@idefix37 I'm unable to reproduce the 3 fold discrepancy I was seeing last week. But here are some screenshots showing a 2x descrepancy for the same location, which is still pretty significant, especially for a forecast for the current time. This is a pretty big difference in visibility when hiking in the mountains.

I'm not exactly sure what it would mean for the grib file to be sent at lower resolution. It was previously mentioned as an explanation for why the interpolated value would be different at different zoom levels. Something seems to be more course at higher zoom levels.

The forecast changed slightly since I last posted but another easy screenshot I can do to show something is off is the hourly forecast for Tuesday
If adding up the hourly precip between 8 am and 6 pm there is well under 1 inch of precip over this interval, only about 1/4 inch.

@Suty thank you for reading this. Its hard to capture this in screenshots just due to the detailed nature of rainfall integration. But here are three screenshots that suggest the narrative.

The first screenshot is the rain/thunder layer at 6pm Tuesday showing the storm system just about to strike the coast of northern California. This layer suggests the start of substantial rainfall is after 6 pm Tuesday with little rain before that.

The next screenshot zoomed in closer on Eureka, CA as an example shows the 4 day accumulation starting at 6pm Tuesday. The 4 day accumulation over this interval is displayed as about 6.3 inches.

The third screenshot with the picker over the same location shows the 4 day accumulation but starting 10 hours earlier at 8 am. The 4 day accumulation over this earlier interval is displayed as about 7.6 inches. Its strange there is over an inch of additional precipitation when the rain/thunder layer shows little activity over these 10 hours. If anything this time interval should show less rain as its missing the end of the storm that lingers into Saturday.

Ive noticed for some time, even after recent changes, that the accumulation tool does not work properly. Across models, the accumulation calculation is not correct based on the underlying data.

It seems the tool is simply integrating the rain rate over a different time period than the slider is set. For example, for a rain event that lasts 4 days, the accumulation should be maximum when the 4 day slider is moved to capture the whole event. But I find the accumulation calculation is higher when the slider is advanced to capture time before it starts raining but misses the tail end of the event.

@idefix37 I'm unable to reproduce the 3 fold discrepancy I was seeing last week. But here are some screenshots showing a 2x descrepancy for the same location, which is still pretty significant, especially for a forecast for the current time. This is a pretty big difference in visibility when hiking in the mountains.

I'm not exactly sure what it would mean for the grib file to be sent at lower resolution. It was previously mentioned as an explanation for why the interpolated value would be different at different zoom levels. Something seems to be more course at higher zoom levels.

Ive noticed this zoom dependence as well. At btschumy mentions its troubling because it unclear which value to trust, especially when there is a large discrepancy. I've seen over a 3x variation in forecasted PM 2.5 where it means the difference between Moderate and straight up Unhealthy air quality. Can this be interpreted as meaning the uncertainty in the forecast is even greater than this 3x discrepancy? This seems reasonable for a forecast days out but I've seen this discrepancy for the current time.

btschumy has a good point on grib file resolution possibly being sent at a lower resolution when zoomed further out. Can you confirm this is the case? Otherwise, I'm struggling to understand why the interpolated value would vary with zoom level. The only other explanation I can think of is if the coordinates of the chosen point are rounded more as you zoom out. For example, at a close zoom, say the interpolation is calculated by rounding the coordinate to the nearest thousandth of a degree. But zoomed further out the interpolation is calculated by rounding to the nearest tenth of a degree. In this case the computed forecast location could vary by a few miles which could make a huge difference.