Forecast Bar Weather Powered By Forecast Io 2 6

Get the forecast for today, tonight & tomorrow's weather for Lisle, IL. Hi/Low, RealFeel®, precip, radar, & everything you need to be ready for the day, commute, and weekend! Dallas sees 80-plus. Be prepared with the most accurate 10-day forecast for Huron, SD with highs, lows, chance of precipitation from The Weather Channel and Weather.com.

pvlib python provides a set of functions and classes that make it easyto obtain weather forecast data and convert that data into a PV powerforecast. Users can retrieve standardized weather forecast data relevantto PV power modeling from NOAA/NCEP/NWS models including the GFS, NAM,RAP, HRRR, and the NDFD. A PV power forecast can then be obtained usingthe weather data as inputs to the comprehensive modeling capabilities ofpvlib python. Standardized, open source, reference implementations offorecast methods using publicly available data may help advance thestate-of-the-art of solar power forecasting.

pvlib python uses Unidata’s Siphon library to simplify accessto real-time forecast data hosted on the Unidata THREDDS catalog. Siphon is great forprogramatic access of THREDDS data, but we also recommend using toolssuch as Panoplyto easily browse the catalog and become more familiar with its contents.

  • Get the forecast for today, tonight & tomorrow's weather for Lisle, IL. Hi/Low, RealFeel®, precip, radar, & everything you need to be ready for the day, commute, and weekend! Dallas sees 80-plus.
  • Specific (deterministic) snow accumulations for locations in the United States can be obtained from the National Weather Service's National Digital Forecast Database. NCEP monitoring of ongoing or imminent (up to six hours in the future) Hazardous Winter Weather can be found at SPC Mesoscale Discussion Link.

We do not know of a similarly easy way to access archives of forecast data.

This document demonstrates how to use pvlib python to create a PV powerforecast using these tools. The forecast and forecast_to_power Jupyter notebooksprovide additional example code.

Warning

The forecast module algorithms and features are highly experimental.The API may change, the functionality may be consolidated into an iomodule, or the module may be separated into its own package.

Note

This documentation is difficult to reliably build on readthedocs.If you do not see images, try building the documentation on yourown machine or see the notebooks linked to above.

Accessing Forecast Data¶

The Siphon library provides access to, among others, forecasts from theGlobal Forecast System (GFS), North American Model (NAM), HighResolution Rapid Refresh (HRRR), Rapid Refresh (RAP), and NationalDigital Forecast Database (NDFD) on a Unidata THREDDS server.Unfortunately, many of these models use different names to describe thesame quantity (or a very similar one), and not all variables are presentin all models. For example, on the THREDDS server, the GFS has a fieldnamedTotal_cloud_cover_entire_atmosphere_Mixed_intervals_Average,while the NAM has a field namedTotal_cloud_cover_entire_atmosphere_single_layer, and asimilar field in the HRRR is namedTotal_cloud_cover_entire_atmosphere.

pvlib python aims to simplify the access of the model fields relevantfor solar power forecasts. Model data accessed with pvlib python isreturned as a pandas DataFrame with consistent column names:temp_air,wind_speed,total_clouds,low_clouds,mid_clouds,high_clouds,dni,dhi,ghi. To accomplish this, we use anobject-oriented framework in which each weather model is represented bya class that inherits from a parentForecastModel class.The parent ForecastModel class contains thecommon code for accessing and parsing the data using Siphon, while thechild model-specific classes (GFS,HRRR, etc.) contain the code necessary tomap and process that specific model’s data to the standardized fields.

The code below demonstrates how simple it is to access and plot forecastdata using pvlib python. First, we set up make the basic imports andthen set the location and time range data.

Next, we instantiate a GFS model object and get the forecast datafrom Unidata.

It will be useful to process this data before using it with pvlib. Forexample, the column names are non-standard, the temperature is inKelvin, the wind speed is broken into east/west and north/southcomponents, and most importantly, most of the irradiance data ismissing. The forecast module provides a number of methods to fix theseproblems.

Much better.

The GFS class’sprocess_data() method combines these stepsin a single function. In fact, each forecast model classimplements its own process_data method since the data from eachweather model is slightly different. The process_data functions aredesigned to be explicit about how the data is being processed, and usersare strongly encouraged to read the source code of these methods.

Bar

Users can easily implement their own process_data methods oninherited classes or implement similar stand-alone functions.

The forecast model classes also implement aget_processed_data() method thatcombines the get_data() andprocess_data() calls.

Cloud cover and radiation¶

All of the weather models currently accessible by pvlib include one ormore cloud cover forecasts. For example, below we plot the GFS cloudcover forecasts.

