Remotely Sensing Temperature and Humidity Profiles in the PBL

Remotely Sensing Temperature and Humidity Profiles in the PBL

Dave TurnerNOAA / Global Systems Laboratory

1

Thermodynamic Profiling: Primary Challenge is in the Lower Troposphere

2

• Huge observational gaps exist in lower trop thermodynamic profiling• Closing these gaps is essential for progress in weather and climate research• Ground-based passive and active remote sensing systems can close these gaps• Marriage of these ground-based systems (and future networks of them) and satellite sensors

enhance information content and utility

Published online Aug 2015

Microwave Radiometer (MWR)

3

• Measures downwelling microwave radiance in dozens of spectral channels• Over three decades of development• Hardened, automated instruments • Commercially available from multiple vendors• Need to calibrate with liquid nitrogen periodically

Solheim et al. Radio Sci. 1998Rose et al. Atmos. Res. 2005

Caveat: Radio frequency interference becoming more relevant!

Infrared Radiance Interferometer

4

25 µm 7.1 µm10 µm15 µm

Rad

ianc

e [m

W/ (

m2

srcm

-1)]

Wavelength [µm]

SGP, 8 Jul 2003PWV: 48.0 mmNSA, 1 Mar 2004

PWV: 1.3 mm

Knuteson et al. JTECH 2004 (two parts)Turner et al. AMS Monograph 2016

• Measures downwelling IR spectra emitted by atmosphere at high temporal and spectral resolution (30 s, 1 cm-1)

• Invented by Univ Wisconsin – Madison in 1990s; matured as part of the DOE ARM Program

• Hardened, automated instrument; self-calibrating• Commercially available from two vendors

AERI

ASSIST

NCAR Water Vapor Differential Absorption Lidar (nDIAL)

5

Spuler et al. AMT 2015Weckwerth et al. JTECH 2016

• Laser-based active remote sensor • Developed at NCAR and Montana State University

– Based upon prototype developed at MSU• Micropulse system using diode-based lasers• Automated instrument; self-calibrating (narrowband approach)• Deployed during FRAPPE, PECAN, Perdigao, CACTI, …• Lowest good data level: ~500 m AGL

Still research based system(NCAR now has 5 of these in their instrument pool)

• Laser-based active remote sensor • Developed by Vaisala; aim is to be commercially available by 2021• Uses two separate systems; one optimized for near surface and one for “far range”• Uses broadband approach; a bit more challenging to calibrate

Vaisala Water Vapor DIAL (vDIAL)

6Newsom et al. JTECH 2020

vDIALRLIDAERI

May-June 2017 at ARM SGP Site

• Demonstration deployments at ARM SGP in 2017, DWD Lindenberg, ECCC

• Lowest good data level: ~20 m AGL

Radio Acoustic Sounding System (RASS)

7

Martner et al. BAMS 1993Bianco et al. AMT 2017

• RASS can be paired with radar wind profiler or sodar• Radar/sodar tracks speed of sound wave, which is proportional to Tv• Higher frequency systems needed to profile near the surface; lower

frequency systems profile to higher altitudes• Most groups typically only collect RASS Tv profiles once / hour (using

5 or 10 min integration)

New Commercially Available Raman Lidar

(Measures q and T)

8This shows 5-min resolution, but 30-s resolution also possible

Similar WV performance as ARM Raman lidar at SGP

This is the best temperature profiling system I have seen

Lange et al. GRL 2019

Passive Remote Sensing

• IR/MW radiometers measure radiation emitted from the atmosphere in channels sensitive to emission from different gases– Radiance contains info on T(z) and q(z) (and clouds, other trace gases, etc)– The channel selection should span a range of optical depth

• Ill-defined problem; retrievals need to be constrained by either a prioridata (e.g., climatology) or model background

• Information content is key: what part of retrieved profile is from observation vs. from the a priori information

• Calibration is absolutely key (both in obs and forward model) • No real information on how temperature and moisture covaries temporally

/ spatially / vertically, which hinders retrievals9

Synergistic Remote Sensing

• Combining active and passive observations into a retrieval can improve accuracy and information content of retrieved profiles

• Consistent forward models and no systematic errors critical• Strength of one observing technology can be used to overcome the

weakness of the other• Uncertainty analysis and information content is important• Retrievals performed using TROPoe algorithm

– Physical-iterative method using optimal estimation framework– Able to combine different types of observations to retrieve T(z) and q(z) – Full error characterization and vertical resolution are standard output products

10

(Turner and Löhnert 2014; Turner and Blumberg 2019; Turner and Löhnert 2021)

Physical-Iterative Retrieval(Optimal Estimation)

11

𝑌𝑆!

