On the effects of cloud water content on passive microwave snowfall retrievals

Highlights

• Passive microwave retrievals of snowfall are prone to miss and false alarm.

• We defined the imposter brightness temperatures that have physical inconsistency.

• Silhouette Coefficient is used to find the imposters in geophysical classes.

• Conditioning cloud water content helps the retrievals to reduce the uncertainties.

• Neyman–Pearson hypothesis testing is used to find those physical conditions.

Abstract

The Bayesian passive microwave retrievals of snowfall often rely on mathematical matching of the observed vectors of brightness temperature with an a priori database of precipitation profiles and their corresponding brightness temperatures. Mathematical proximity does not necessarily lead to consistent retrievals due to limited information content of passive microwave observations. This paper defines imposter (genuine) vectors of brightness temperature as those that are mathematically close but physically inconsistent (consistent) and characterizes them through the Silhouette Coefficient (SC) analysis. The Neyman–Pearson (NP) hypothesis testing is used to separate the imposter and genuine brightness temperatures based on their associated values of cloud ice (IWP) and liquid water path (LWP), given by coincidences of CloudSat Profiling Radar (CPR) and the Global Precipitation Measurement (GPM) Microwave Imager (GMI). The study determines thresholds for IWP and LWP that allow optimal identification of imposter brightness temperatures of non-snowing and snowing clouds, which can mislead the passive microwave retrieval algorithms to falsely detect or miss the snowfall events. It is demonstrated that emission signal of supercooled liquid water in snowing clouds can lead to improved passive microwave retrieval of snowfall and conditioning the retrievals to the cloud IWP and LWP can result in marginal correction of the snowfall detection probability; however, reduce the probability of false alarm by 6%–8% over sea ice and open oceans.

Introduction

Falling snow occurs in response to a complex cascade of macro and microphysical processes in response to variations of moisture and temperature profiles in atmospheric column (Birkeland and Mock, 1996, Homan and Kane, 2014). Once on the ground, terrestrial snow changes diurnal cycle of surface temperature (Cohen and Rind, 1991), controls dynamics of surface mass and energy budget (Gallée et al., 2001, Sade et al., 2014), regulates the extent of permafrost (Stieglitz et al., 2003), and influences the nutrient seasonal cycle across soil–atmosphere–vegetation continuum (Niu and Yang, 2004, Grünberg et al., 2020).

Over oceans, snow on sea ice has a cooling effect on the Arctic climate system by increasing the surface albedo (Ingram et al., 1989, Ledley, 1991). At the same time, snow cover acts as an insulating layer and suppresses the latent and sensible heat fluxes to the atmosphere (Holtsmark, 1955), regulating the seasonal life cycle of sea ice (Gradinger et al., 1991). Within the past few decades, observational evidence suggest that the snowfall has been changing due to global warming (Kapnick and Delworth, 2013, Tamang et al., 2020, Zhang et al., 2021), especially over polar climate regimes (Danco et al., 2016). Changes of polar snowfall can amplify the loss of sea ice (Liu et al., 2012, Screen and Simmonds, 2012) and alter mass balance of ice sheets and glaciers (Eicken et al., 1994). Global monitoring of space–time dynamics of high-latitude snowfall is critical to expand our understanding of these changes (Kulie et al., 2016, Milani et al., 2021, Guilloteau et al., 2021).

Variability of snowfall is one of the least understood components of hydrologic water cycle over the Arctic and Southern Oceans – chiefly because of limited historical ground-based measurements (Milani et al., 2018, Dong, 2018) and errors in those data due to blowing snow, vegetation transition, and complex topography (Wen et al., 2016, Gossart et al., 2017). Microwave remote sensing using spaceborne satellites can provide a global picture of space–time variability of high-latitude snowfall (Foster et al., 2005, Gonzalez and Kummerow, 2020, Kulie et al., 2020) to overcome these challenges. Among existing satellites, the Global Precipitation Measurement (GPM, 2014-present) and CloudSat (2006–2020) satellites provide frequent passive and active observations of precipitation.

