Fire Locating and Modeling of Burning Emissions (FLAMBÉ): 7 years of lessons and prospects

Transcription

1 Fire Locating and Modeling of Burning Emissions (FLAMBÉ): 7 years of lessons and prospects Jeffrey S. Reid, Edward J. Hyer, Douglas L. Westphal, Cynthia A. Curtis, Jianglong Zhang and Kim Richardson Naval Research Laboratory, Monterey CA Elaine M. Prins and Christopher C. Schmidt CIMSS, University of Wisconsin, Madison WI Sundar A. Christopher and Jun Wang* University of Alabama, Huntsville AL *Now at Harvard University University of Maryland MODIS Team

2 FLAMBÉ Program Multi agency sponsorship: NASA, ONR, NOAA Common cause consortium approach in basic and applied research with a final goal of operational transitions to NWP. Adapt best practices where feasible. Technology push rather than pull. Open data and methods policy.

3 FLAMBE Timeline 1999: FLAMBE project conceived to fill the independent needs of core investigators. Fire product development, radiative forcing, and operations. 2000: Program funded by NASA IDS with leveraging with ONR and NOAA. 2001: FLAMBE runs quasi operationally at NRL for NAAPS using GOES WF_ABBA 2002: Near real time fire data distribution online. Initiated second stage temporal filtering 2003: U of Maryland MODIS added to quasi-operational runs 2005: FLAMBE ported to RAMS-AROMA (Meso-scale) 2006: FLAMBE transitioned to FNMOC, first truly operational global smoke model. NAAPS has 10,000 hits per day, FLAMBE website pushes ¼ Gb day. data 2007: Quasi-operational AOT assimilation. Include MSG and Asian Satellite?. Version 2 of source function.

4 Requirements for an Operational System Needs to be truly operational. Case study or regular human interactions is not an option, neither is regular tweaking. 24/7/365/fully automated/ no downtime. Do you sleep at 4 AM? Needs operational data feeds. Remember, MODIS is not operational, although it acts like one. Algorithms need to be robust and consistent. Operations folks do not like change. Simple works. Consider how often geostationary satellites change. Needs to be fast: <15 minutes. Fire hot-spot locations by operational satellite is the only option for fire identification. Geostationary systems most likely to be the standard bearer.

5 FLAMBÉ Step By Step Remote Sensing Fire Hotspot Detection: WF_ABBA & MODIS from SSEC/NESDIS and Maryland/NESDIS NRPTE (other geostationary data coming) Fire Characteristics: ABBA, Dozier fire size; MODIS: Static (FRP in 2007?) Ecosystem Assignment: AVHRR GLCC (MODIS by 2007) Emissions Product: Forward model. Literature values for intensive properties, add diurnal cycle for MODIS. Transport: NAAPS (Global-offline); RAMS-AROMA, COAMPS (mesoscaleonline). Run starts at T=-24 and uses persistence at T=0+ AOT Assimilation: J. Zhang prototyped over ocean MODIS AOT globally, J. Wang nudging in RAMS-AROMA at mesoscale Radiation: Fu-Liou

6 Lesson 1: Proper Uncertainty Analysis (Westphal s First Law) No aerosol model can be any better than the underlying meteorology model. Or It does not matter how sophisticated your aerosol parameterizations are if the wind is blowing in the wrong direction. Corollary: The same can be said for any underlying premise. Emissions: Land cover?

7 An Example: Model Precipitation Differences between model and observed rainfall NOGAPS 24 hr Precip. Oct : 0.5 degree Turk 24 hr Blended Microwave Product Oct : 0.25 degree Scale differences important for wet deposition

8 An example showing the convse: Long range transport mostly impacted by synoptic patterns rather than any source function or microphysics skill.

9 Lesson 2: Biomass burning, and aerosol systems in general, is a highly underdetermined system (Westphal s Second Law) The global aerosol system is so underdetermined that one should take consistency over accuracy. Corollary#1: By tweaking your model to work you have probably broken something along the way. Corrolary#2: Spend your time characterizing the product you have rather than chasing butterflies.