Observation

Obs Uncertainty

𝐹𝑋"

Forward Model

State Vector (nth iter)

𝑋#𝑆#

Mean Prior

Prior Uncertainty𝐾" =

𝜕𝐹𝜕𝑋"

Jacobian of F

𝑋"$% = 𝑋# + 𝑆#&% +𝐾"'𝑆!&%𝐾" &%𝐾"'𝑆!&% 𝑌 − 𝐹 𝑋" +𝐾" 𝑋" −𝑋#

Iterative solution (n to n+1)

𝑆( = 𝑆#&% +𝐾'𝑆!&%𝐾 &% Uncertainty in X

A= 𝑆#&% +𝐾'𝑆!&%𝐾 &%𝐾'𝑆!&%𝐾 Averaging Kernal

Combining Observations within the Retrieval

• Combine the observations into the obs vector Y

• Use the appropriate forward model for each element of Y

• Combine the observational uncertainties

12

𝑌 = [𝑌)"*+%, 𝑌)"*+,, 𝑌)"*+-]

𝐹 𝑋 = [𝐹)"*+% 𝑋 , 𝐹)"*+, 𝑋 , 𝐹)"*+- 𝑋 ]

𝐹 𝑋 = [/𝑌)"*+%, /𝑌)"*+,, /𝑌)"*+-]

𝑆! =𝑆)"*+%

𝑆)"*+,𝑆)"*+-

0 0

00

0 0


13

Example #1: XPIA on 17 Mar 2015 at 1830 z

Combined MWR+RASS retrieval described in Djalalova et al. AMT 2021 (in review)


14



15

Combining Observations within the RetrievalExample #1: XPIA on 17 Mar 2015 at 1830 z



16




18

Example #2: PECAN on 20 Jun 2015 at 0248 z

Combined IR+DIAL retrieval described in Turner and Löhnert AMT 2021


19




20



So How Good are these Retrievals?

21xkcd.com/1478

𝑆( = 𝑆#&% +𝐾'𝑆!&%𝐾 &% Uncertainty in X

A= 𝑆#&% +𝐾'𝑆!&%𝐾 &%𝐾'𝑆!&%𝐾 Averaging Kernal

• What is the inherent uncertainty?• How many independent pieces of info are in

the retrieved T and q profiles?• How is this information distributed vertically?• What is the true vertical resolution of these

profiles?• How should I use these data?

Example: Passive-Only Retrieval

22From Perdigao Field Campaign in Portugal, where an AERI, MWR, nDIAL and sondes were collocatedCombined IR+DIAL retrieval described in Turner and Löhnert AMT 2021

Example: Active + Passive Retrieval

23From Perdigao Field Campaign in Portugal, where an AERI, MWR, nDIAL and sondes were collocatedCombined IR+DIAL retrieval described in Turner and Löhnert AMT 2021

Bias and RMSE using Radiosondes as Truth

24Blumberg et al. JAMC 2015

AERIMWRe

AERIMWRe

1-Sigma Uncertainty Profiles

25

(Square root of the diagonal of Sx)

Turner and Löhnert AMT 2021


26Turner and Löhnert AMT 2021

Information Content and Vertical Resolution

• Retrievals are performed on a discrete vertical grid– Often has 40 to 100 vertical layers

• The amount of information in the observations is well less than the number of layers– E.g., the number of channels on a MWR is always less than 25

• Ill-posed problem• To constrain the retrieval, a priori data (e.g., radiosonde climatology)

is used to provide guidance on the level-to-level correlation• So what is the true IC and Vres of the retrieved profiles?– Can be determined directly from the averaging kernel!

