31
Developing the next generation of fuel loading data useful for emissions modeling Nancy French (PI), Susan Prichard (Co-I) Anne Andreu, Michael Billmire, Paige Eagle, Eric Kasischke, Maureen Kennedy, Sim Larkin, Don McKenzie, Roger Ottmar, and Kjell Swedin JFSP project # 15-1-01-01

Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

Developing the next generation of fuel loading data useful for emissions modeling

Nancy French (PI), Susan Prichard (Co-I)Anne Andreu, Michael Billmire, Paige Eagle, Eric Kasischke, Maureen Kennedy, Sim Larkin, Don McKenzie, Roger Ottmar, and Kjell Swedin

JFSP project # 15-1-01-01

Page 2: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

2

Outline

• Background• Emissions modeling &

uncertainty• Wildland fuel

characterization• Project objectives

• Project status • Database development• Data visualization &

distribution• Sensitivity analysis

• Expected applications• Other Considerations

Page 3: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

3

Estimating Emissions

Et = A·βB ·Efg

Et is the total EmissionsA is the total Area burned (ha)β is the fraction of biomass/fuel consumed during fireB is the Biomass/fuel loading (t ha-1)Efg is the Emission Factor for each gas species

(g gas/kg fuel) [e.g. CO2, CO, CH4, NMHC]

Total Emissions:

CombustionFactors - ß

Area Burned - A

Fuel Consumption

Emission FactorsEmissions

Fuel Loading - B

Et = A·(βaBa + βbBb …)·EfgPartition fuels into strata

Page 4: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

4

FCCS Fuelbed Strata

Page 5: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

5

Estimating Emission Source

Fuel loading and the proportion of the fuel that is combusted (consumption) have highest uncertainty.

Fuel Loading

Fuel Consumption

Emission Factor

Emission Produced

Largest Error (CV= 83)

Second Largest Error (CV=30)

Smallest Error (CV=16)

Area Burned

Page 6: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

6

Predicted ECO2

Predicted ECH4

Predicted ECO

Emissions Modeling Uncertainty

French, N.H.F., P. Goovaerts and E.S. Kasischke (2004). Uncertainty in estimating carbon emissions from boreal forest fires. Journal of Geophysical Research 109: D14S08 doi: 10.1029/2003JD003635.

Using an Uncertainty Model for carbon emissions (next slide) we computed the distributions of predicted mean annual emissions of (a) CO2, (b) CO, and (3) CH4 for Alaska.

Page 7: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

7

Emissions Modeling Uncertainty

Uncertainty Model Method• Used a stratified random sampling of probability distributions for

each input parameter;• Monte Carlo simulation;• Each combination of sampled values was combined to retrieve

the corresponding simulated emission value.• Monte Carlo simulation provides a range of possible

outcomes and the probabilities they will occur for all combination of variables

PROBLEM: Implementation of the Monte Carlo simulation required

information regarding the characteristics of the probability distributions (shape, spread) of each fuelbedand strata.

Probability distributions of input parameters were not available, so we assumed Guassian distributions with spread based on published CVs

Page 8: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

8

Fuelbed Map•Fuelbeds are mapped via crosswalks to satellite-derived vegetation and land cover, at scales from < 25 m (landscape applications) to >36 km (continental and global applications).

•USGS Landfire 30-m scale existing vegetation maps were used for mapping FCCS

•A 1-km scale version is available for regional-scale modeling

http://www.fs.fed.us/pnw/fera/fccs/index.shtml

includes fuel loadings by type

Fuels Mapping: Mapping standard FCCS fuelbeds

Page 9: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

9

Standard Fuelbeds1-km resolution

234 fuelbeds (5 lost in aggregation)

Page 10: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

10

Emissions modeling

FCCS-based WFEIS/Consume (French et al.) BlueSky Framework (Larkin et al.)

Others: CanFIRE (de Groot et al.)

GFED (van der Werf et al.)

FINN (Wiednmyer et al.)

French, N.H.F., D. McKenzie, T. Erickson, B. Koziol, M. Billmire, K.A. Endsley, N.K.Y. Scheinerman, L. Jenkins, M.E. Miller, R. Ottmarand S. Prichard (2014). “Modeling regional-scale fire emissions with the Wildland Fire Emissions Information System”. Earth Interactions 18: 1-26 doi: 10.1175/EI-D-14-0002.1.

Page 11: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

11

Variability of Fuels

Forest/vegetation typeDuff depthConifer vs. deciduousForest structure & density

Lower Duff

Upper Duff

Live MossDead Moss

Mineral Soil

Boreal black spruce sites, for example, have varying amounts of duff.

Page 12: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

12

Existing Vegetation Type – FCCS Fuelbed Crosswalk

Southeast conifer sites can have sparse or dense understory shrubs.

