standard meteorological data and ocean data was downloaded. Downloaded started at 1980, which is the first full year of collecting wave information.

Data Fields

{'TSTMP': 'timestamp', '#YY': '#yr', ' MM': 'mo', 'DD': 'dy', 'hh': 'hr', 'mm': 'mn', 'WDIR': 'degT', 'WSPD': 'm/s', ' GST': 'm/s', ' WVHT': 'm', 'DPD': 'sec', 'APD': 'sec', 'MWD   ': 'degT', 'PRES': 'hPa', ' ATMP': 'degC', ' WTMP': 'degC' }

Dataset Creation

Curation Rationale

The original data has inconsistent delimiters, different and inappropriate missing data values, and was not harmonized across years. Pre-2023 was edited in the same way as the previous data but kept separate to allow for train and inference.

Source Data

Initial Data Collection and Normalization

Data Downloaded on Oct 12 2023

All code used to transform the data can be found in the buoy-python directory. This is NOT production code and the focus was on correct results and minimizing time spent writing cleaning code.

1. #YY, MM, DD, hh, mm were concatenated to create a timestamp and stored in a new column.
2. From 1980 until 2005 there was no recording of minutes. Minutes for those years was set to 00.
3. All missing data was set to a blank value rather than an actual number
4. Remove all rows without wave data from all the data sets ( missing value in WVHT and DPD)
5. Columns MWD, DEWP, VIS, and TIDE were removed because of consistent missing values
6. From 2005 -> 2006 Wind direction goes from being called WD to WDIR
7. From 2006 -> 2007 Header goes from just 1 line with variable names to 2 lines with the second line being units.

These steps were used to create full_2023_remove_flawed_rows, the 2023 months, and full_years_remove_flawed_rows the previous data going back to 1980

Since the original purpose of this data was anomaly detection. The two data sets above received further processing.

1. All data values were converted to Z-scores (file named zscore_2023)
2. For 1980 - 2022, all rows with 2 or more fields with Z-scores > 2 were removed from the dataset (file named trimmed_zscores_years )

Uses

Direct Use

Primary use is working with tabular data and embeddings, particularly for anomaly detection