Use of open source tools to estimate global GHG emissions

1. State of global wildfires and carbon emissions.
2. Role of FOSS in geospatial applications.
3. How ghg emissions are measured from biomass burning ?
4. Pros and cons of existing approaches.
5. Our approach, results and validation.
6. Case studies and collaboration.
7. How to access the data ?
8. Challenges and what’s next for us ?
9. QA

Table of Contents

It is estimated that climate change has amplified global wildfires by 600%, emerging as the largest and most critical climate risk.

17M +

fire events as per NASA FIRMS

6.6B

tonnes estimated CO2 emissions from 2021 wildfires

4M +

km2 area burnt as per ESA CCI

600%

rise in wildfires due to climate change

State of global wildfire & carbon emissions

Role of FOSS in geospatial application

​

Enable easy transition for early stage

startups into the geospatial ecosystem.

​

​

Provides foundation through necessary tools and data, to explore use cases & applications.

​

​

Major FOSS projects that have enabled this study are QGIS, GDAL, PostgreSQL, PostGIS, and OSM.

1

​

Biomass emissions are calculated using the product of total fuel burnt and emission factor.

The emission factor represents the fraction of gas emitted per kg of fuel burnt.

Traditionally there have been two approaches for estimating biomass emissions i.e. Top-Down and Bottom-up.

2

3

4

How emissions are measured ?

​

Total fuel burnt represents the amount of vegetation being burnt.

​

• Emissions are estimated using burned area, biomass density, combustion efficiency and emission factor (Seiler and Crutzen 1980).

​

• Requires ground truth and small scale lab experiments. Prone to uncertainty.

​

• Works well overall but significantly underestimates small-scale fires contribution.

​

• Not appropriate for use in near real-time systems.

​

• Global Fire Emissions Database (GFED; van der Werf et al., 2010) is based on a similar technique.

Bottom Up

Time Consuming

Ground Truth & Small Scale Experiments

Data available months after fire event

​

• Top Down is based primarily on satellite observations.

​

• Rely on fire radiative power (FRP) observed through satellites, representing the rate of emitted radiative energy by the fire.

​

• Links FRP directly to the rate of fuel burnt using aerosols (AOD) (Wooster et al., 2005; Freeborn et al., 2008; Hudak et al., 2016).

​

• Eliminates the need for small-scale experiments.

​

• Global Fire Assimilation System (GFAS; Kaiser et al., 2012) is based on a similar technique.

Near Real Time

Top Down

Satellite Data

Active Fire Detection

Aim to develop a global dataset capable of generating hourly CO2 estimates across different biomes.

​

Choosing the top-down approach with select modifications to improve accuracy.

​

Overcoming limitations from the previous approaches and revising previous estimates.

Tracking a billion tonnes of GHG

​

• Detect Small Scale Fires - Ingesting hourly observed FRP from VIIRS SNPP satellite. Excluding any low-confidence fires and classifying vegetation types using high-resolution landcover maps.

​

• Eliminate Small Scale Experiment & Develop Generic Global Solution - Using machine learning to derive biome-specific combustion factors.

​

• Estimating the total fuel burnt directly from fire radiative energy (time integrated FRP) and total particulate matter (TPM) derived from aerosols (AOD).

​

• Use Updated Ancillary Data - Using the latest IPCC recommended emission factors for each trace gas.

Our Approach

Machine Learning

Satellite Data

Small Scale Fire Detection

Workflow

Fire Radiative Power

Provides the amount of radiative energy emitted from a fire and as such act as a quantitative measure of the strength of a fire.

Aerosol Optical Depth

Derived from MODIS MAIAC 1km at 550nm, validated against Aeronet. Combination of FRP & AOD are used to derive combustion factors using machine learning.

Vegetation Type

The 100m CGLS landcover map enables fire classification at high resolution. Helps in deriving combustion factor per vegetation type.

GFAS v1.2 (Kaiser et al. 2012)

GFED v4.1 (van der Werf et al. 2010)

​

​

Results & Validation

Avg. CO2 Emissions 2015-2020 (Gigatonnes)

Avg. CO2 Emissions 2015-2020 (Gigatonnes)

R2 Score: 0.7; Mean Absolute Error: 14 Mt CO2e/yr

R2 Score: 0.9; Mean Absolute Error: 9 Mt CO2e/yr

Regional View

Global View

​

​

More Results !

More Results !

Global View

Provides country level emissions statistics per vegetation type.

Country View

Provides states level emissions statistics per vegetation type.

County View

Provides county level emissions statistics per vegetation type.

Emission

• The dataset has been a subject of several case studies such as Understanding the impact of wildfires and importance of monitoring carbon sinks.

​

• More info at Research - Blue Sky Analytics.

​

• Also part of global coalition Climate Trace (climatetrace.org), an independent group for monitoring & publishing GHG emissions across different sectors.

​

• Annual countrywise emissions dataset available under Forestry and Land Use (except Forest Clearing) and Agriculture (cropland fires) via CC-By-4.0 framework.

Case Studies & Collaboration

Where

How

How to get the data ?

Access data through a single line of code via an API portal.

Python tutorials and documentation are available on blueskyhq.io/tutorials.

​

• “Snapshots” captured by polar satellites failed to include fire diurnal cycle.

​

• Use of geostationary can fix the issue but brings it own issue of coarser resolution and local coverage.

​

• Lack of actual ground truth often limits the scope of emissions estimates and the methodology.

​

Challenges

What’s next for us ?

Harmonised Fire Data

Harmonise geostationary and polar satellite data to improve fire diurnal cycle.

Carbon Sink

Ground Truth

Include carbon accounting from vegetation regrowth and estimate net carbon emissions.

Combine ghg observations from commercial satellites to validate estimates

QnA

QnA