WHY DIFFERENT PASSIVE MICROWAVE ALGORITHMS GIVE DIFFERENT SOIL

WHY DIFFERENT PASSIVE MICROWAVE ALGORITHMS GIVE DIFFERENT SOIL MOISTURE RETRIEVALS Xiwu Zhan, Jicheng Liu NOAA-NESDIS Center for Satellite Applications and Research, Camp Springs, MD Thomas Holmes, Wade Crow, Tom Jackson USDA-ARS Hydrology and Remote Sensing Lab, Beltsville, MD Steven Chan NASA-JPL, Pasadena, CA IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011 X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 1

OUTLINE v Current PM SM Data Products v Single-Channel vs Multi-Channel Algorithms v Uncertainty Sensitivity Analysis v Summary X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 2

Current Satellite Soil Moisture Data Products: ü GSFC SMMR (Owe et al, 2001) ü USDA TMI (Bindlish et al, 2003) ü Princeton TMI (Gao et al, 2006) ü NASA AMSR-E (Njoku et al, 2003) ü USDA AMSR-E (Jackson et al, 2007) ü VUA AMSR-E (Owe et al, 2008) ü USDA Wind. Sat (Jackson et al, 2008) ü NRL Wind. Sat (Li et al, 2008) X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 3

Multi-channel Inversion (MCI) Algorithm : (Njoku & Li, 1999) TB, icmp = Ts {er, i exp (- i/cos ) + (1 – ) [1 – exp (- i/cos )] [1 + (1 -er, i)exp (- i/cos )]} i = b *VWC er, i = f(es, h) es = f(ε) -- Fresnel Equation ε = f(SM) -- Mixing model (Dobson et al) TB, iobs = TB 06 h , TB 06 v , TB 10 h , TB 10 v , TB 18 h , TB 18 v X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 4

Land Parameter Retrieval Model (LPRM) : (Owe, de Jeu & Holms, 2008) TBhcmp = Ts {eh, r exp (- /cos ) + (1 – ) [1 – exp (- /cos )] [1 + (1 - eh, r)exp (- /cos )]} = f(MPDI) , MPDI = (TBv-TBh)/(TBv+TBh) eh = f(es, h, Q) es = f(ε) -- Fresnel Equation ε = f(SM) -- Mixing model (Wang & Schmugge) Ts = f(TB 37 v) or Ts. LSM TBhobs = TB 06 h , TB 10 h or TB 18 h X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 5

Single Channel Retrieval Algorithm (SCA) : (Jackson, 1993) TB 10 h = Ts [1 –(1 -er) exp (-2 /cos )] = b * VWC, VWC = f(NDVI) eh es ε Ts = f(ev, h, Q) = f(ε) -- Fresnel Equation = f(SM) -- Mixing model = f(TB 37 v) or Ts. LSM X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 6

Retrieval Results: SM: Aug 4, 2010 SCR X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. MCI LPRM 7

Retrieval Results: SM: Aug 5, 2010 SCA X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. MCI LPRM 8

Retrieval Results: SM: Aug 6, 2010 SCA X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. MCI LPRM 9

Retrieval Results: NDVI/VWC/tau: Aug 4, 2010 MCI SCA X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. LPRM 10

Retrieval Results: NDVI/VWC/tau: Aug 5, 2010 MCI SCA X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. LPRM 11

Retrieval Results: NDVI/VWC/tau: Aug 6, 2010 MCI SCA X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. LPRM 12

Uncertainty Sensitivity Analysis Procedure: MCI and LPRM: 1. LPRM converges while MCI sometimes not; 2. Remove tau=f(MPDI) from LPRM and use Ts = f(Tb 37 v) for MCI; 3. Perturb Tb 37 v, Tbh & Tbv for LPRM and MCI to test how they are sensitive to their errors. X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 13

Land Parameter Retrieval Model (LPRM) : (Owe, de Jeu & Holms, 2008) TBhcmp = Ts {eh, r exp (- /cos ) + (1 – ) [1 – exp (- /cos )] [1 + (1 - eh, r)exp (- /cos )]} = f(MPDI) , MPDI = (TBv-TBh)/(TBv+TBh) eh = f(es, h, Q) es = f(ε) -- Fresnel Equation ε = f(SM) -- Mixing model (Wang & Schmugge) Ts = f(TB 37 v) TBhobs = TB 10 h X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 14

Multi-channel Inversion with LPRM (MCI) : TBhcmp = Ts {eh, r exp (- /cos ) + (1 – ) [1 – exp (- /cos )] [1 + (1 - eh, r)exp (- /cos )]} eh = f(es, h, Q) es = f(ε) -- Fresnel Equation ε = f(SM) -- Mixing model (Wang & Schmugge) Ts = f(TB 37 v) TBiobs= TB 10 h and TB 10 v X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 15

Impact of Tau = f(MPDI) on SM Retrievals: LPRM with tau = f(MPDI) MCI without tau = f(MPDI) X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 16

Impact of 2 K Ts error on LPRM/MCI Retrievals: Ts + 2 K Ts – 2 K X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 17

No Ts errors X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 18

Impact of 2 K Tb error on LPRM/MCI Retrievals: Tbh + 2 K Tbv - 2 K Tbh - 2 K Tbv + 2 K X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 19

No Tb errors X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 20

Uncertainty Sensitivity Analysis Procedure: SCA: 1. Use GLDAS SM inverse tau with SCA eqns; 2. Use the inversed tau to retrieve SM as reference; 3. Perturb Tb 37 v, Tbh for SCA retrievals to test how they are sensitive to these errors. X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 21

Inversed SM and Tau using SCA equns: tau SM X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 22

Impact of Tau error on SCA Retrievals: Tau + 0. 01 No Tau error X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 23

Impact of Tau error on SCA Retrievals: Tau + 0. 05 No Tau error X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 24

Impact of Tau error on SCA Retrievals: Tau + 0. 1 No Tau error X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 25

SUMMARY v v The difference of current satellite soil moisture products may confuse users. Single-Channel Algorithm relies heavily on accuracy of tau estimates. LPRM algorithm uses a tau-MPDI relationship and TB 37 v for Ts estimate to reduce iteration variable numbers in solution procedure. Its sensitivity to TB calibration, Ts estimate and other parameter errors needs to be assessed. Multi-channel Inversion algorithm is similar to LPRM algorithm when using the same Ts estimates. Thus, the tau-MPDI relationship may not be the key for the LPRM success. X. Zhan, NOAA-NESDIS-STAR, IGARSS 2011, Vancouver, Canada, 24 -27 July, 2011. 26