Forecast Bar Weather Powered By Forecast Io 2 6 Ft

However, many of forecast models do not include radiation components intheir output fields, or if they do then the radiation fields suffer frompoor solar position or radiative transfer algorithms. It is often moreaccurate to create empirically derived radiation forecasts from theweather models’ cloud cover forecasts.

pvlib python provides two basic ways to convert cloud cover forecasts toirradiance forecasts. One method assumes a linear relationship betweencloud cover and GHI, applies the scaling to a clear sky climatology, andthen uses the DISC model to calculate DNI. The second method assumes alinear relationship between cloud cover and atmospheric transmittance,and then uses the Campbell-Norman model to calculate GHI, DNI, andDHI [Cam98]. Campbell-Norman is an approximation of Liu-Jordan [Liu60].

Caveat emptor: these algorithms are not rigorously verified! Thepurpose of the forecast module is to provide a few exceedingly simpleoptions for users to play with before they develop their own models. Westrongly encourage pvlib users first read the source code and secondto implement new cloud cover to irradiance algorithms.

The essential parts of the clear sky scaling algorithm are as follows.Clear sky scaling of climatological GHI is also used in Larson et. al.[Lar16].

The figure below shows the result of the total cloud cover toirradiance conversion using the clear sky scaling algorithm.

The essential parts of the Campbell-Norman cloud cover to irradiance algorithmare as follows.

The figure below shows the result of the Campbell-Norman total cloud cover toirradiance conversion.

Most weather model output has a fairly coarse time resolution, at leastan hour. The irradiance forecasts have the same time resolution as theweather data. However, it is straightforward to interpolate the cloudcover forecasts onto a higher resolution time domain, and thenrecalculate the irradiance.

Users may then recombine resampled_irrads and resampled_data usingslicing pandas.concat() or pandas.DataFrame.join().

We reiterate that the open source code enables users to customize themodel processing to their liking.

Lar16

Larson et. al. “Day-ahead forecasting of solar power outputfrom photovoltaic plants in the American Southwest” RenewableEnergy 91, 11-20 (2016).

Cam98

Campbell, G. S., J. M. Norman (1998) An Introduction toEnvironmental Biophysics. 2nd Ed. New York: Springer.

Liu60

B. Y. Liu and R. C. Jordan, The interrelationship andcharacteristic distribution of direct, diffuse, and total solarradiation, Solar Energy4, 1 (1960).

Weather Models¶

Next, we provide a brief description of the weather models available topvlib users. Note that the figures are generated when this documentationis compiled so they will vary over time.

GFS¶

Weather - Dark Sky

The Global Forecast System (GFS) is the US model that provides forecastsfor the entire globe. The GFS is updated every 6 hours. The GFS is runat two resolutions, 0.25 deg and 0.5 deg, and is available with 3 hourtime resolution. Forecasts from GFS model were shown above. Use the GFS,among others, if you want forecasts for 1-7 days or if you want forecastsfor anywhere on Earth.

HRRR¶

The High Resolution Rapid Refresh (HRRR) model is perhaps the mostaccurate model, however, it is only available for ~15 hours. It isupdated every hour and runs at 3 km resolution. The HRRR excels insevere weather situations. See the NOAA ESRL HRRR page for more information. Use theHRRR, among others, if you want forecasts for less than 24 hours.The HRRR model covers the continental United States.

RAP¶

The Rapid Refresh (RAP) model is the parent model for the HRRR. It isupdated every hour and runs at 40, 20, and 13 km resolutions. Only the20 and 40 km resolutions are currently available in pvlib. It is alsoexcels in severe weather situations. See the NOAA ESRL HRRR page for more information. Use theRAP, among others, if you want forecasts for less than 24 hours.The RAP model covers most of North America.

NAM¶

Forecast Bar Weather Powered By Forecast Io 2 6 0

The North American Mesoscale model covers, not surprisingly, NorthAmerica. It is updated every 6 hours. pvlib provides access to 20 kmresolution NAM data with a time horizon of up to 4 days.

NDFD¶

The National Digital Forecast Database is not a model, but rather acollection of forecasts made by National Weather Service officesacross the country. It is updated every 6 hours.Use the NDFD, among others, for forecasts at all time horizons.The NDFD is available for the United States.

PV Power Forecast¶

Finally, we demonstrate the application of the weather forecast data toa PV power forecast. Please see the remainder of the pvlib documentationfor details.

Forecast Bar Weather Powered By Forecast Io 2 6 Days

Now we plot a couple of modeling intermediates and the forecast power.Here’s the forecast plane of array irradiance…

…the cell and module temperature…

…and finally AC power…