OZ-ENERGY-ANALYSIS.ORG   -   open science for the new millennium

THE STORIES   |   DATA   |   ANALYSIS   |   MODELS   |   LITERATURE   |   DISCUSSIONS

status: [2nd March 2011] awaiting overhaul

Wind Speed and Direction Data

• Introduction
• Data Selection - BtCC - Round 1

Introduction

This is the wind data page, and is currently focused on The BtCC. Earlier work and data (the initial five) is here in supplementary materials.

The wind is a major source of renewable power, due principally to its lower cost relative to other renewable energy sources (putting aside Hydro Power). It is also a highly variable source of power. The Australian Bureau of Meteorology (BoM) has a large store of wind data.

While not entirely straightforward, BoM wind data can be used to estimate the (variability of) power that might be obtained if a Wind Farm were built in the vicinity of the BoM station. Unfortunately, BoM stations are not generally located in the places that Wind Farm builders would choose; BoM stations are generally located in or near towns and airports, while the ideal positioning for a wind farm is on the top of a ridge that runs perpendicular to the prevailing wind. None-the-less, we work with what we have keeping in mind the limitations.

The primary use of Wind Data in the OzEA project is to examine spatial smoothing, that is, the extent to which geographically distributed wind farms balance out each-others lull periods.

Figure (DWSF) 01: Locations of the 825 BoM stations with wind speed data covering 2009. This small image file is for fast page load, and links to a pdf file (only 90 KB). This PDF file allows you to zoom in and in, until you can see the station ID numbers. In some cases stations are so close that the numbers overlap - a problem we may attempt to resolve at a later date.

If you go to: http://www.bom.gov.au/climate/cdo/about/sitedata.shtml and select "Wind Speed" you can obtain a list of 1838 BoM stations [June 2010]. These are interesting to look at in general, but here we restrict immediately to those stations with data covering 2009; this reduces the list to 825 stations, as plotted in the Figure above (be sure to click on the image so as to obtain the pdf version - into which you can zoom and zoom).

Data Selection - BtCC - Round 1

This work is complete, as below, for now [July 2010]. However, the purpose of the Round-1 data selection is as 'a start'. As work proceeds the weakness and limitations of this particular selection can be expected to become more apparent, and in time we will return to this step and make an improved selection for processing through the workflow we are currently developing. That is, the current analysis is focused on: (i) understanding the data, and (ii) developing a pipeline for data processing and analysis. Of course, the Round-1 work will also provide a first tentative look at the results of this project.

The following figure (and linked PDF) now represents the conclusion of the (round 1) data selection. We have obtained the Wind Speed data for the 40 selected BoM stations (red crosses, and as listed in the table following) and these data are now [July 2010] being examined and characterised. This selection was aided by using the generators data from which we extracted the locations of the current and proposed Wind Farms (lime green and cyan blue respectively), and with a lower cutoff of 20 MW (exception made for Windy Hill in QLD).

Figure (DWSF) 03: The Round 1 selection of 40 BoM stations with wind speed data covering 2009 (red crosses). Shown also are the current and proposed sites for Wind Farms of capacity 20+ MW (an exception for Windy Hill in Qld, which is only 12). The magenta crosses represent BoM stations that were candidates for consideration, but which do not form part of the current data processing. This small image file (for fast page load) links to a (100 KB) pdf file. This PDF file allows you to zoom in and in until you can see the station ID numbers.

Using the PDF it was possible to identify (somewhat roughly) a selection of BoM stations that trace out the coast (and now include some inland areas) from Broome to Cooktown (and a little further). Any given latitude and longitude can easilly be copy and pasted into Google Earth, and this was the primary mechanism for sanity checking the selections and comparing them with nearby alternatives. In this way the Round One Selection was developed. Note that the locational precision given in the BoM data is not usually high enough to give a direct hit onto the stations. In addition to the table following (the selection), a further table listing those BoM stations that were part of these considerations, but that did not end up in the following selection, can be found on supplementary materials page.

Table DWST01: The Round One Selection; 40 AWS Wind Speed measuring stations from Broome to Cooktown. Note that the locational precision given in the BoM data is not high enough to give a direct hit onto the stations.

[23rd July 2010] Here is the base round-one wind speed data for the above BoM stations for 2009:

• BtCC_R1_BASE_hr_2009.txt.gz [332 KB] - (csv version)

The data for all stations is given in a single flat file, in the OzEA 'line per day' format; the data points correspond to hour values (i.e. 12am, 1am, ... 11pm) in AEST, have units of m/s, and with missing data given as 'NaN' (Not a Number). As described in what follows the above data includes interpolated values for isolated missing data points.

For anyone who is interested, the following sequence of steps and files outline the process of transforming the data provided by the BoM into the data file given above.

First, here is a zip file containing 40 individual files (for the 40 BoM stations), each giving the June 2009 data, in half hourly steps, as given by the BoM: BtCC_sample_BoM_data_v1.zip [280 KB].

Second, here is the very same data transformed into the 'Line per Day' format that we use here for such data: BtCC_sample_LpD_data_v1.zip [58 KB].

Third, here is the same data again, now expanded to include all of 2009 and consolidated into a single flat file: BtCC_R1_RAW_30min_2009.txt.gz [510 KB gziped flat file]. This is the raw data; it is numerically identical to the BoM data, but in a different format.

And now the above data is processed to give the Base Data according to the following three step process: (i) using this half hourly data interpolate any single missing value (as the average of the adjacent values); (ii) discard the half hour values, leaving only the on-hour values; and (iii) interpolate again as above, but now with the hourly data, before writing out.

