This post explores data-filling methods for missing data in a GOES-17 Level 1B radiance time series dataset.

Background: SSEC’s Data Center is home to a near complete record of GOES-16 and GOES-17 Full Disk Level 1B data. With these data capabilities, it is possible to compile a time series or ‘short climatology’ of GOES data over a given location. In examples shown here, Pago Pago has been chosen. (Pago Pago, the capital of American Samoa, holds significance due to its data sparseness. Weather forecasting and nowcasting in remote Pacific regions such as Pago Pago are heavily reliant on satellite data because of a lack of ground-based observations.)

Long-term purpose: The purpose of building a continuous satellite time series is to perform a time series analysis, i.e. to determine any periodicity or patterns in the data.

Short-term goal: The goal is to build a continuous, gap-free, time record for GOES-17 Band 13 radiance over Pago Pago from 2019-2021. Brightness temperature is computed from radiance. Compiling the existing data is trivial, but when data is missing, it can present problems. The data must be gap-free for a time series analysis. For the GOES-17 Level 1 record from January 1, 2019 – December 31, 2021, 1.14% of radiance data is missing. Some gaps are as large as 31 hours.

Approach: To build a continuous data record, data gaps must be identified and filled. The original raw data, which exists as frequently as every ten minutes, is first sampled at every fifteen minutes. Then, it is smoothed using a moving mean method. The moving mean computes a mean over a sliding window of length k. As a last step, the smoothed data are run through a shape-preserving piecewise cubic spline interpolation (nicknamed ‘pchip’ interpolation) to fill remaining gaps [1].

Results: In examples below, we investigate samples of the time series for k = 6 hours, 12 hours, 24 hours, and 48 hours over the two example gaps: January 2019 and July 2021. As the data is smoothed, precision is lost when comparing to the original time series. However, smoother data also results in a more “natural” looking interpolation. For example, when examining the 48-hr smoothed data [red lines in panels (d)], the data hardly looks interpolated.

How smoothed must the data be for the time series to look convincingly ‘real’ after interpolation? And will the data smoothing and loss of precision affect results from a time series analysis? These questions remain unanswered.

View only this post Read Less

GOES-18  images in this blog post are preliminary and non-operational

GOES-18 “Red” Visible (0.64 µm), Shortwave Infrared (3.9 µm) and “Clean” Infrared Window (10.35 µm) images (above) showed the formation of a small pyrocumulonimbus (pyroCb) cloud — generated by a wildfire that was burning near the northwest coast of Great Slave Lake in the Northwest Territories of Canada — on 24 June 2022. The pyroCB cloud then drifted southeastward across the lake, toward Resolute Bay (CYFR). Incidentally, this was Canada’s first documented pyroCb of the 2022 wildfire season.

Suomi-NPP VIIRS Visible (0.64 µm), Shortwave Infrared (3.74 µm) and Infrared Window (11.45 µm) images valid at 1930 UTC (above) showed the pyroCb shortly after it formed (when it exhibited a minimum cloud-top 11.45 µm infrared brightness temperature of -49C, lighter shades of red) — and Suomi-NPP VIIRS images valid at 1930 UTC (below) displayed the pyroCb over Great Slave Lake at 2110 UTC (when it exhibited a minimum cloud-top 11.45 µm infrared brightness temperature of -54C, darker shades of violet).

View only this post Read Less

The radar at NWS Grand Junction, Colorado (KGJT) was down for scheduled maintenance in late June 2022 — and on 23 June, overlapping 1-minute Mesoscale Domain Sectors provided 30-second GOES-17 (GOES-West) “Red” Visible (0.64 µm) images (above), which showed the development of thunderstorms across a portion of their County Warning Area (which includes eastern Utah and western Colorado) to help fill their gap in radar coverage. These particular storms did not reach severe levels, but some produced small hail, strong winds and heavy rainfall (Local Storm Reports).

View only this post Read Less

GOES-18 images in this blog post are preliminary and non-operational

GOES-18 True Color RGB images created using Geo2Grid (above) showed smoke from Alaskan wildfires that had been transported southward over Kodiak Island and the northern Gulf of Alaska. Most of this smoke remained aloft — note that although Kodiak began reporting smoke late in the day, the surface visibility remained at 10 miles (below).

A composite of VIIRS True Color imagery (below) showed the location of active fires over Alaska on that day.

View only this post Read Less