Using the Autonomous Modular Sensor (AMS) to Validate Satellite-Retrieved Sub-Pixel Fire Area: Radiative Flux of Wildfires and Fire Weather
David PetersonNational Research Council – Monterey, CA
Edward Hyer, Naval Research Laboratory – MontereyJun Wang, University of Nebraska – LincolnCharles Ichoku, NASA Goddard Space Flight CenterVincent Ambrosia, NASA Ames Research CenterAutonomous Modular Sensor Airborne Science Applications Use Workshop, 04/18/2013
NRL’s FLAMBE
General Goal: Improve the Prediction of Smoke Emissions
Reid et al. (2009)
Smoke Transport Modeling
http://alg.umbc.edu/usaq/images/
Highlights of this Talk…
1. Sub-pixel-based calculation of fire intensity
2. AMS validation: general
3. AMS validation: background temperature
4. Short-term predictor of satellite fire activity
Advantages of FRPp over Standard Fire Counts Quantitative indicator of fire intensity (Ichoku et al, 2008) Proportional to amount of biomass consumed (Wooster et al., 2005) Proportional to amount of smoke released (Ichoku and Kaufman, 2005) Related to the smoke plume height (Val Martin et al., 2010)
High fire temp.
Small fire area
Cooler fire temp.
Large fire area
MODIS Pixel #1 MODIS Pixel #2
These pixels have equal FRP?
We need FRP per fire area!
Current MODIS Fire Radiative Power (FRPp)
MODIS pixel-level FRPp (Kaufman et al., 1998)
Fit-line to many theoretical fire
scenarios!
Current FRP Limitation (collection 5)FRP is currently derived over the pixel area
Ap Ap
TfTb
Af
FRP and Initial plume buoyancy, Kahn et al. (2007) & Val Martin et al. (2010)
Sapkota et al. (2005)
We need high-resolution validation data for fire area, temp., and FRP!
Improved Sub-Pixel-Based Fire Radiative Power (FRPf)
Based on retrieved fire area (Af) & temperature (Tf):
The flux of FRPf can also be calculated:
Retrieval details are provided in: Peterson et al. & Peterson and Wang (2013), Remote Sens. Env.
Is the smoke contained within the boundary layer?
n
if
n
if
f
i
i
A
FRPFluxFRP
1
1
Units: MW
Units: Wm-2
AMS Pixel Resolution• Varies from 3 - 50 meters• Scan-to-scan differences• Topography• Flight Altitude varies
Limitations…• 4 µm channel saturates at
~510 K!• Can’t use for FRP
validation!
4000 to 9000 AMS data points per MODIS fire pixel!
AMS fire area assessment algorithm developed byPeterson et al. (2013), Remote Sens. Env.
NASA’s Ikhana
MODIS Tb window: 8-21 valid pixels (Giglio et al., 2003)
Non-Fire Background Warmer than the MODIS Fire Pixel (11 µm)?
White = Tb error
All 3 fire pixels with Tb > Tfire contain diffuse or pixel-edge hot spots!
Cooler AMS fire temps…
Peterson and Wang (2013), Remote Sens. Env.
Background Temperature Errors
Day Error: 151 (26%)
Night Error: 6 (2%)
Background Temperature Investigation
2011 Texas Wildfires
The In-Pixel Background Temperature
Error bars show the variability within the background region of a fire pixel
MODIS vs. AMS Background Temperature (California, 2007)
Variation: 5-10 K Variation: 1-5 K
Peterson and Wang (2013), Remote Sens. Env.
Retrieval’s Sensitivity to Background Temperature
ΔTb = ± 5.0 K ΔTb = ± 1.0 K
Simulate potential errors in background temperatureRetrieved Fire Area (4 and 11 µm)
Large sensitivity to a small Tb error
Incomplete error bars indicate Tb > Tpixel
Small sensitivity to a large Tb error
2011 Texas Wildfires
Fire Pixel Clustering Alleviates Random Error
1. Do fire observations contain information to identify potential for high injection/blowups?
2. How can we use weather information to make automated short-term forecasts of emissions for AQ models?
3. How can we use weather information to improve smoke emission estimates in near-real-time and retrospectively?
Fire Weather Application
NARR Domain (~32 km)
Choosing FRPf flux over fire counts?
Ongoing fire growth/intensity
inflow/circulation
MODISAlaska
Observed (Day 2/Day 1)
Growth
Decay
Small symbols: < 10 fire counts on day 1
Toward Developing a Short-Term Predictor of Fire Activity
Peterson et al. (2013)
Atmospheric Environment
Toward Developing a Short-Term Predictor of Fire Activity
Persistence All Inputs Observation N RMSE RMSE % Change
2004 Development Test (MODIS)Overall 3192 18.3 15.9 -13.1Growth/Ignition 435 34.6 32.9 -5.0
Persistencea 2328 4.0 4.4 11.2Decay/Extinction 429 34.6 26.2 -24.3
2005 Independent Test (MODIS)Overall 1302 21.2 20.1 -5.2Growth/Ignition 228 33.5 34.0 1.5
Persistencea 857 4.1 4.1 0.5Decay/Extinction 217 38.1 33.8 -11.4
a Observed persistence is bounded by ±10 fire counts for MODIS.
RMSE statistics for the fire count prediction model compared to persistence…
We need multiple AMS observations for the same fire event!
Highlights
• Fire prediction model is an improvement over persistence.
• Best with cases of decay/extinction!
• Must overcome scan-to-scan variations!
• Can also be applied to geostationary data.
ASTER 8.3 µm image
VIIRS 4 µm image
VIIRS DNB image
VIIRS 11 µm image
Fort Collins
Denver
Greeley
Loveland
Boulder
High Park Fire
~20 km
~23 km x 23 km
~37 km x 37 km ~37 km x 37 km VIIRS DNB image of the
Denver / Front Range areaT
Images by: Tom Polivka, UNL
~37 km x 37 km
Potential VIIRS Day-Night Band (DNB) Applications
Value of AMS Data Collocated with a Satellite Overpass• Valuable validation tool for retrieved fire area, background temp., etc.• Non-saturated 4 µm data are required for fire temp and FRP validations!• Repeat looks are very useful for both applications and validation!
Background Temperature• The AMS can identify reasons for errors in the MODIS background temp.• Important component of the sub-pixel retrieval's sensitivity analysis!
Fire Weather and Changes in Smoke Emissions• A short-term predictor of fire counts has been developed, may also use FRPf • We need daily AMS observations from the same fire event!• Can we use the AMS before and after a significant change in meteorology?
Future Goals• Retrieve FRPf flux using the next generation satellite sensors • Investigate potential applications using the VIIRS DNB• We need AMS collocations with VIIRS, especially at night!
VIIRS
Summary and Conclusions
Thank You!
Acknowledgements and Related PublicationsNational Research Council Postdoctoral Fellowship
NASA Earth and Space Science FellowshipNaval Research Enterprise Intern Program
NASA Nebraska Space Grant
Peterson, D., Wang, J., Ichoku, C., Hyer, E., & Ambrosia, V.: A sub-pixel-based calculation of fire radiative power from MODIS observations: 1. Algorithm development and initial assessment, Remote Sensing of Environment, 129, 262-279, 2013.
Peterson, D., & Wang, J.: A sub-pixel-based calculation of fire radiative power from MODIS observations: 2. Sensitivity analysis and potential fire weather application, Remote Sensing of Environment, 129, 231-249, 2013.
Peterson, D., Hyer, E., & Wang. J.: A short-term predictor of satellite-observed fire activity in the North American boreal forest: toward improving the prediction of smoke emissions, Atmospheric Environment, 71, 304-310, 2013.
Tf & P
Lb
1 km2
Modified from Dozier (1981)Calculations per MODIS pixel:
Pixel radiance = fire + background
Bi-Spectral Retrieval:L4 = τ4PB(λ4,Tf) + (1-P)L4b
L11 = τ11PB(λ11,Tf) + (1-P)L11b
Where: Tf = fire (kinetic)
temperatureLb = background
radianceP = fire area fractionB(λ,Tf) = IR Planck FunctionΤ =
atmospheric transmittanceL = pixel
radiance
Radiance (L) or Brightness Temperature?
Predefined Lookup Tables:(4 and 11 µm)
• SBDART Model• Atmospheric effects• Geometry• Surface temp. variations
MODIS Sub-Pixel Retrieval Inputs1. Geolocation data (solar/sensor zenith, azimuth)2. Level 1B pixel radiances3. Fire product background temperatures (4 and 11 µm)
MODIS Pixel Overlap Correction and sub-pixel calculations (iterations)
Pixel-Level Retrievals:One output per pixel
Single Retrieval via Averages:One retrieval for all fire pixels corresponding to the cluster
Output:•Fire area fraction and retrieved fire area (Af, in
km2)•Surface kinetic fire temperature (Tf)
Calculation of Sub-Pixel-Based FRPf
General Summation Method: All pixel-level fire area retrievals are summed
Clustering-Level Retrievals:
Zaca Fire Example
Instrument DetailsAmbrosia & Wegener (2009)
• Range: ~ 4000 miles
• Flight altitude can vary
• 12 spectral channels
• Fires detected at 4 (3.75) and 11 (10.76) µm
• Flight domain: western United States
Autonomous Modular Sensor (AMS)Flight Path: 8/16/2007
NASA’s Ikhana
Creating an AMS Fire Mask for Each MODIS pixel
Goals • Obtain actively burning fires
- Remaining data are disregarded• Obtain background temperature
Challenges• Saturation at 4 µm (not at 11 µm)• Scan-to-scan variations• Diurnal effects
Approach• Calculate minimum thresholds• Two fire thresholds (4 and 11 µm)• Search for regions of low density
within the histograms- Fire thresholds vary per MODIS pixel
• Day/night algorithm• Consider variation of AMS and MODIS
pixel size• Calculate AMS fire fraction
Fire Hot Spots
Background
Saturation Problem
Smoldering or cooling
AMS Data Within Several MODIS Pixels
4 µm
11 µm
AMS Data Within Several MODIS Pixels
Fire Hot Spots
Background
Smoldering or cooling
No Saturation