Logout succeed
Logout succeed. See you again!

NASA Technical Reports Server (NTRS) 20120004163: Application of Artificial Neural Networks to the Development of Improved Multi-Sensor Retrievals of Near-Surface Air Temperature and Humidity Over Ocean PDF
Preview NASA Technical Reports Server (NTRS) 20120004163: Application of Artificial Neural Networks to the Development of Improved Multi-Sensor Retrievals of Near-Surface Air Temperature and Humidity Over Ocean
Application of artificial neural nneettwwoorrkkss ttoo tthhee ddeevveellooppmmeenntt ooff improved multi‐sensor retrievals of near‐surface air temperature and hhuummiiddiittyy oovveerr oocceeaann J. Brent Roberts1, Franklin R. Robertson1, Carol Anne Clayson* 11 NNAASSAA//MMSSFFCC, EEaarrtthh SScciieennccee OOffffiiccee * Woods Hole Oceanographic Institution Outline • BBackkgroundd • Approach • Results • CConcllusiions andd FFutture WWorkk Motivation for Retrieving Surface Parameters • Estimating the surface heat fluxes from satellite requires: • Sea surface temp (SST) • Specific humidity (Qa) • Air temperature (Ta) • WWiindd speedd ((WWspdd)) • Current estimates show ssyysstteemmaattiicc ddiiffffeerreenncceess ooff 25‐50Wm‐2 • QQa & Ta differences are a major driver of the differences between these products. Large‐scale patterns are similar but amplitudes can be very different. Retrievals of near‐surface parameters from microwave brightness temperatures • Observations at microwave frequencies shhow ddependdenciies on: • Water Vapor (QV) • Surface wind speed • SSea SSurfface Temperature • TThhiis sensiittiiviitty iis sttatte dependent • Presence of clouds • Sensitivity to surface layer (i.e. within 10m) is llooww Based on simulations from CRTM Forward and Jacobian model. Sources of information in successful retrievals of near‐ surface temperature and humidity • There is a strong connection between the near surface air‐temperature and humidity. • Clausius‐Clapeyron • The sea surface temperature and air temperature are typically strongly correlated • Narrow distribution of (SST‐TA) • SSttuddiies hhave shhown ttottall collumnar watter vapor (precipitable water) and surface air temperature to be highly correlated (Liu, 1988). • Nonlinearity arises: • Dependence on atmospheric state • Deppendence on surface conditions • Inherent relationships between moisture and temperature. From Roberts et al. (2010) Inverse retrieval approach TB F ( X ) 1 X F (TB) GOAL: FIND F‐1() LINEAR NON‐LINEAR • Stepwise linear regression • Neural Network (Jones et al., (Jackson et al.,2006) 1999) •• GGeenneettiicc AAllggoorriitthhmmss ((SSiinngghh eett al., 2006) • Neural Network with first guess (Roberts et al., 2010) Data Fusion: Merging AMSR‐E and AMSU‐A Training dataset • Direct in situ measurements are co‐ llooccaatteedd wwiitthh ssaatteelllliittee‐ observations. • CCRRTTMM‐bbaasseedd ssiimmuullaattiioonnss can be used to AMSR‐E and AMSU‐A sensors on AQUA have co‐ supplement the in situ located footprints with minimal time between ddaattaasseett. samples. • Co‐located measurements between AMSR‐E and AMSU‐A are available from mid‐2002. Improved Surface Humidity Retrieval AQUA Advantage • AMSU‐A contains channels sensitive to lower troppospphere temperature • AAMMSSRR‐EE ccoonnttaaiinnss channels sensitive to PW, CLW, and SST • Results in improved surface humidity retriievalls. IImmpprroovveedd SSuurrffaaccee TTeemmppeerraattuurree RReettrriieevvaall • Overall improvements are ffoundd ffor near‐surfface temperature • The near‐surface stability is also better represented. • Improved by taking information directly related ttoo tthhee ssuurrffaaccee tteemmppeerraattuurree and temperatures in the lower troposphere. Conclusions • A statistical retrieval methodology for surface parameters iis iimprovedd usiing a nonlliinear approachh • Due to nonlinear nature of the problem • Retrieval of the near-surface parameters is improved through use of multiple sensors • AAddddiittiioonnaall iinnffoorrmmaattiioonn iiss aavvaaiillaabbllee ffoorr iinnvveerrssiioonn • It is important to include a synthetic component of the ttrraaiinniinngg ddaattaasseett;; cchhooiicceess aarriisseess rreeggaarrddiinngg ssaammpplliinngg • Future work : add a priori information to help regularize the nneettwwoorrkk ((ii.ee. aa BBaayyeessiiaann aapppprrooaacchh)).