Wind Data Characterisation

[Jump to bottom]

7 OzEA_DBWD0007

John Morgan Subject: Broome to Cooktown Date: 2010-06-10 (at 00:33:51)

Here's an early entry in the Broome to Cooktown challenge. It may be running in the wrong direction, and probably doesn't have much staying power.

I was looking at the Cleve_Aerodrome data, since it seems well behaved. Plots are in Fig 1, attached.



Data is for one year, and 250 hours. There seems to be a daily cycle, so lets look at the autcorrelation. I calculated the autocorrelation function as the Pearson correlation coefficient for the windspeed array against itself with a time lag, using a periodic boundary condition, ie, assuming each year is a repeat of the last. I can get six months of autocorrelation this way, see Fig 2.



Looking in at the ACF the daily correlation is nicely resolved. What I wanted to see was the underlying decorrelation. Wind speeds at this station appear to decorrelate over about four days. Looking at just the daily peaks, the decorrelation time constant is about 19 hours (running an exponential through the peaks), so call it 60 hours for decorrelation. (I tried to filter out the periodic component but wasn't very successful - its been a long time since I've done this sort of data analysis.)

I take this decorrelation time as the synoptic timescale. I assume a large scale weather system takes approximately 60 hours to pass across this monitoring station. How fast is it travelling?

The average wind (ground)speed over the year was 5.5 m/s. Assume this is a lower bound for the weather system speed. Then we can estimate the synoptic correlation length constant as being not smaller than ~380 km, or, the size of a weather system as not smaller than 1200 km.

This is quite consistent with the correlation length constant reported by Kempton et al. in PNAS as ~430 km, as determined from wind farm power output along the US eastern seaboard.

Windspeed at modest height is estimated by the Hellman equation as log(v) = a . log(h), with a value of a~0.3 typical. Assuming the BOM station heights are 10 m, the velocities are 2, 4 and 8 times faster at 100m, 1000m, and 10000m. (Why not extrapolate beyond the bounds of validity - lets go for gold!) So the synoptic weather system size is about a small integer multiple of my minimum estimate.

So, from a single weather station, I've deduced the spatial smoothing scale is at a minimum ~1200 km, consistent with the PNAS study, though more likely several times that. Lots of dodgy assumptions, but I'll bet its about right.

Another quick point from the ACF - any spatial smoothing over a large scale is going to have to take account of the time differences between regions,


fc: thanks John, and a note to all that you get figs up by mailing them to me (for now)

22 OzEA_DBWD0022

John Morgan Subject: exploration of wind properties Date: 2010-06-24 (at 17:05:49)

I was noodling around with the Cleve Aerodrome data I analysed above since I wanted to answer the questions I raised at the top of the thread, namely, does the data conform to the Rayleigh distribution, and what does the distribution of lulls look like? ie, how often does this station show extended durations of low wind.

This is somewhat tangential to the aim here, but, I think its worthwhile doing a deep dive into wind at a single site because it should help inform the analysis when putting multiple sites together. And because I'm curious.

So, are the windspeeds Rayleigh distributed? You get the Rayleigh distribution if the x- and y- components of the wind velocity are (i) independent, and (ii) normally distributed. It would be handy if true, since a simple closed form expression for the wind probability makes for easy analysis. The distribution is often used for windspeeds, so I'd like to see how well it stacks up.

First up, are the x- and y- components independent? I plotted all hourly wind vectors on polar axes, to show speed and direction (Fig 3).



Units are degrees from true north, and m/s.

It looks pretty uncorrelated to me. I expected to see a prevailing wind, especially so close to the coast, but there is no evidence at all of such.

Next, lets look at the velocity distribution (Fig 4):



The windspeeds look to be pretty well fit by the Rayleigh distribution (red line, least squares fit of Rayleigh function to data), except at the lowest speeds. The number of hours of zero wind is much higher than the model. So maybe using the Rayleigh distribution would miss the lulls. But up to about 2.5 m/s counts are lower than the model by an amount that compensates. So unless you're interested in the detail at the lowest speeds, the Rayleigh distribution looks good.

Another way to visualize this is sort all the data into order and plot them over the interval [0,1). For the Rayleigh distribution this should be fit by sigma*sqrt(-2*ln(1-x)). This shows the quality of the fit more clearly than the histogram. The red line is the same distribution as in Fig 4. The fit is good except at the lowest speeds (Fig 5).



Finally, how are those low wind events distributed? Are they grouped together into extended lulls, which would be a problem for power production? Our are they shortlived transient lows, which aren't?

I counted the number of lulls of different durations and plotted them in Fig 6. A 'lull' is a group of adjacent windspeed measurements less than some threshold. Lulls that are completely becalmed are interesting, but if your windfarm can't generate power below 3 m/s, then thats the threshold of interest. I plotted thresholds of 0, 3 and 6 m/s, though 6 m/s is not really a lull.



So for instance, there were 11 occasions in 2009 where the wind at this site was less than 3m/s for (exactly) 9 straight hours. There was one occasion where it was less than 3 m/s for 21 hours. In theory, it should be possible to calculate the 3 m/s curve from the Rayleigh distribution, but not the 0 m/s curve.

So, for this site,

• there appears to be no prevailing wind direction,
• the Raleigh distribution is a very good predictor of windspeeds except at low windspeeds
• but this probably doesn't matter for power and power outage modelling
• which is good news for modelling power generation if you know the turbine characteristic
• windspeeds below 3 m/s for longer than, say, six hour s, would happen a number of times a month, and might last up to about 20 hours.

Things I'm interested in looking at next are correlations between two sites and whether the lull distribution for a site can be calculated from the Rayleigh parameters.

[Show Full Lists]

 

  TOP   |   REFRESH PAGE   |   DATA

fc - 16th June 2010