Onboard the GPM core satellite (Hou et al., 2014, Skofronick-Jackson et al., 2017), the GPM Microwave Imager (GMI, 10.65 to 183 ± 7 GHz), with a coverage up to 68°S-N, scans an outer swath width of 931 km with a mean footprint resolution ranging from 25 km at 10.65 GHz to 6 km at 183.3 ± 7 GHz channels (Draper et al., 2015). The GPM Dual-Frequency Precipitation Radar (DPR) measures precipitation reflectivity at Ka (13.6 GHz) and Ku (35.5 GHz) bands with a vertical resolution of 250 m over a footprint size 5 km (Iguchi et al., 2012, Toyoshima et al., 2015) within a swath width of 125 and 245 km, respectively. The coincident active-passive observations by GMI and DPR have been instrumental in developing algorithms for passive microwave retrievals of precipitation (Kummerow et al., 2015, Ebtehaj et al., 2016, Le et al., 2017, Turk et al., 2018). At the same time, the non-scanning W-band (94 GHz) CloudSat Profiling Radar (CPR) (Stephens et al., 2002) is a near-nadir-pointing radar with 16-day orbital cycle. The measured radar reflectivity profiles by CPR provide information of clouds and light precipitation profiles throughout the atmosphere column with vertical resolution of 240 m over a 1.8 × 1.4 km (along-track$\times$cross-track) footprint (Tanelli et al., 2008) with a quasi-global coverage 82°S-N.

Studies suggest that DPR lacks sensitivity to light snowfall events (Casella et al., 2017) with reflectivity values below 12 dBZ over oceans (Hamada and Takayabu, 2016, You et al., 2021). Even though CPR has a limited temporal frequency, it is highly sensitive to cloud ice particles and snowflakes with reflectivity values as low as −30 dBZ (Liu, 2008). Therefore, coincidences of CloudSat and GPM satellites not only allow to expand algorithmic capabilities in passive microwave retrievals of snowfall (Rysman et al., 2018, Takbiri et al., 2019, Ebtehaj et al., 2020, Vahedizade et al., 2021, Turk et al., 2021) but also help to explain when, where and why GMI passive microwave retrievals of snowfall might be highly uncertain through new programs such as the Atmosphere Observing System mission (AOS, Braun et al., 2022).

The study by Bennartz and Bauer (2003) revealed that channels around 150–170 GHz, which are only moderately affected by variations in surface emissivity, generally exhibit the strongest scattering signature due to precipitation-sized ice particles. These signatures have been used for passive microwave snowfall retrievals (Noh et al., 2006, Liu and Seo, 2013, Skofronick-Jackson et al., 2013, Panegrossi et al., 2017). It is worth noting that cloud ice particles and snowflakes both scatter the upwelling emission at high-frequency channels, above the oxygen absorption line 60 GHz, as the wavelengths approach the ice particles size (Bennartz and Petty, 2001, Noh et al., 2006, Skofronick-Jackson and Johnson, 2011, You et al., 2017).

As the surface emission increases, the strength of the scattering signatures also increases, in terms of a more significant depression in the observed brightness temperatures compared to the surrounding background emission. Thus, over radiometrically cold open oceans, any potential snowfall scattering signatures are naturally weaker than over land. However, even a thin 10 cm layer of young ice can increase the ocean surface emissivity from 0.4 to 0.95 at frequency channels above 10 GHz (Ulaby et al., 2014). As soon as snow begins to accumulate on ice, the surface emission begins to decrease over frequencies from 18 to 200 GHz (Stiles and Ulaby, 1980, Tsang et al., 2008) as the snow particles scatter the upwelling sea ice emission. Nevertheless, observations show that a snow-covered sea ice remains radiometrically warmer than open ocean surfaces (Vahedizade et al., 2021). Therefore, while the snowfall signatures are generally stronger over sea ice, they can be confused with snow-cover scattering signatures.

High amount of supercooled liquid water path (LWP) is one of the main reasons that lead to ice crystal growth, especially in mixed phase clouds. In particular, high concentration of supercooled liquid water provides an ideal environment for deposition of water vapor on ice crystals and thus their rapid growth, when the cloud environment is subsaturated for liquid water and supersaturated for ice, through the so-called Wegener–Bergeron–Findeisen process (Storelvmo and Tan, 2015). Moreover, when LWP is high enough, the likelihood of the collisions of liquid water droplets with ice particles increases substantially, giving rise to a faster growth of ice particles with surface temperature approaching to 0 °C – a process known as the wet growth regime (Pruppacher and Klett, 2010). For instance, over the Korean Peninsula in winter 2017–18, observations show that LWP in three types of snowing clouds (i.e., near-surface, shallow, and deep clouds) can reach to 500 g m⁻² with a high probability (Wang et al., 2013, Jeoung et al., 2020). The challenge is that emission of cloud liquid water can extend to higher frequencies and weakens the scattering signatures of ice aloft and snowfall particles (Kneifel et al., 2010, Liu and Seo, 2013, Ebtehaj and Kummerow, 2017). This masking effect can be naturally more significant over oceans than land surfaces because oceans are radiometrically colder and supply more moisture to the overlying atmosphere. It is important to note that the water vapor absorption line peaks at 183.3 GHz and spreads to nearby frequency channels. Therefore, snowfall scattering at nearby frequencies can be weakened due to a significant increase in total water vapor content of the atmosphere as well (Panegrossi et al., 2017, Millán et al., 2020).

Therefore, the content of the observed microwave brightness temperatures within a field of view (FOV) captures signals of surface and different atmospheric constituents (e.g., water vapor, cloud water/ice content, precipitation water/ice, supercooled water) through emission, absorption, and scattering processing resulting in an ill-posed inverse problem with non-unique solutions. In other words, different combinations of surface and atmospheric conditions can lead to exactly the same or very similar observed brightness temperatures. Thus, in the precipitation passive microwave retrieval algorithms (Petty and Li, 2013a, Petty and Li, 2013b, Utsumi et al., 2020, Milani et al., 2021, Panegrossi et al., 2022, Sanò et al., 2022), additional a priori information is necessary to constrain the solution space and tackle the ill-posed nature of the inverse problem.

The Bayesian retrieval algorithms, which are widely used inverse models to cope with ill-posed nature of passive microwave precipitation retrieval problems (Kummerow et al., 2001, Kummerow et al., 2011), often match the observed brightness temperatures (TB) with an a priori database that links a statistically representative number of TBs to their precipitation profiles. The matching process is often based on a mathematical proximity measure to detect the occurrence and phase of precipitation and to estimate its rate in a probabilistic sense (Petty, 1997, Petty, 2013, Kummerow et al., 2015, Ebtehaj et al., 2015, Ebtehaj et al., 2020). As explained, background surface and atmospheric emission signals often confuse the matching process and increase the retrieval uncertainties. For example, over snow-covered sea ice, in a non-snowing cloud with relatively low values of ice water path (IWP) and LWP, algorithms might confuse surface and ice aloft scattering and falsely detect snowfall. In a snowing cloud, relatively high values of LWP can alleviate ice aloft and snowfall scattering and lead to a high probability of miss in those algorithms that solely rely on snowfall scattering signals.

This paper mainly investigates the following questions using a GMI-CPR coincidence dataset over open oceans and sea ice: (1) Under which range of LWP and IWP, are the Bayesian snowfall retrieval algorithms might be prone to miss (blind region) or false (bright region) detection of snowfall events? (2) How can these blind and bright spectral regions be detected and corrected in the passive microwave retrievals? (3) Where and to what extent can the proposed approach correct the retrievals over the blind and bright spectral regions on a global scale? The paper is organized as follows. Sections 2 Data, 3 Methodology describe the data and methodology that rely on the Silhouette Coefficient (SC) analysis and the Neyman–Pearson (NP) hypothesis testing. Optimal values of lower and upper thresholds of LWP and IWP are identified to separate snowing clouds from non-snowing atmosphere. The results, findings and discussions are presented in Section 4. In this section, these thresholds are used to inform a snowfall retrieval algorithm (Vahedizade et al., 2021) about the underlying physical conditions of atmosphere to improve its detection capability. Section 5 provides a summary and concludes the study.