How was channel 3 of MHS on satellite NOAA18 performing overall? Can I use the data? – Scientists ask or should ask these questions when they want to use operational data for their research. As a general question one could ask: How does the performance of the microwave (MW) humidity sounding instruments MHS, AMSU-B and SSMT-2 evolve over their life time? Operational level 1c data, i.e. radiance in units of brightness temperature, of the MW instruments comes with sparse uncertainty information only, if they come with uncertainty information at all. But the knowledge about the uncertainty is crucial for any measurements. Within the FIDUCEO project we have tackled the issue about the uncertainty due to noise of the MW sounders and recently published our analysis on that (Ref. 1) to answer such kind of questions. Here, we will briefly present the main outcomes of our study: First, what do we call noise and how do we calculate it? Second, how does the noise evolution actually look like? Third, what data can be used? And last, does the story end here?
What do we call “noise”?
Let’s first clarify our use of the notion “noise”: In our analysis, we consider two quantities related to noise. First, we define the count noise as the fluctuations in instrumental counts. That means, we look at the fluctuations in the counts of the deep space view (DSV) measurements and the view of the internal warm calibration target (IWCT), both views are used for the instrumental calibration and therefore directly impact the quality of the final measurement of the Earth scene. Second, we consider the Noise Equivalent Differential temperature (NEDT) that can be deduced either from DSV (“cold NEDT”) or IWCT counts (“warm NEDT”).
How do we determine noise?
We computed the count noise for each channel, instrument and mission with the help of the Allan Deviation (see also the corresponding FIDUCEO report by J. Mittaz, Ref .2). The Allan Deviation is a well-known tool in other disciplines, but has only been recently suggested for use on Earth observation data (Ref. 3). The advantage of the Allan Deviation over the widely used standard deviation is that it will give a noise estimate unbiased from the natural underlying variations of the data. Those variations occur for measurements from polar orbiting satellites which these MW instruments fly on, but the variations are not due to a noise signal. Hence, using the Allan Deviation we get a better noise estimate on the raw measurement counts.
As a second quantity, we compute NEDT. This measure translates the count noise to a brightness temperature. Thus, it measures the sensitivity of the instrument to small changes of the signal and therefore closely relates to the gain of the instrument. In fact, NEDT is computed from the quotient of actual count noise and the gain. Hence, a large gain means small NEDT and vice versa. Now, a small NEDT means that the uncertainty due to noise on the actual measurement is small since the small changes in the signal due to noise correspond to a very small changes temperature only. A large NEDT means that the uncertainty due to noise is rather large: changes in the signal due to noise will be translated to huge changes in temperature.
How does the noise evolution look like?
These two quantities, the count noise and NEDT, being closely related by the gain, are the basis for our analysis of noise evolution over the life time of the MW sounders: Computing both quantities and additionally investigating all relevant calibration measurements over the life time, we get a comprehensive picture of the performance of the instruments. This answers to the second and third aspects of the question asking for all instruments and all times.
Figure 1 Left: DSV count noise for channel 3 of all instruments. Right: deduced cold NEDT.
Looking at the evolution of NEDT, tells us about the evolution of the sensitivity of the instrument. This is a very important measure and serves as main quantity to provide an overview of the usability of data. Figure 1 (right) displays the evolution of cold NEDT over the life time of all 11 considered instruments for channel 3, i.e. 183±1 GHz channel. Figure 1 (left) displays the same, but for DSV count noise. The behavior of both quantities is sometimes similar, sometimes different: On mission time scales the count noise is rather stable for most cases, whereas the NEDT shows distinct evolution. This is due to the gain that enters the NEDT-calculation. Wherever the gain shows variations, NEDT will change also, although the underlying count noise does not necessarily need to change (see the green lines for NOAA15 and NOAA16). Nonetheless, the count noise itself shows variations on shorter time scales of years or months (see blue lines of F14, F15 or orange line of Metop-A), that also impact NEDT.
What data can be used then?
The evolution of NEDT and count noise for all instruments and channels reveals a strong degradation of NOAA15 and NOAA16 AMSU-B instruments: NEDT increases strongly and surpasses 5 K in 2010. But it is not the actual noise level that increases (see Figure 1 dark green lines), but it is the gain that decreases dramatically and makes NEDT surpass 1 K in 2006. We took this threshold of 1 K to display in Figure 2 a set of usable data from which even a time series for all channels without gaps may be constructed. Of course, changing this threshold for higher/ lower measurement precision would alter the image of Figure 2. The users would have to decide on the acceptable level of uncertainty for their purposes and set their threshold accordingly.
Figure 2 Periods of NEDT < 1K
Does the story end here?
The usability of this noise analysis does not end here, no. The computing method for the count noise, making use of the Allan Deviation, is used in the FIDUCEO processing to produce new Fundamental Climate Data Records (FCDRs). The new FCDRs will provide information on uncertainty due to noise and all other sources identified within the project. Thus, the scientists in climate research get direct access to valuable uncertainty information together with the actual measurement data within one dataset.
The publication, which this blog entry is referring to, contains useful information for users already as they can quickly estimate whether the chosen instrument, channel and time period might be of use for their study or whether the instrument’s data should not be used for those times. Hence, we can answer the initial question: Looking at the evolution of NEDT, Channel 3 of MHS on NOAA18 shows some ups and downs. Nothing seriously bad happens, but from 2012 on, the amplitude gets larger and NEDT reaches 1K in 2016. So, one should be aware of the changing uncertainty due to noise when using the data.
In future, the information on uncertainty due to noise will be available in the FCDRs for every pixel recorded by the instruments. This will be highly valuable information as complement to the overview provided in the publication.
Ref. 1. Hans, I., Burgdorf, M., John, V. O., Mittaz, J., and Buehler, S. A.: Noise performance of microwave humidity sounders over their lifetime, Atmos. Meas. Tech., 10, 4927-4945, https://doi.org/10.5194/amt-10-4927-2017, 2017.
Ref. 2. Mittaz, J.: Instrument Noise characterization and the Allan/M-sample, available at: http://www.fiduceo.eu/sites/default/files/publications/noise_and_allan_v..., 2016
Ref. 3. Tian, M, Zou, X. & Weng, F., 2015, Use of Allan deviation for characterising satellite microwave sounder noise equivalent differential temperature (NEDT), IEEE Geoscience and Remote Sensing Letters, 12, 2477 –2480 doi: 10.1109/LGRS.2015.2485945