29

Cumulative Degrees of Freedom for Signal

30

(Diagonal of the Averaging Kernel)

Turner and Löhnert AMT 2021

Cumulative Degrees of Freedom for Signal

31

• DFS provides a measure of the number of independent pieces of information

• In this example, the T(z) retrieved from the– MWR has DFS = 2.3 below 3 km– AERI has DFS = 5.4 below 3 km

• But below 1 km, the– MWR has DFS = 1.7– AERI has DFS = 4.4

• If you wanted to assimilate these profiles, you would want to use levels that are separated by at least 1 DFS

• In this example, if we take the first level at 100 m AGL then we would assimilate– 2 levels from the MWR– 4 levels from the AERI

Interpreting the cumulative DFS profile

Cloud Impacts on the Retrieval: MW vs IR

32

• Clouds are much more opaque in the IR than MW– Cloud is opaque in IR when LWP > 60 g m-2

– MWR remains very transparent until LWP is very high (>1000 g m-2)

• IR has no ability to profile above cloud base in these situations, but MW can

• In this example (assume LWP > 60 g m-2)– DFS below cloud base:

• AERI 4.3, MWR 1.7– DFS above cloud base:

• AERI 0.0, MWR 0.6• True information content above/below

cloud depends on CBH and LWP• The points one would assimilate are

determined the same way as before

Comparing Similar Instruments• 3 ASSISTs operated several hours in clear sky

conditions at Quebec City on 12 July 2021• How well do they compare?

37

Examples of How these Retrievals are Being Used

38

This is just a subset.There are many I could have chosen to highlight

Composite Radar Image: 0425 UTC on 10 Aug 2014

Thermodynamic Profiles and LWP from the AERI

46

Toms et al. MWR 2017Haghi et al. BAMS 2019

Cross-comparisons with UAS Observations

48

Koch et al. JTECH 2018

de Boer et al. BAMS 2018

DA Impact on a Tornado Forecast

49Storm of interest

MRMS Composite Reflectivity (dBZ)

2330 UTC

Storm-scale Domain (3 km)

Hu et al. WAF 2019

• On 13 July 2015, a tornado formed near Nickerson, KS• Near surface air was very warm (>35 C) and dry• Environmental storm-relative helicity was very small• Neither a tornado or severe thunderstorm watch was issued• After the supercell formed, it moved towards the SW

• Event occurred during PECAN, when there were 6 fixed sites with ground-based AERIs and wind profilers

• Used the NSSL Warn-on-Forecast ensemble system (NEWS-E) using GSI-EnKF

Nickerson Storm (After 2-h DA)Forecast initialized at 2000 z

50

MRMS CREF at 2330 UTC

Probability of CREF > 25 dBZ (3 km neighborhood)

CNTL

%

CIAERI_DL

1-h Fcst valid 2100 UTC

1.5-h Fcst valid 2130 UTC



3.5-h Fcst valid 2330 UTC

Black is 25 dBZ contour from MRMS composite reflectivity

Hu et al. WAF 2019

Many important quantities depend on spatial gradients of the kinematic and thermodynamic fields:

• Divergence• Advection

We can calculate this at SGP using Green’s Theorem:

The right-hand side of this equation looks a lot like the definitions of these quantities.

If we can use observations to approximate the left-hand side, we can get advection and similar properties.

Quantifying WV and Temperature Advection

51

!𝑃 𝑑𝑥 + 𝑄 𝑑𝑦 =)𝜕𝑄𝜕𝑥 −

𝜕𝑃𝜕𝑦 𝑑𝐴

ARM SGP Site

Wagner et al. JTECH 2021 (in review)

Quantifying Land-Atmosphere Interactions• Mixing diagram approach to study

energy and moisture budget evolution in the CBL pioneered by Betts (1992)

• Santanello et al. (2009) showed utility of approach to evaluate models

52

Wakefield et al. J Hydro Met 2021 (in preparation)

Summary• Ground-based thermodynamic (TD) profiles in the PBL very useful for a range

of operational and research objectives• Only passive TD profilers are currently commercially available

– Active lidar-based systems coming in next few years• Understanding the information content, vertical distribution of this

information, and resolution is critical to use the obs correctly– Variational methods like the TROPoe retrieval provide this info– Combining observations from different instruments can increase the information

content and improve the retrievals– The retrieval method matters! Working to make TROPoe available to all via Docker

• Utility of TD profilers demonstrated in several different applications• Continuing to use these data for both process-study research and in limited-

area data assimilation experiments53

Remotely Sensing Temperature and Humidity Profiles in the PBL

Documents

Transcript of Remotely Sensing Temperature and Humidity Profiles in the PBL