Loading...

Dynamical and statistical high resolution downscaling approaches for the surface wind

In the framework of the Work Package 4 Atmosphere for Energy from the High Performance Computing for Energy (HPC4E) project, the Centre for Renewable Energy from the Federal University of Pernambuco (CER-UFPE) is working on the development of low computational cost dynamical and statistical downscaling approaches to serve as a reference to assess the results from the HPC approaches which are being developed by the same work package, also by CER-UFPE. The main goal is to describe the behavior of the surface wind (e.g., in the surroundings of an anemo mast) with emphasis to the models’ skill to reproduce the amplitude of variation and frequency/phase structure of the observed wind in the intraday and daily scales.

The dynamical approaches are based on Planetary Boundary Layer (PBL) modelling with the aim of coupling Global Circulation Models (GCMs) output with the local scale. Such approaches are being validated over the Northeast Region of Brazil. Different PBL model families are being assessed for different zones of such region of Brazil. Four families are defined according to the importance of the temperature gradient:

  1. no temperature gradient;
  2. vertical temperature gradient;
  3. horizontal temperature gradient;
  4. horizontal and vertical temperature gradients.

A statistical sensitivity analysis is also made so as to improve the skill of vertical temperature gradient based models. Preliminary results for Paracuru (precisely, at the coastline of the State of Ceará) are presented over a 1-year validation period (observational data at 60 m above ground level for Paracuru are available at the Secretary of Infrastructure from the State of Ceará covering the period from August 2004 to May 2006). For ERA-INTERIM reanalysis at 00 06 12 18 UTC, Figure 1 shows that horizontal temperature gradient based models (see blue dots with labels 12 and 14) have a slightly higher correlation with observations than the simplest approach, the bilinear interpolation over the four ERA-INTERIM grid points closest to the anemo mast (see magenta dot with label 2). On the other hand, horizontal temperature gradient based models are quite better reproducing the standard deviation of observations. For the daily averaged analyses, Figure 2 shows that the bilinear interpolation is harder to be hit. The sensitivity analysis (see cyan dots with labels 8 and 34) only leads to results slightly better than the bilinear interpolation in terms of correlation as well as standard deviation.

The statistical approaches also aim at coupling GCMs output with the local scale. However, statistical models are very sensitive to both GCM output and observational data, different from the dynamical models which are much more sensitive to GCM output than to the observations. This strongly limits the generalization capability over space of such statistical approaches. But, they are usually more skillful for the very local scale. The statistical approaches are also being validated over the Northeast Region of Brazil. In this sense, three main model families are being assessed for different zones of such region of Brazil:

  1. regression models (multilinear regression with or without previous principal component analysis);
  2. classification models (analog based models);
  3. combination models (linear and non-linear combinations over different model outputs).

Preliminary results are also presented for Paracuru, but over a slightly different time series than that used for models validation at Figures 1 and 2. For ERA-INTERIM reanalysis at 00 06 12 18 UTC, Figure 3 shows that both linear and non-linear (neural net based) combinations lead to significant better results than the bilinear interpolation (compare black dot with label 1 to both green dot with label 9 and cyan dot with label 11). In its turn, Figure 4 shows, for the daily scale, the same tendency as in Figure 3.

Next steps on dynamical approaches will be introducing higher resolution effects related to orography, ground aerodynamic roughness and breeze. Concerning statistical approaches, non-linear (neural net based) regression will be checked as well as a spectral analysis to select complementary model outputs.

Fig1

Figure 1. PBL model families based on ERA-INTERIM reanalysis at 00 06 12 18 UTC for Paracuru, State of Ceará, over a 1-year validation period at 60 m above ground level (magenta dots stand for observations –which are the target– and bilinear interpolation over the 4 ERA-INTERIM grid points closest to the anemo mast; black dots stand for no temperature gradient; red dots stand for vertical temperature gradient; blue dots stand for horizontal temperature gradient; green dots stand for horizontal and vertical temperature gradients; cyan dots stand for statistical sensitivity analysis). Note that no statistical corrections were applied to the models output.

 

Fig2

Figure 2. PBL model families based on ERA-INTERIM reanalysis at daily scale for Paracuru, State of Ceará, over a 1-year validation period at 60 m above ground level (color legend is the same as for Figure 1). Note that no statistical corrections were applied to the models output.

 

Fig3

Figure 3. Statistical downscaling approaches based on ERA-INTERIM reanalysis at 00 06 12 18 UTC for Paracuru, State of Ceará, over a 1-year validation period at 60 m above ground level (magenta dot stands for observations which are the target; black dot stands for bilinear interpolation over the 4 ERA-INTERIM grid points closest to the anemo mast; red dots stand for multilinear regression approaches; blue dots stand for analog based models; green dots stand for linear combinations over different model outputs; cyan dots stand for non-linear combinations over different model outputs).

 

Fig4

Figure 4. Statistical downscaling approaches based on ERA-INTERIM reanalysis at daily scale for Paracuru, State of Ceará, over a 1-year validation period at 60 m above ground level (color legend is the same as for Figure 3).

 

Leonardo Aquino, Valentin Perruci and Alexandre CostaCentre for Renewable Energy from the Federal University of Pernambuco (CER-UFPE) – Recife – Brazil

 

This article also appeared in the LinkedIn GroupJoin us!

Other LinkedIn articles:

Link Wind energy in HPC4E
Link Investigation of flame structure of biomass-derived gaseous fuels
Link Finding the "not so easy" oil
Link How supercomputing can help improving the energy sector
Link Effects of fuel composition on biogas combustion in premixed laminar flames
Link New generation subsurface imaging gets a boost from HPC
Link Novel Hybridizable Discontinous Galerkin method paves the way to tackle realistic 3D problems in seismic imaging
Link Improving short-range wind intensity prediction based on multimodel meteorological ensemble forecasts and Genetic Programming
Link Biogas Utilisation for Sustainable Power Generation
Link Preparing the oil and gas industry for the Exaflop era
Link Innovative multiscale numerical algorithms for highly heterogeneous media extended to seismic problems
Link Do we really need exascale computers? Geophysicists say yes, we do
Link Providing resilient executions to the energy simulations