Page 13: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

13

Page 14: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

14

Project Overview

Page 15: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

15

Fuel Loadings Database

Data Sources:• FIA plot data• LANDFIRE reference database• Natural fuels photo series data• Continuous Vegetation Survey Plots (USFS)• Source data for FOFEM development

(courtesy of Bob Keane)• Source data for Fuel Loading Model development

(courtesy of D. Lutes)• FCCS fuelbed development referencesCurrently “complete” but considered a “work in progress”

Page 16: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

16

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0

CANOPY

0 100

0.6

0.5

0.4

0.3

0.2

0.1

0 0 10 20

SHRUB3.5

3.0

2.5

2.0

1.5

1.0

0.5

00 2 4 6

HERB

Mg/ha

dens

ity

Sample distributions

Page 17: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

17

0.16

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0

COARSE WOOD

0 100

Mg/ha

0.25

0.20

0.15

0.10

0.05

0 0 20

LITTER

3.5

3.0

2.5

2.0

1.5

1.0

0.5

00 100

DUFF

dens

ity

Sample distributions

Page 18: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

18

Sensitivity Analysis

Example: Emissions (y-axis) and fuel loading (x-axis)

not sensitive to shrub loadings

but are to duff loadings

Page 19: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

19

Sensitivity Analysis

Explore the sensitivity of emissions estimates to uncertainty in fuel loadingsCandidate distributions:• normal • lognormal • gamma • weibull

weibull

gamma

Page 20: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

20

Estimate emissions for each sampled

value

Calculate distribution of

emissions

Sample loading values from fit distributions

Sensitivity Analysis

As in French et al. 2004

Page 21: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

21

Web-based application for visualizing uncertainty and fuel loading distributions by region and fuel category

Data access and visualization

Page 22: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

22

Two different uncertainties to be represented:1. Uncertainty of fuel loading Represented as probability

distributions

Visualizing uncertainty

2. Uncertainty of emissions resulting from fuelbed variability Represented as distribution of

values resulting from Monte Carlo simulation of fuel loading Probability Density Functions (PDFs)

Page 23: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

23

Local and global sensitivity analysis ranks fuels categories for their contribution to variability in emissions predictions

Cross-reference important fuels categories with data gaps found in Task 1a

Draw randomly from empirical joint distributions of important fuels categories, predict emissions for each

Prioritize resources for data acquisition

Produce expected distributions and prediction intervals for emissions estimates

Using the Database & Sensitivity Results

Page 24: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

24

Visualizing uncertainty

Challenges for geospatial display of uncertainty:– Each map pixel is multidimensional, having values for:

• multiple fuel strata• multiple measures of uncertainty for each strata (i.e.

lower bound, upper bound, standard deviation)• representations of uncertainty that are themselves

multidimensional (PDFs for each variable)– For a single map pixel, what is the optimal way to

simultaneously show pixel value + some measure(s) of pixel uncertainty?

Page 25: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

25

Possible solutions:– HSI model (aka whitening)

• hue represents the value• whiteness/opacity represents pixel uncertainty• Requires two-dimensional legend

Visualizing uncertainty

Hengl, T., 2003. Visualisation of uncertainty using the HSI colour model: computations with colours. In 7th International Conference on GeoComputation.

Page 26: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

26

Possible solutions (continued):– 3D mapping

• Where “elevation” can be used to represent value or uncertainty

• Example: value=elevation uncertainty=hue

Visualizing uncertainty

Sanyal et al. 2009. A user study to compare four uncertainty visualization methods for 1d and 2d datasets. IEEE transactions on visualization and computer graphics, 15(6) Figure 5c

Page 27: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

27

Possible solutions (continued):

– Animation• Automatic cycling between lower/mid/upper bounds

– Side-by-side views• Evidence that this is more effective than whitening

(Gerharz and Pebesma 2009) for targeted data analysis (as opposed to broad pattern display)

Visualizing uncertainty

Page 28: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

28

Applications

Emissions estimation

Page 29: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

29

Smoke and Air-quality Modelsdraw from probability distributions of mapped fuels rather than single-average values.Community Multiscale Air-quality System (CMAS)

Global Climate Modelsenable coupled models* to incorporate spatial variation in fuels when projecting uncertainty in GHG emissions.(*e.g., GCMs + land-surface models + smoke dispersion models) Note: this methodology can be extended, in theory, to coarse-scale GCMs

Applications

Page 30: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

30

Other Considerations

Improving methods for characterizing & mapping– Advancing measurement methodologies, such as LiDAR-

based fuel density– Map improvements and validation (a part of this project)

Quantifying consumption & emissions with Thermal IR Fire Radiative Energy (FRE)– Proven satellite-based method operationally used in Europe– Method is reliable independent fuel-loadings

• more energy = more fuel consumption– Fuels are still important to know for

• Emission factors depend on material burning• Flaming vs. smoldering is not well studied• Need for uncertainty (this study)

Page 31: Developing the next generation of fuel loading data …...1.5 1.0 0.5 0 0 100 DUFF density Sample distributions 18 Sensitivity Analysis Example: Emissions (y-axis) and fuel loading

31

Thank-You

31