Development update

Paul

For the dynamic range palettes, this is a very unfinished feature but already shows value in expanding the dynamic range around dusk/dawn of visible satellite imagery.

Here is static palette range - note the extremely dark/dusk values off Nova Scotia to Sable Island:

Now here is the same visible satellite imagery frame when zoomed into that area with dynamic range palette on:

Notice how all that variation is from 0.0 - 0.1, scaled from the 1.3 max range in the first screenshot, where the same variation is completely drowned by the full static range and therefore imperceptible.

I have plans to experiment with a spatially variant mode of dynamic palette range, where different areas of the screen get their own dynamic ranges depending on the content therein. I have no idea what this will look like but it could be very useful for visible satellite imagery based on the value already demonstrated.

Paul

The next build has improved SDF styling semantics enough to finally semi-operationally release the experimental 2023 SDF line styling special effects for e.g. seamless blending road lines, never released until now but previewed in the first-gen forum in summer 2023, screenshots reproduced earlier in this thread. These are visually optimized techniques for crisp rendering of contrast-boosted/blurred background underneath-line scene color, for minimum visual clutter, using distance fields. The result is both an aesthetic optimization, and an informational one. When the lines clutter the data less, it becomes possible to reduce the map size without losing as much perceptive visual information. This concept will be exploited with the 8x multiple map mode (1, 2, 4, 8x rapid switching) planned for implementation shortly. But the SDF lines are looking quite nice.

Paul

I have initial positive results with multi-domain rendering. Having MESO 1, MESO 2, CONUS, and Full Disk all concurrently rendering on the map in their own full resolutions, with realtime per-pixel evaluation and zero naive, ugly overlaying like how first-gen lets you have 2x domains with the EGR add-on - that is cheap one on top of another, not true per-pixel selection of the domains by priority. E.g. you can have transparent palettes such as for IR and they do not overlap, multiplying transparent values, like you see with first-gen. Instead the GPU is selecting only one domain to render at each pixel so the original projection grayscale values of each seamlessly fuse into one coherent scene. The screenshot below shows both MESO domains loaded, but to really see this it requires a video, because the spatial resolutions are mostly the same - only while looping, you see the higher temporal resolution of MESO over CONUS (1-min vs 5-min) and CONUS over Full Disk (5-min over 10-min). I think this is a meaningful advancement in using GPU for weather visualization since it eliminates the "which single domain to load" limitation. This supports up to 4x concurrent domains per layer currently, but I could possibly bump that to 8, so if someone purchases both GOES 18 and 19 streampacks, they could have all 8x domains open at the same time, covering half the planet. That's a lot of data but I am testing special GPU algorithms to get this multi-domain rendering working without exploding the frame time. This is a highly novel ability I am not sure if major TV system have.

Paul

Concurrent multi-domain rendering is a quintessential concept/ability in WSV3 Tactical 2025 engine. The planned RadarHalo Polyradial Mosaic system is merely an extension to this ability, where polar Level 2 single-site domains will blend into MRMS reflectivity at range. Beyond that and the inherently multi-domain GOES-R cases, the third usage for multi-domain rendering is with future precip-typing datasets (or any NWP data) where a hi-res subset grid supplies focused data for e.g. CONUS but inset into global e.g. GFS domain for a seamless gap-free global lower res fallback.

radioman10

[unknown] Looking great. Thank you for your work!

Paul

The multidomain rendering update will be released soon - not impossibly later today.

I believe this is unique and not available elsewhere, even in TV systems: full-resolution original-projection GPU-based realtime display of multiple original GOES-R domains with spatial resolution based selection for seamless multidomain fusing, but without any reprojection of underlying fullres data, running at 120fps on commodity hardware. And there's even an optimization where tiles from loser-domains where fully obscured by winner-domains are excluded from loading, so the maximum bandwidth/overhead hit is limited and reasonable.

I did a whole refactor/partial rewrite of the raster tile streaming/ingest/rendering code after the multiple domain dimensions required a complex refactor for optimum code minimality, binary size reduction, and efficiency. Currently finishing and stress testing that refactor.

Paul

As shown in previous screenshot and below, multi-domain rendering (for next update) supports 8x concurrent domains. So if you have both G18 and G19 separate streampacks, you will automatically get one giant seamless fused timeseries dataset with all 8 domains, 4 from each satellite, with the independent domains updating during loops at their native 1, 5, or 10 minute frame cadences (MESO, CONUS, and Full Disk domains respectively):

Paul

Just need a few more days with this to be ultra-robust. These are the internals of the raster tile streaming engine, to support multiple pyramids and ensure the download ordering is fully optimized round-robin inter-pyramid top-down spatial psychovisual-forward-weighted temporal order. And ensuring the ingest scheduler is fully optimized, something I wisely deferred until the multipyramid domain context, knowing how much the code in this area would need to change and saving that work for now.

I also want the multidomain debut update release to have nice domain fading around the edges of far stretched areas, which I found does best with a distortion-based factor as opposed to screenspace edge convolution. This is important to retain aesthetic appearance of map where the full disk domains degrade rapidly on the global periphery where the projection angles become acute or whatever. Also the NW panhandle like corner of the weird geostationary projection
for G19 CONUS benefits from this fading.

I also want quality vector line rendered outline domains for MESO and CONUS (interior domains) so we can support effects for indicating which domain is which and even eventual new-frame update effects. That's purely visual so if that takes more than a day I'd get the base multidomain release out first. It will probably be out by Sunday night but early next week at latest I'd say.

