search

openclimatefix / satellite_image_processing

Apache License 2.0
12 stars 3 forks source link

Why does the SEVIRI HRV image occasionally drift ~ 3km south? #1

Open JackKelly opened 5 years ago

JackKelly commented 5 years ago

Why does sat image drop downwards? Does the SEVIRI header / trailer give any info? Check the MSG Level 1.5 Image Data Format Description docs?

This southwards shift (by about 3 km) happens at these times (UTC):

You can see these shifts in this video of EUMETSAT images on YouTube.

JackKelly commented 5 years ago

Hmm, in the Native files headers, LowerSouthLinePlanned and LowerNorthLinePlanned don't appear to change. Here's the header data for every image in the hour of 2019-06-20T10. (Specifically, the header['15_DATA_HEADER']['ImageDescription']['PlannedCoverageHRV'] header)

  datetime LowerSouthLinePlanned LowerNorthLinePlanned LowerEastColumnPlanned LowerWestColumnPlanned UpperSouthLinePlanned UpperNorthLinePlanned UpperEastColumnPlanned UpperWestColumnPlanned
2019-06-20 10:00:00 6961 11136 2768 8335 0 0 0 0
2019-06-20 10:05:00 6961 11136 2767 8334 0 0 0 0
2019-06-20 10:10:00 6961 11136 2766 8333 0 0 0 0
2019-06-20 10:15:00 6961 11136 2765 8332 0 0 0 0
2019-06-20 10:20:00 6961 11136 2765 8332 0 0 0 0
2019-06-20 10:25:00 6961 11136 2764 8331 0 0 0 0
2019-06-20 10:30:00 6961 11136 2763 8330 0 0 0 0
2019-06-20 10:35:00 6961 11136 2762 8329 0 0 0 0
2019-06-20 10:40:00 6961 11136 2761 8328 0 0 0 0
2019-06-20 10:45:00 6961 11136 2760 8327 0 0 0 0
2019-06-20 10:50:00 6961 11136 2760 8327 0 0 0 0
2019-06-20 10:55:00 6961 11136 2759 8326 0 0 0 0
2019-06-20 11:00:00 6961 11136 2758 8325 0 0 0 0
JackKelly commented 5 years ago

The trailer contains ['15TRAILER']['ImageProductionStats']['ActualL15CoverageHRV'] (as opposed to the planned data in the header). But the trailer CoverageHRV stats appears identical to the planned stats.

JackKelly commented 5 years ago

The only trailer data which looks vaguely relevant is:

'MisregistrationResiduals.NorthSouthMaxDeviation. For 2019-06-20 10:00 vs 10:20:

2019-06-20 10:00 array([-0.40437472, -0.15457937, -0.19689183, -0.17919952, 0. ,

  1. , -0.23171186, -0.23598963, -0.2490123 , -0.27155465, 0.4767555 , 0.20391025], dtype=float32),

2019-06-20 10:20 array([-0.6090172 , -0.5180342 , -0.40122026, -0.61327034, 0. ,

  1. , -0.3549063 , -0.43837735, -0.4319517 , -0.3388781 , -0.46477982, -0.34633762], dtype=float32))
JackKelly commented 5 years ago

I've emailed EUMETSAT to ask for help!

(If they can't help then I should check the PNG images I downloaded for 2019-06-20 10:00 to 11:00. If those images contain the same southwards shift then it's likely that EUMETSAT don't capture the southwards shift and we either live with it (probably fine), or we manually clean the data.)