I’ve long touted the advantages of using NDBC data for introducing students to programming and data analysis, with an oceanographically focused dataset. In particular, their DODS Server makes it fairly easy to access decades of data from hundreds of stations using the xarray library and a few lines of code in Python.
Apparently, people have been listening. Over the last year, we’ve heard from several faculty in our community who have started down this path. But, a few of them have run into the same confusing issue: If you try to request a full dataset from the NDBC DODS server, you often get the (unhelpful) error…
RuntimeError: NetCDF: file not found
Here’s an example of what it might look like.
Yeah…. Errors are never fun when you’re just starting out. And running into an error you don’t know how to deal with (and where googling isn’t helpful either), is even worse. This is a ripe cause for frustration and demotivating for learners.
To be honest, I’m not entirely sure why this error occurs, but there are a few workarounds you can use, which I’ve detailed in the example notebook below. If you use the DODS server with Python, this will hopefully help you understand some of its limitations, as well as some tricks that might help you debug future errors you run into.
Another Approach: Pandas + ERDDAP
Another option is to skip using xarray and DODS altogether, and instead use the Pandas library to access the same dataset on the CoastEatch ERDDAP server. Pandas tends to be a little more intuitive for students, and so this approach might even be preferable for those just starting out.
If you’re interested in this approach, here’s a simple example.
Most of my example notebooks to date have relied on xarray, because it’s required for accessing netcdf data files and multidimensional datasets. Spend any time in oceanography and you’ll definitely have to learn this library, but it’s not necessary on day 1.
More fun with NDBC
If you’re interested in some of our other NDBC dataset examples in Python, you should check out our Python Tutorials (Part 1, Part 2, and Data Analysis), as well as this example on Seasonal Cycles. Just keep in mind, that these examples use the same xarray/DODS technique below, so if you try to load a full dataset, you will need to keep these tricks in mind.
As always, I hope this helps!
A quick fix for a common NDBC DODS error¶
By Sage Lichtenwalner, March 9, 2021
The NDBC DODS OpenDAP server is a great resource, providing decades of weather and ocean data from hundreds of stations around the globe. This dataset is a great resource for students interested in accessing a large and intuitive dataset to learn programming and data analysis skills.
Unfortunately, if you try to request a long timeseries from some stations, you might run into the following error:
RuntimeError: NetCDF: file not found
If this is the case, here are 2 workarounds that might help.
The problem¶
First, let's setup our notebook like we normally do for the Google Colab environment.
# Notebook setup
import xarray as xr
!pip install netcdf4
import matplotlib.pyplot as plt
# Let's make our plots pretty
import seaborn as sns
sns.set()
Next, let's try to load a full timeseries for Station 44065, which has data that goes back to 2008.
data = xr.open_dataset('https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065.ncml')
# data['sea_surface_temperature'].plot() # This line will fail
The .ncml file is really an aggregation of the other individual yearly files available for this station (see here). For some reason, when you try to load a large time range with the .ncml file, you get the "netcdf file not found error.”
I'm not quite sure if this is because the lat/lon coordinates or another variable changes in the long series, which causes an aggregation issue on the server.
But as unintuitive as the error may be, we can in fact still grab all the data.
Option 1 - Make a smaller time range request¶
Bizarrely, and perhaps annoyingly, if you change your date range to a smaller period, it works.
After a bunch of testing with this station, I found I can request 1/1/2009 to the present without issue. But as soon as I try to add 2008, I get the error. But 2008 to a 2-3 months shy of the present day also loads. This is the kind of inconsistent bug that makes debugging hard. (But at least in this case the error is consistent.)
data = xr.open_dataset('https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065.ncml')
# data = data.sel(time=slice('2008-01-01','2021-03-30')) # This fails as of 3/9/21
# data = data.sel(time=slice('2009-01-01','2021-03-30')) # Removing the first year works
data = data.sel(time=slice('2008-01-01','2021-01-30')) # Removing the last 1-2 months works
# data = data.sel(time=slice('2008-01-01','2021-02-01')) # But this also fails
df = data.to_dataframe().reset_index().set_index('time') # Convert to dataframe to force loading of the full dataset
df.tail(3)
# Quick plot
data['sea_surface_temperature'].plot();
Let's try the same approach for Station 44025.
data = xr.open_dataset('https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44025/44025.ncml')
# data = data.sel(time=slice('1975-01-01','2021-03-30')) # This fails as of 3/9/21
# data = data.sel(time=slice('1976-01-01','2021-03-30')) # Removing the first year works
data = data.sel(time=slice('1975-01-01','2021-01-30')) # Removing the last 1-2 months works
# data = data.sel(time=slice('1975-01-01','2021-02-01')) # But this also fails
data.sea_surface_temperature.plot();
So, if you don’t need the full range, you can leave off a year on either end. Or alternatively, if you can find the offending time breaks, you might be able to make smaller requests and then merge them together.
Option 2 - Don't use the aggregate ncml file¶
Another approach that seems to work, is to specify the individual data files using open_mfdataset()
like this...
urls = ['https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2008.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2009.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2010.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2011.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2012.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2013.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2014.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2015.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2016.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2017.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2018.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2019.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2020.nc',
'https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44065/44065h2021.nc',
]
data = xr.open_mfdataset(urls, combine='by_coords')
df = data.to_dataframe().reset_index().set_index('time')
df.tail(3)
A Note about Realtime Data¶
In the examples above, you'll notice that the most recent data available for both stations is from 1/31/2021, even though I'm running this script on 3/9/21.
This is because the "yearly" netcdf files on the DODS server only seem to include "verified" data, which is added to the archive in monthly chunks. (My guess is that this typically occurs approximately 45 days after the end of each month.)
If you want real-time data, you will need to use the "9999" files available in each station's directory. Apparently these files include some historic data as well, but how much varies from station to station.
So, if you want a full archive up to the present day, you'll have to make and merge multiple requests, as these recent/real-time data points are not available through the .ncml aggregation file.
Here's a quick example for Station 44025. You can see that this file includes several years of data, but nothing like the decades available in the example above.
data = xr.open_dataset('https://dods.ndbc.noaa.gov/thredds/dodsC/data/stdmet/44025/44025h9999.nc')
data.sea_surface_temperature.plot();
df = data.to_dataframe().reset_index().set_index('time')
df.head(1)
df.tail(1)
For more guidance, check out my Accessing NDBC Data tutorial, as well as the other Python Examples on the Ocean Data Labs site.
Leave a Reply
Want to join the discussion?Feel free to contribute!