| 05 Oct 2026
Evaluating the scanning capabilities of an MWR to estimate temperature and humidity advection from a single site
Abstract. Advection plays a fundamental role in the transport of heat and moisture within the Atmospheric Boundary Layer (ABL). Recently, horizontal advection has been quantified using ground-based remote sensing networks deployed at multiple locations. This study presents a novel single-site methodology to estimate horizontal temperature and humidity advection. Horizontal gradients are derived from a microwave radiometer (MWR) by exploiting its azimuthal scanning capability, while wind components are obtained from a Doppler wind lidar (DWL). Advection is estimated and a comprehensive uncertainty quantification framework is developed, based on standard error propagation and a Monte Carlo approach that accounts for correlated uncertainties in the MWR's observations. The resulting uncertainties are evaluated relative to the magnitude of advection over a range of horizontal gradients and wind speeds.
For typical ABL conditions, the uncertainty associated with humidity advection is approximately 60 % of its estimated value, while temperature advection uncertainties reach 140 %. However, under conditions characterized by strong horizontal gradients and high wind speeds, the observed magnitude of advection exceeds its uncertainties, enabling the detection of significant events such as frontal passages. As proof of concept, two strong advection episodes are analyzed. On 19 June 2022, a pronounced cold advection event was observed, and the current estimation captured intense cooling throughout the ABL, with vertically averaged values reaching −10 K h-1, with relative uncertainties generally below 40 %. A second case, on 8 October 2021, highlights humidity advection, with dominant zonal contribution at around 2300 UTC. The height-averaged estimated humidity advection reached -10 g kg-1 h-1, with uncertainties typically below 30 %. The present study provides a novel advection estimate and its robust assessment of capabilities and limitations, highlighting the clear identification of strong advection events in the ABL, reducing the logistic complexity of capturing this variable.
This paper presents a clever and elegant method for retrieving advection magnitudes from a single point without relying on spatially distributed observation nodes like existing methods. Unfortunately, the method appears to be of limited utility given the large error bars produced by the error analysis (and I fear there are some unaddressed sources of error that also need to be integrated). The method is of greatest use when the spatial gradients are large, but paradoxically that is when advection observations are needed the least as the advective signal is drowning out any local effects in the total derivative. It is when gradients are weak and advection is on the same order as diabatic heating or evapotranspiration that quantifying the advection is of the greatest importance. Still, none of this should preclude this well-written paper from being published, if for no other reason than to provide a springboard for future investigations by this or other research groups.
With all this being said, there are still several important issues that need to be addressed before the paper can be accepted for publication in AMT. This includes a discussion of some other sources of error and a further justification of the method itself.
1. This method fundamentally relies on the ability of a scanning MWR to detect and quantify horizontal gradients in thermodynamic scalars. Once you have a gradient, then it’s relatively easy to merge it with a vertical wind profile and get the advective tendency. Therefore, it is critical to demonstrate that the MWR is capable of accurately measuring horizontal scalar gradients at different heights using the method presented.
This is where I’m getting a little lost. I’m less familiar with horizontal scanning as I am zenith stares, so forgive me if I’m missing something fundamental here. But vertically-pointing passive remote sensing profilers struggle greatly with detecting unambiguously sharp vertical gradients in temperature that occur over a few dozen meters. How is it that one can rely on the MWR’s observation of weaker, more diffuse horizontal gradients? When scanning at a non-zenith angle, the weighting functions are going to peak at lower heights due to increased opacity along the observation path and therefore information content is going to be biased toward the surface. Figure 1a implies that the MWR is capable of equally detecting the target scalar at Z1 as it is at Z3. But that isn’t true: Z3 is going to be much more challenging to get useful information out of than Z1, and by the time a slant path gets more than a few hundred meters above the ground almost all the temperature and humidity information will be coming from the prior. Does the horizontal difference in temperatures in such conditions accurately measure the gradient, or is it just an assessment of random noise?
At the very least, the paper needs to demonstrate that useful information is coming from the higher altitudes such that differencing the observations at those heights provides actionable data. I think a figure that picks a subset of channels and shows how those weighting functions change from zenith to 30 degrees elevation could provide some very useful context as to where the information is coming from in the MWR observations.
2. Because of all of the issues raised in Point 1, I believe that the uncertainty analysis is missing some key points. Table 1 notes the values used for the uncertainty analysis, but the theta and q uncertainties are strongly dependent on height. Furthermore, the horizontal separation has a large influence on the uncertainty of the calculated advection: instrument noise dominates the uncertainty at low levels because even a tenth of a degree error in the observed theta results in a large d(theta)/dx when dx is so small. At high elevations the uncertainty of the scalars themselves is quite large. As a result, I’d expect there to be a sweet spot in the uncertainty in the middle of the advection profile where the spatial gradients are occurring over a larger distance but the scalars themselves are known with a decent degree of confidence. It appears that the uncertainty analysis shown in Fig 3 takes some mid-level assumptions about the performance of the advection calculations and then applies them geometrically to all heights, but I feel that this underestimates the actual variability in the calculated values due to the neglect of parameters that have strong height dependence.
3. A stronger paper would demonstrate that the method works by comparing it to some known values. I think Wagner et al. 2023 (cited in the original paper) shows a useful way forward. In that paper, the authors first demonstrated their method in the 3 km HRRR model in the United States. In a model the advection can be calculated directly, so by applying their technique to the model output and comparing it to the known advection (which matched very closely) they were able to demonstrate with a high degree of confidence the utility of their method. Given the short horizontal scale of this method, a 3 km convection allowing model won’t be appropriate (according to my basic trigonometry calculations, to get 1 km above the surface with a 30 degree elevation angle you need to go out 1.73 km; therefore you’re comparing adjacent grid cells at best). However, a large eddy simulation model would certainly provide the needed resolution. ARM provides dozens of pre-computed cases from their LASSO project at 100 m resolution, ready to download and investigate. I think it’s okay if the synthetic demonstration of this technique is at a different location than its real-world implementation.
Minor issues:
This is more of a philosophical question than a scientific one, perhaps, but how representative is this method, really? The highest heights of this technique are evaluating a gradient over no more than a couple of kilometers. The minimum valid height for this technique is going to depend on the configuration of the Doppler lidar, but is probably around 80 m AGL which means that the gradient is being assessed over just a couple hundred meters. Are gradients evaluated over such small distances truly representative of synoptic scale forcing? One way to investigate this would be to do the model investigation described above, but move the observing site within the domain to many different locations and illustrate how much the observed advection values change.
Figures 5 and on: the vertical coordinate for this appears to be height MSL, but that’s not explicitly stated. I feel that height AGL is a better coordinate as the end user is going to be more interested in what’s happening a kilometer above them than what’s happening at a specific height relative to sea level. This would also help better discern the influence of the doppler lidar dead band at the surface.