10 An Example: Aerosol Optical Depth Because AOT is one of the few observables, every model uses it as a key goal. But what are the uncertainties of the input terms? Intensive: Fuel Load: Factor of 2-3 Emission Factors: 30-60% Combustion fraction:20% NAAPS AOT MODIS AOT Extensive: Fire Sampling: Factor of 2-4 Burn Area: Factor of 2 Navigation/Ecosystem:? Ratio AOT CS Bias

11 Lesson 3: The biomass burning problem is fundamentally hindered by scale differences in research Different tools at different scales The world is a non-linear place. Mesoscale Grid Aircraft MODIS L2/LES Domain Photo: ISS Fire Tower AERONET Global Grid

12 Near Field Considerations Particle formation essentially a condensational process. It is not unreasonable that material still condenses for ~ an hour after emission. Particle mass growth on the order of 40% observed directly and through receptor analysis. Coagulation and chemistry are strongly non-linear in evolving plumes. Can t measure most of reactive species. Emissions are a result of non-linear combination of flaming/smoldering Bottom line, emission factors may need to be adjusted up from the beginning. Strong impact on interpretation of mass extinction efficiency and absorption Variable impact in emissions inventories dv/dln dp (CO2 Normalized) AVIRIS Quinault Gasso et al C0 Normalized Volume hours Source Diameter, d (mm) p

13 Mesoscale is where scale issues are most difficult, consider Central America. Demonstration of RAMS-AROMA Simulation (Wang et al., 2006 JGR) MODIS Obs. 12:00 CDT, 10 May 2003 Modeled smoke mass near sfc. L H

14 Lesson 4: On a case by case basis simulation can look remarkably good even at mesoscale (J. Wang et al. 2006, JGR) Smoke path way comparison in 30days Correlations between daily PM 2.5 and modeled smoke mass are significant in 23/36 stations. In 30day avg., smoke events increased PM 2.5 concentration by 40%-60%. Diurnal information in GOES beneficial to the simulation

15 Lesson 5: Be realistic. We need to examine the best ways to utilize multiple satellite data sets and land databases in a simple consistent manner. For smoke source functions, this is where the rubber meets the road. Propagating the error will difficult. MODIS ABBA

16 FLAMBÉ s Future: Emissions product Version 1.1 by February. South America Regional correction. Incorporate WF_ABBA metadata. Switch to MODIS landcover. Version 2 by summer. Test FRP and other fire model formulations. Determine what are the true uncertainties in fire characterization. Include WF-ABBA SEVIRI? Version 2DA: Inverse modeling on the fly???

17 FLAMBÉ Future: Data Assimilation Data quality is finally to a point where data assimilation is possible (not just fire product utilization). There are two immediate applications: Extensive aerosol optical properties Emissions model Operational AOT assimilation is coming soon DA based emissions model is more of a long term goal. Working to develop cal/val programs in Africa (SAMI-B) with Witts and SE Asia (7-SEAS) with BPPT-Indonesia.

18 AOT Over Ocean Data Assimilation J. Zhang, UCAR/NRL NAAPS 120 hr Forecast 3 day Composite MODIS ECL=100 km; BEV=0.1τ Forecast with NAVDAS 2-D var data assimilation ECL=385 km; BEV=0.3τ

19 Over Land? MODIS ver. 4 and AERONET AOT Need time to evaluate ver. 5 J. Zhang (NRL/UCAR) R > < R < 0.9 Mixed Dust Sulfate Smoke 0.5 < R < 0.7 R < 0.5

20 Should we utilize data assimilation for source model? Where our future probably lies, but have a number of issues: Goal is data assimilation: circular feedback with model Scale representation: Vector versus gridded (or both)? Source function feedback: After each AOT innovation how do we communicate with the source function? High AOT: Retrievals rejected by algorithm or at best highly uncertain. Multitude of data sources: We need to develop a framework for including varying fire datasets. Realistic error matrices: If not careful one can easily degrade your model.

21 FLAMBÉ: the Bottom Line Programs like FLAMBE are forcing the formation of new innovative shared resources and operational systems. Fire products are such a case. We need to work together. What often passes for aerosol model validation is really an assessment of advection and the baseline meteorology models. Even the worst model can look good with decent meteorology. There is no such thing as a meteorology data independent model. Indirect or receptor validation schemes are misleading and do not address the true degeneracy of the biomass burning system. More time needs to be spent on fundamental uncertainties. This means teaming up with colleagues. Scale bias has creped into most models. As a community we should set guidelines and test cases. This does not mean we should stifle innovation. NPOESS will not give us what we need. Geostationary fire products need to be fully utilized. Even if geostationary data providers have their own pet product, agreement must be made as to universal product guidelines.

22 Data Availability FLAMBÉ Archived fire products, fluxes and optical depths: FLAMBÉ Homepage: NRL Aerosol Homepage: NOAA/NESDIS ABBA Homepage: Also MODIS Fire: GSFC/Univ of Maryland: