Wildland Fire Emissions Study – Phase 2 For WRAP FEJF Meeting Research in progress by the CAMFER fire group: Peng Gong, Ruiliang Pu, Presented by Nick Clinton

U.C. Berkeley –2 Purpose “…to develop a method for producing coherent, consistent, spatially and temporally resolved GIS based emission estimates for wildfire and prescribed burning.”

U.C. Berkeley –3 User Interface Vegetation Crosswalk Fuel Models Emission Estimation Fuel Loading Fuel Consumption Vegetation Coverage User Parameters Sum Modular System Fire History Map Emissions Reporting

Vegetation Data The GAP vegetation layer –Statewide coverage –Less complex than other vegetation layers such as CALVEG –1990 source data

National Inputs The spatial inputs are the NFDRS fuel model grid (seen left) and a grid of remotely sensed fire detections (both 1km resolution). Utilizes the same emissions equations as with polygon processing. Requires crosswalk of FOFEM fuel models to NFDRS fuel models (proof of concept).

Fire History – Agency Data CDF fire polygons Historical database Completeness?? Remote sensing based fire map

Algorithms A. Hotspot Detection (modified to CCRS’) Y ES NO AVHRR data preparation Algorithm applied to each pixel Test # 1 T3 > 315 K? Test # 2 T3 –T4>=14 K? Test # 3 T4>=260 K? Fire clear pixels Eliminate cloudy pixel Eliminate warm background, e.g., bare soil

YES NO YES NO YES NO YES NO YES NO YES NO Test # 4 Contextual info R2<=30%?R2<=8 neighb P ave-1? T3>8 neighb P ave+5? Test # 5 Wild land cover types? Test # 8 |R1-R2|>1%? Test # 7 R1+R2<=75%? Test # 6 T4-T5<4.0 K and T3-T4>=19 K? Test # 9 One of neighbor P passes the 8 tests above? True fire pixels False fire pixels Eliminate highly reflecting clouds & surface and warm background Eliminate urban, agriculture, dune, desert, water body Eliminate single fire pixel Eliminate sunglint pixels Eliminate highly reflecting clouds & surface Eliminate thin clouds with warm background Single date fire mask

Algorithms B. Burnt Scar mapping (modified to CCRS’ HANDS) with - Two NDVI composites of an interesting interval - One corresponding hotspot composite (fire mask) Step 1. Normalize NDVIpost to NDVIpre normalized NDVIpost = Ratio.C * NDVIpost Step 2. Calculate NDVI difference normalized NDVIpost – NDVIpre Step 3. Confirm hotspot pixels using NDVI difference (CBP)

A CBP is assumed to have a negative NDVI difference Step 4. Calculate NDVI difference statistics (mean, SD) of CBP for each landscape type Step 5. Select potential burnt scar pixels (BSPs) A BSP NDVI difference <mean + c*SD (CBP), c can be 0~1 Step 6. Apply a sieve filter to BSPs Filter out a burnt patch of < 2 pixels Step 7. Confirm a BSP with a neighbor CBP and later on a neighbor BSP to create connected burn patches One to four neighbor CBP, BSP to be used for the confirmation Step 8. Filter out a BSP patch of < 2 pixels and false burnt patch Step 9. Output burnt area mask (in TIFF format)

Fire History – RS Data Overlay of CDF and CAMFER data 1996 and 1999 (big fire years)

Overlay of CDF and CAMFER

Quantitative Comparison

Variation in mapping success between different ecosystem types. The amount of variation differs between methods (monthly or annual differencing), and between years. In general, the CAMFER method is more successful in the forest type.

Overlay of CDF and CAMFER RED is now RS detections. Green is Jepson ecoregion Lambert Conformal Conic Projection No Post-processing (filtering, nearest neighbor relationship to hotspots) Slightly reduced accuracy Potential for more data refinement by incorporating hotspots…

Overlay of CDF and CAMFER Green is annual NDVI differencing. Blue is monthly NDVI differencing Neither method is effective in detecting the entire burn area

Overlay of CDF and CAMFER Hotspots (Red) overlaid on the monthly and annual NDVI differencing Increase or at least negligible decrease in NDVI, especially over an annual time scale Problems with temporal resolution in hotspot detection Potential for more dynamic thresholding in burn scar mapping?

Temporal Decomposition of RS Data Remotely sensed burn scar polygons can be decomposed to daily polygons based on a nearest neighbor relationship using hot spot detections Facilitates temporal allocation of emissions Useful to dispersion modeling, emissions tracking