MET Office data API and hours of sunshine forecast

[dnb]

Has anyone played with the Met office data API? I am looking and can't see anything that can be used as a predictor for hours of sunshine other than perhaps visibility (but not sure what the units are!) Was hoping for a cloud cover number but I can't see it in the 3 hour spot forecasts.

[Radian]

11 minutes ago, dnb said:

can't see anything that can be used as a predictor for hours of sunshine

Might Solcast be of use to you?

[SteamyTea]

Many years ago, @Ed Davies and myself, looked into predicting light levels.  After looking at thousands of data points, we came to the conclusion that it was not very accurate.  The best correlation was wind direction.

I did think at the time that cloud base temperature, or lack of cloud base temperature, may be a good indictor, but never followed it up.

This came about as I was trying to find the algorithm that my cheap weather station uses and is spookily accurate.

Bought an IR temperature sensor for my RPi, think it is still in a packet somewhere.

Maybe this year I shall get around to doing it.

[Marvin]

Hi @dnb

Can I ask what the info is for. We live at the bottom of the island. If you require detailed info for calculation PV generation or solar gain I have no quick answers but have some suggestions.

 

Marvin

 

[SteamyTea]

37 minutes ago, Marvin said:

but have some suggestions

Let's hear them.

'One Point' weather prediction would be a useful tool, and with today's cheap hardware, should be easy to implement.

[Marvin]

Reading the original post first I would like to clarify:

 

If the aim is a prediction for the next 3 hours I can't help. If your looking for general predictions of daily sunlight over a month I can do this.

 

Predicting sunlight is easy on fully cloudy days and completely clear sky days otherwise good luck! 

 

The value of any information can only be decided when the purpose is confirmed. 

 

If I was anal about needing the house to respond to sunlight I would use sensors.

If it was solar gain deflection I would use sensors. 

If it was to calculate solar gain or PV output this can be done.

If you want to know exactly how long sunlight hits a particular spot on your building you can install a 12PV panel obscured to only produce enough power when lit by sunlight (or add a relay that need a minimum voltage to start the meter)and a ayron JR-HM001 Snap in Mechanical Hour Meter Rectangular Hour Meter for DC 6V to 80V and an appliance to use up the power produced.

 

Hope this helps someone.

 

We live on a windy island, where expected hit and miss weather days are frequently wrong.

 

M

 

 

Edited January 1, 2023 by Marvin
further thought

[SteamyTea]

I have found that weekly predictions are pretty accurate, daily a lot less so, but can be useful if standard error is taken into account.

[Marvin]

1 hour ago, SteamyTea said:

I have found that weekly predictions are pretty accurate, daily a lot less so, but can be useful if standard error is taken into account.

Ditto.

[dnb]

Sorry for the very late replies. Sometimes work drags me away from house building for extended periods.

 

On 31/12/2022 at 22:13, Radian said:

Might Solcast be of use to you?

I think so! It certainly answers the solar PV aspect of the question.

 

On 01/01/2023 at 12:16, Marvin said:

Can I ask what the info is for.

You can certainly ask! From other posts you made here we appear to have similar interests in using data for efficient housing. 

 

I am looking at using several parameters from weather forecast data (the MET office API appears to give the most likely state of their monte-carlo weather model at 3 hour intervals for the next week) to decide on an energy optimisation strategy for the next few days.  Amount of sunshine is just one (probably derived or inferred) parameter of many.  E.g. a rule set such as "It's going to be overcast all week, so keep the hot water tank topped up with cheap-rate electricity until Tuesday when it should be sunny."

 

This is the data collection stage so I can prove algorithms against it and with other data I am collecting measure how well the house performs against its heat loss model. I also have a set of candidate algorithms to model energy demands. I believe I can express the system as something related to a Kalman filter (at it's simplest this is an iterative form of least squares regression) that will answer the question of what action the house should take to stay comfortable for us using the least energy (on average) for the next N hour time period given a particular set of inputs and a current state, where N will be a significant fraction of a day.

 

It is definitely NOT about immediate responses since the best "nowcast" for the weather is to look outside!

 

On 01/01/2023 at 14:04, SteamyTea said:

I have found that weekly predictions are pretty accurate, daily a lot less so, but can be useful if standard error is taken into account.

True. This is why I am hopeful the Kalamn filter type approach will work since it smooths the daily data and can account for random errors. (But far less good at bias errors!)

[SteamyTea]

4 hours ago, dnb said:

True. This is why I am hopeful the Kalamn filter type approach will work since it smooths the daily data and can account for random errors. (But far less good at bias errors!)

The analysis side is similar to a Bayesian result i.e. if this has happen, what is the probability that another thing is true.

Weather prediction uses, and is checked with Bayesian statistics.

Another way is to make up some probability tables from known variables and just look up results.  That can be quite effective and uses little computational power.

[Marvin]

Hi @dnb

 

Our weather is often so variable during a day ( in relation to sun and wind) that we've given up trying to predict it.  Using brush strokes has been our best strategy. Using the results from the PGIS on a monthly basis has given us a general picture of what to expect regarding PV energy production.

 

Here's our results for last year:

 

Year               Settings

2022           ERA5 6% loss    Running     Act         sub      diff       %

                         kWh               Total                                            

Jan 2022          216                 216          204        204     -12     94.4%
Feb 2022         316                 532          294        498     -22     93.6%
Mar 2022         552              1,084          601       1099      49     101.4%
Apr 2022         774               1,858          744       1843    -30     99.2%
May 2022        809              2,667          818       2661        9     99.8%
Jun 2022         821              3,488          902      3562       81     102.1%
Jul 2022          836              4,324          816      4379      -20     101.3%
Aug 2022        724               5,048         770       5149       46     102.0%
Sep 2022        582               5,630         536      5603     -46     99.5%
Oct 2022        393               6,023         362      5965      -31     99.0%
Nov 2022        252               6,275         296      6262       44     99.8%
Dec 2022        190               6,465          171      6433       -19     99.5%

 

So in the year we produced 99.5% of what was expected.

 

Temperatures are reasonably as expected except for the wind chill factor!

 

We think that the winter production is lower than estimated because we are in a shallow with higher ground to the east and west so during the winter a bigger percentage is not produced.

 

Again, if we decide to cook a cake for instance, because of the good insulation this can overheat the bungalow.

 

Solar gain?

 

Vacuuming!

 

Ironing!  

 

Jolly good argument!

 

All heat the house.

 

Too many variables for my liking. 

 

I am now concentrating on reducing the purchased overnight energy using battery storage.