The next update will truly justify V0's existence as a legitimately valuable supplemental tool for the state of the art in realtime full-resolution, progressive-downloading, low-latency raw display of satellite imagery. No other program is composing original projection raw full-resolution tiled timeseries GOES-R datasets in realtime on GPU. The nicest competitive solutions in my opinion, like windy.com's enhanced/optical flow modified satellite imagery, are all reprojected into their own mapping tiles and non-dataset-native resolutions. We are preserving the original projections and actually mathematically calculating them on the fly on GPU, which makes rendering moving domains free and automatic. Right now MESO 1/2 in GOES-R benefit from the native support for moving domains, which was from the begging of the R2 engine, not recent for multipyramid.

In the future when we get into NWP data, the HREF hurricane model simulations use storm-centered moving domains which vary on the forecast timeline, so that's where the moving domain support will really shine.

Paul

3 days ago I had a private internal multidomain build I will now share (not polished and not verified as tactical-grade for public version update), because I am now looking at multiple more days of delay. I need to excise and rewrite vast swaths of code which became unacceptably convoluted and suboptimal during the recent work to finalize multi-domain.

It's frustrating because this is blocking all sorts of otherwise parallel projects: 1) GIS/Vector tile line preprocessing and client downloading for CONUS-wide street-level mapping 2) same, but city/road label tiles 3) re-preprocessing the static elevation background data for the new R2-based path, so we can eliminate the original R1 engine and attain full unification with temporal raster data 4) progressing to global mapping raster polygon (land/sea mask) and orthoimagery background mapping preprocessing, which will be exceptionally hi-res and beautiful compared to first gen, which I previewed here in a different thread in September 2024.

So here is the temporary multidomain 3/10 private release, which does work decently, so current streampack holders can actually use it, as I take longer to perfect and polish the code to be industrial strength:

s3.us-east-005.backblazeb2.com/wsv3-tactical-public/D/Temp/WSV3.Tactical.03102026.zip

P.S. - Good news is, my perfectionism-absolutist coding standards for Tactical Mesoanalyst require such an intense rejection of the superficially working but not tactical-grade code for the multidomain rendering in the private 3/10 release above, that I am strongly considering moving directly to the postponed work for 8x multimap support in R2 engine, earlier than expected, as part of the scope of a broad rewrite. I will know by end of today if this makes sense to move up in priority.

JPWX

Paul I mentioned a while back about how I would like to see WSV3 able to be draggable from one screen to another (computer to TV). Is this in the works currently or planned to be implemented?

Paul

JPWX Multimonitor support should already be working, by simply launching the application graphically from the different monitor. It's now based on which monitor the mouse cursor is in at time of launch.

This gives full control over which monitor is your output. To change, close and relaunch on other monitor.

Let me know if this doesn't already work for you.

JPWX

It doesn't work for me on my end

Paul

[unknown] Can you be more specific? Did you launch the application with cursor on monitor 2? And it launched still on monitor 1?

JPWX

I have three monitors. I still opens up on my main monitor 1 even though I have the taskbar on at the bottom of each monitor. And still does even if I move the desktop icon to another monitor and open it from there

Paul

[unknown] OK, I haven't tested it myself and will do so soon. This should be very easy to get working properly.

Paul

The link to the 3/10 multidomain update is in my post from 3/13 above. I did not push that to the main public download link.

The raster engine refactor has expanded context, not just to 8x multimap support, but further, it's clear now I need to take 1-3 weeks working on a whole host of related things which break the wire format between client, server, and ingest. This includes user stats/auto update notifications/support for 16 scenes (the new raster engine will support dual-basis architecture where each map 0-7 has two scene slots, current and loading, so it can actively load new scenes onto GPU while still displaying the old) and also changes to the ingest system, where we want to pack neighbor tile edge texture data to speed up our seam-free gbuffer upscaling shaders.

This "big push" requires me to keep separate tactical/ingest applications on my local dev server, without ruining live data for users on current build.

Then, once I release this, it will break all data for existing versions and you will have to force-update. Because the wire format to server is changing. And part of this work - the automatic update notification system - will have support for warning the client about any future planned wire-format-breaking updates where an update will be needed or else data lost. This needs to be alerted to the user so that the 24/7/365 use case isn't disturbed. For now, I will have to manually email all paid licensees to apprise of the coming total loss of data, until you update.

The 8x multi-map generic raster engine rewrite, supporting live temporal and static background data in same engine, is still front-and-center of this massive refactor. I am doing this headless-first, for the industrial automation version, so the engine will support ability to export graphics.

The new raster engine will allow for preprocessing of our global mapping data, the same elevation tiles you've seen for a year now, but also, finally, 1) getting back the hi-res orthoimagery background mapping and 2) raster polygon tiles, for land/sea mask.

The mapping has been significantly held back from not having this infrastructure.

This does push the GIS/Vector mapping engine into April. This is a necessary development that will result in very rich raster mapping graphics and mature preprocessing ability before we go forward. I had been handicapped by the March/April 2025 system (Raster1) for too long. The new engine shares all the highly efficient Raster2 code and makes it standard across both static and temporal raster loading.

Thank you for your patience. The 3/10 build above shows you first-in-the-world 8x concurrent domain display of full res GPU rendered multiresolution raster pyramids, and I am excited to then have 8x distinct multimap with 8x domain rendering, which will be another industry first.

Kirk

How do I get my hands on this? Been a minute since I’ve been here — a lot to read. We have an event here this evening; I’d like to see it in the next-generation product, if possible.

Paul

Kirk See WSV3.com/Tactical and use the special 3/10 multidomain release above. Send me an email once you launch the EXE, with static-only background mapping. Live data streampack for full resolution G-19 imagery is available for $20/month/PC purchase but let me send you a 24 hour live data account if you want to try it.

JPWX

Hey Paul, quick question. Is that download link back on the 13th suppose to be a hot link?

Paul

JPWX No, it's just text.

« Previous Page Next Page »