Astronaut Ellison Shoji Onizuka Memorial Downtown Los Angeles, CA Los Angeles City Hall
Passive Microwave Radiometric Observations over Land The performance of physically based precipitation retrievals from the GPM constellation of passive microwave (PMW) radiometers hinges on the capability to replicate the wide variety of observed multi-channel brightness temperatures (TB), under changing land surface conditions. How can one isolate and study the land surface effect? For radiometer, the surface boundary condition is the multi-channel emissivity, and for radar the backscatter cross-section σ 0. Procedure Construct a light version, mimicking the GPROF-GPM retrieval framework currently in place, using synthetic retrievals to trace back the effect of changes to surface properties upon precipitation retrieval performance. Use a land surface classification, land surface properties, and land surface emissivity principal components.
Use what others have discovered Use findings from the various NASA Precipitation Measuring Missions (PMM) working groups: To introduce hydrometeor size distribution (HSD) variability in a-prioir databases, follow the correlations noted amongst the HSD parameters (Williams et. al., 2014). For GPM-era radiometers, 10H channel is most sensitive to surface (Bytheway et. al., 2010). When building databases, use covariances with 10H to jointly adjust higher frequency emissivities for forward RT model. For physical surface, try at least two control parameters (related to soil moisture and vegetation water content) (Ringerud et. al., 2013; Turk et. al., 2014; others). To capture changing surface conditions, when possible jointly utilize the radar and radiometer observations to inform the retrieval.
Leading Emissivity Principal Components Have Similar Patterns As Soil Moisture and Vegetation Water Content Test 1: To vary the emissivity, vary the soil moisture (SM) and vegetation water content (WC) (Turk, Li, Haddad. et. al., 2013)
Leading Emissivity Principal Components Have Similar Patterns As Soil Moisture and Vegetation Water Content Test 2: To vary the emissivity, vary the first two emissivity principal components (Turk, Haddad, You et. al., 2013)
Radar-Based Land Surface Classification (Durden et. al., 2012) Classify surfaces by how surface backscatter varies as a function of radar incidence angle. Used for surface reflection technique (SRT) radar retrievals from TRMM-PR (now DPR) data (Meneghini et. al., 2004). 0.1-degree TRMM-PR temporal SRT σ 0 database used (8 land classes defined). Examine response to rain using two years of matched TMI-PR, with previous time accumulated 1-km NMQ precipitation over US-NEXRAD coverage area. Cluster 3 looks like bare soil (large change with incidence angle, high variability) Incidence angle (deg) Cluster 8 looks like forest (little change, low variability)
Durden et. al. 2012 Land Classes 2-9 JULY Class 6-7 separation generally coincides with the change from grass savanna to grass prairie Class 3 includes irrigated areas, and some urban areas
Durden et. al. 2012 Land Classes 2-9 JULY Class 6-7 separates the transition between cerrado and caatinga type vegetation southern grasslands
Joint PDFs Class 4 No-Rain Prev 24-hrs 0 < θ < 4 degrees 4 < θ < 8 degrees 8 < θ < 12 degrees 12 < θ < 16 degrees
Joint PDFs Class 4 > 25-mm Prev 24-hrs 0 < θ < 4 degrees 4 < θ < 8 degrees 8 < θ < 12 degrees 12 < θ < 16 degrees
Joint PDFs Class 8 No Rain Prev 24-hrs 0 < θ < 4 degrees 4 < θ < 8 degrees 8 < θ < 12 degrees 12 < θ < 16 degrees
Joint PDFs Class 8 > 25-mm Prev 24-hrs 0 < θ < 4 degrees 4 < θ < 8 degrees 8 < θ < 12 degrees 12 < θ < 16 degrees
All data where PR=0 (not raining at observation time) Effect on TMI 10H emissivity: CDF for each class rain persistence effect most noticeable in class 3 and 4 No rain previous 24- hrs All 2 3 No rain previous 72- hrs No rain in previous 24- hrs, but > 1 mm rain in the 48-hr period before that 4 5 6 > 10-mm previous 3-hrs > 25-mm previous 24- hrs 7 8 9
Surface Offline Lookup Table Calculations Hydrometeors physical emissivity LUT Hydrometeor Size Dist LUT soil moisture veg water content PC-based emissivity LUT surface temp (MERRA) Two emis PCs (nonlinear TB combinations) total column vapor (MERRA) Two surface control variables, in addition to T sfc and total column vapor (TQV) temp density mass cont Per-particle scattering LUT s exist at the GPM frequencies for the Kuo and Lu databases of non-spherical hydrometeor shapes. For this exercise, using simple spherical shapes. PR-based classification (joint σ 0, emissivity) (Durden et. al., 2012) Will use this to separate geographically at the end
A-Priori Database: Hydrometeor Size Distribution Variability Conventional modified gamma: Use mass spectrum parameters: Value and breadth of σ m increases as D m increases Value and breadth of μ decreases as D m increases (Haddad et. al., 1996) C. R. Williams, V. N. Bringi, L. D. Carey, V. Chandrasekar, P. N. Gatlin, Z. S. Haddad, R. Meneghini, S. Joseph Munchak, S. W. Nesbitt and W. A. Petersen, "Describing the Shape of Raindrop Size Distributions Using Uncorrelated Raindrop Mass Spectrum Parameters", Journal of Applied Meteorology and Climatology, 53, no. 5, p.. 1282-1296, 2014. Introduce HSD variability by varying the coefficients of the mass spectrum power law relations
A-Priori Database Construction TRMM Orbit 64546 15 March 2009 04:09:22 UTC Over Argentina at 32.7S, 61W Mean absolute Z14 diff= 1.9 db, max= 5.2 db TB RMS diff= 9.3 K, max= 11.7 K CDF of overall agreement: 95% of profiles match within 11K RMS difference (no-rain scenes, black line) and 20K (rain scenes, red line). 95% of PR reflectivity profiles agree to within 3 db mean absolute difference (blue line). Radar modeling of heavy rain events and ice phase needs improvement (Kuo HSDs, mixed phase)
A-Priori Database Construction two years 1B11 and 2A25, over-land only colocate TMI/PR The PR-only (2A25) mass content retrieval is used to build the HSD profile. HSD LUT Vary c 1, c 2 = mass spectrum coefficients for M-D m and σ m -D m relations physical emissivity LUT Vary the two surface control indices add MERRA reanalysis (c 1,c 2 ) to best match PR Z 14 vertical bin profile start with mean(e), vary by cov(e), based off e 10H best match TMI 9xTB in RMS diff Z 14 simulations to match observed (Z 14, M) Then, do TB simulations to match TMI TB PCA-based emissivity LUT database with physical emissivity database with PCbased emissivity
15-March-2009 10H TMI 15-March-2009 10H Simulated
15-March-2009 85H TMI 15-March-2009 85H Simulated
Synthetic retrieval using 2008 PC-based emissivity database Comparisons of simulated retrievals to actual TRMM observations, for all TRMM orbits on the 10 th day of each month 2009. All orbit data in black dots, and the subset of lightly vegetated scenes (VWC < 2 kg/m 2 ) in red. Synthetic retrieval using 2008 physical propertiesbased emissivity database Rain Column Ice Column Prob(Precip)
Search nearest SM bin, nearest VWC bin Class N RMSD Correlation Bias All 55620 0.569 0.848-0.100 2 1414 0.400 0.801-0.082 3 5515 0.609 0.858-0.116 4 2913 0.449 0.843-0.091 5 10481 0.606 0.861-0.097 6 12201 0.578 0.853-0.108 7 10145 0.522 0.836-0.072 8 7546 0.561 0.839-0.112 9 5405 0.627 0.833-0.117 All SM bins, nearest VWC bin Class N RMSD Correlation Bias All 58006 0.574 0.881-0.095 2 1475 0.441 0.862-0.072 3 5832 0.607 0.886-0.116 4 3046 0.484 0.875-0.095 5 10937 0.600 0.897-0.097 6 12763 0.578 0.886-0.104 7 10532 0.504 0.889-0.059 8 7833 0.592 0.862-0.108 9 5588 0.650 0.838-0.112 Summary from 336 TRMM orbits spread across all months of 2009 As a function of the PR land classification index Not much statistically significant difference in column rain. Why? Nearest SM bin, all VWC bins Class N RMSD Correlation Bias All 57928 0.568 0.882-0.100 2 1478 0.378 0.901-0.079 3 5815 0.606 0.895-0.131 4 3049 0.472 0.890-0.097 5 10915 0.587 0.898-0.101 6 12708 0.572 0.886-0.105 7 10545 0.516 0.881-0.063 8 7829 0.574 0.869-0.113 9 5589 0.654 0.842-0.110
Summary In these cases, no clear distinction between either type of surface control parameters on overall retrieval performance. Even if the PR profiles are well-matched, the Bayesian TB weighting is heavily affected by the wide variability in the 85-GHz simulations, overwhelming any genuine surface effects. Improved high-frequency hydrometeor modeling and land surface modeling go hand-in-hand. PR (DPR) radar classification appears to identify km-scale surface properties related to dynamically changes to land surface. Marry classification and land surface physics through a dynamic, adaptive land classification? (this has been tested for scatterometer-based soil moisture). Comments for Research Working Group Simply total column water vapor may be insufficient to capture the information embedded in the nature of a 166/183 GHz type observation. Possible effects of missing cloud upon high-frequency passive observations.
2014/05/07 2143 UTC N. Russia Light Rain w/bright Band The cloud that is really above where DPR says the cloud top is Value of 166 GHz for over-land high latitude discrimination ICAP-2014 21-23 October 2014 Boulder, CO
2014/06/07 0724 UTC Bhutan Himalayas Orographic convective structure smoothed over at 4- km DPR resolution ICAP-2014 21-23 October 2014 Boulder, CO
2014/06/03 0603 UTC E. Kalimantan Isolated Small-Scale Convection convective structure shows up at single pixel resolution in 89 and 166 GHz ICAP-2014 21-23 October 2014
Synthetic Retrieval different year 1B11 and 2A25, over-land only Identical procedures as for database generation colocate TMI/PR add MERRA reanalysis optional test Adjust PCs Y Locate entries within (Ts, PC 1, PC 2, TQV) bins Rain > threshold? Y Set SM to saturation Locate entries within (Ts, SM, VWC, TQV) bins Compare to observed Bayesian retrieval Bayesian retrieval
Replicating Observations: Simulated vs. Observed 2008 clear scenes (no columnar water) Ts < 273K in red 2008 cloudy scenes (any columnar water) T2m < 275 K and (T2m-Ts) > 2 in red (snow) In general, at cold end 10H underestimated, 85H overestimated