A unified high-resolution wind and solar dataset from a rapidly updating numerical weather prediction model

Highlights

A unified wind/solar dataset is derived from hourly forecast model guidance.

The 3-km resolution model incorporates advanced data assimilation and model physics.

80-m wind verification indicates a bias of less than 1 m s−1 at a site in Colorado.

Spatial patterns of 80-m wind and solar resource agree well with previous studies.

Abstract

A new gridded dataset for wind and solar resource estimation over the contiguous United States has been derived from hourly updated 1-h forecasts from the National Oceanic and Atmospheric Administration High-Resolution Rapid Refresh (HRRR) 3-km model composited over a three-year period (approximately 22 000 forecast model runs). The unique dataset features hourly data assimilation, and provides physically consistent wind and solar estimates for the renewable energy industry. The wind resource dataset shows strong similarity to that previously provided by a Department of Energy-funded study, and it includes estimates in southern Canada and northern Mexico. The solar resource dataset represents an initial step towards application-specific fields such as global horizontal and direct normal irradiance. This combined dataset will continue to be augmented with new forecast data from the advanced HRRR atmospheric/land-surface model.

Keywords

  • Wind resource;
  • Solar resource;
  • NWP forecast;
  • Unified wind/solar

1. Introduction

Today’s global economy depends to a large extent upon a reliable electricity generation and distribution system. In the United States (US) and indeed around the world, traditional power generation technologies such as coal, natural gas, and nuclear energy are gradually being supplemented, and in some cases replaced, by renewable energy generation systems [2] and [30]. Advancing technology, as well as growing concern over anthropogenic global climate change, is accelerating this transition to increasingly affordable new power-generation systems.

While more desirable in terms of its smaller anthropogenic footprint, renewable energy generation such as wind, solar, tidal, and geothermal power rely on geophysical phenomena that are variable in space and intermittent in time, posing unique challenges to the generation and distribution system. Among renewable resources, the US Department of Energy (DOE) has highlighted wind energy in particular, setting a goal of achieving 20% of the nation’s electrical energy from wind by the year 2030 [32] and [31]. While less variable diurnally than solar energy, wind energy production is still highly dependent upon favorable weather conditions. Grid operators must anticipate so-called “ramp events”, in which the renewable resource (wind or solar) undergoes a large change (either positive or negative) in a short period of time [20].

Since power-generation infrastructure must generally be fixed in location, it is very important for decision makers to have access to the most accurate resource estimates achievable. For wind, resource assessments are typically done using the “measure-correlate-predict” (MCP) method, wherein a short period of wind tower measurements is correlated with a long-term near-surface wind record in order to predict the wind climatology at turbine height [7]. The observation-based turbine-level wind datasets used to identify potential sites for wind power installations are typically brief in duration, and more importantly are very sparse in their spatial coverage. Furthermore, one of the largest practically accessible wind resources within US territory is offshore of the East Coast (particularly in the vicinity of New England; [24], [27] and [3]). Observations in this area are very limited. Satellite-based synthetic aperture radar measurements (e.g. [22]), have shown promise for oceanic regions; however, the associated retrieval techniques involve many assumptions.

Observation-based solar resource assessments take advantage of long-term radiation measurement networks such as the 14-station SURFRAD/ISIS network in the US [4]. Other observation sources include the Cooperative Networks for Renewable Resource Measurements (CONFRRM; [35]), the University of Oregon Solar Radiation Monitoring Laboratory [36], a number of state monitoring networks (such as the Illinois Climate Network; [37]), and observations supporting solar development by private companies. Complementary satellite observations, conventional surface observations, and model analyses and forecasts [26] also play a role in solar resource assessment.

In addition to a number of resource datasets developed by private companies (e.g., 3Tier, SOLARGIS, Meteonorm), the DOE’s National Renewable Energy Laboratory (NREL) has developed wind and solar resource maps for the period since the early 1990s. Maps of the long-term average 80-m wind speed over the contiguous US (CONUS), with 2.5-km horizontal resolution, were derived using fine-resolution simulations and validation-based bias corrections [12]. In addition, the National Solar Radiation Data Base (NSRDB), containing data for over 1400 stations, has been developed based on a number of statistical models [34]. Extending this effort for wind resources, Draxl et al. [10] have recently produced a grid integration dataset for wind energy, referred to as the WIND Toolkit, which includes model-derived meteorological data and simulated forecasts for over 100 000 land-based and offshore wind power production sites.

For grid integration studies in a scenario with significant penetration of both wind and solar, it is important to have access to time-matched, gridded wind and solar data, which in turn determine wind and solar generation profiles (e.g., [25] and [19]). While NREL’s wind and solar datasets were derived using state-of-the-art statistical and modeling techniques based on available observations, they have been derived independently of one another, meaning that the wind and solar resource are not necessarily meteorologically consistent when combined. That is, in regions with few observations, the wind and solar resources could evolve independently of one another, without necessarily representing a meteorologically realistic state. In this paper, we present a unified wind and solar dataset derived from a real-time, hourly-updating numerical weather prediction (NWP) model developed by the National Oceanic and Atmospheric Administration (NOAA). Modern NWP systems continue to have both known and unknown errors associated with their data assimilation and modeling components, but they are able to perform increasingly well for diverse meteorological situations (e.g., [6]). Our goal is to compare these new results with the NREL maps over the CONUS land area and offshore regions, and to demonstrate the expanded utility of a physically consistent hourly solar and low-level wind dataset using recent NWP refinements.

Section 2 describes background information for our study: the configuration of the NWP model and its changes during the 3-year period examined here, and some objective statistical verification of the solar and low-level wind forecasts. Section 3 summarizes the development of the archive of model output used for this study, and describes the procedures for calculating the metrics to be shown. Section 4 presents low-level wind results, and Section 5 presents solar results. These results are summarized and discussed in Section 6.

2. Background on the weather model/assimilation system

In this section, we provide background on the NWP system configuration used for this study, and present objective verification of the model’s low-level wind and solar forecasts.

2.1. Hourly updating numerical weather prediction system – the HRRR

The NOAA Earth System Research Laboratory (ESRL) Global Systems Division (GSD) has developed a 13-km horizontal grid NWP model called the Rapid Refresh (RAP; [6]). The RAP is run every hour, out to 21 forecast hours, over a domain that covers all of North America. The RAP replaced the earlier smaller domain Rapid Update Cycle (RUC; [5]) as NOAA’s operational rapidly-updating forecast system in May 2012. Since 2010, a nested version of the RAP on a 3-km horizontal grid has also been run hourly in experimental mode at GSD, over a domain the covers the entire CONUS; this nested version of the RAP is called the High-Resolution Rapid Refresh (HRRR). The model domains are shown in Fig. 1. An older version of the HRRR was implemented operationally within NOAA in September 2014; however, all of the results presented here are based on the experimental HRRR at ESRL/GSD. The HRRR and RAP use specially developed versions of the WRF-ARW model [28].

Be the first to comment

Leave a Reply

Your email address will not be published.


*