This document discusses different strategies for modeling greenhouse gas emissions and carbon stocks, including empirical and process-based models. Empirical models rely on statistical relationships between activity data and emission factors, like those used in IPCC Tiers 1 and 2. Process-based models attempt to simulate underlying biogeochemical processes. Examples of common process models described are DAYCENT, DNDC, and RothC. The document outlines steps to find an appropriate process model, set it up, calibrate it using measurement data, and evaluate its performance through statistical tests. Challenges include limited data for model validation and parameterization for tropical conditions.
2. GHG emissions and carbon stocks are the result of
biogeochemical processes.
(Li, 2000)
Soil C stocks
3. Basic structure of models for GHG emission and C stocks
Strategies for modeling GHGs and C stocks dynamics can be
classified as empirical or process-based.
GHG emission | Δ C Stocks = Activity data x Emission Factor (EF)
Quantity of GHG | Δ C stocks
associated the Activity Data
tCO2 activity data-1 y-1
Activity influencing GHG Quantity
of GHG | Δ Soil C stocks
Number of animals (head)
Area (ha)
Fertilizer (t)
…
4. Empirical models rely on statistical relationships for which data is
available.
Activity data
Emission Factor (EF)
GHGemission
(Shcherbak et al., 2014)
5. (IPCC Tier 1) (IPCC Tier 2)
The IPCC Tiers 1 and 2 methodologies are examples of
empirical models.
(Giltrap et al., 2013)(FAO)
6. The IPCC Tiers 1 and 2 methodologies are examples of
empirical models.
7. Least squares fitting to a linear function result in equation below (r2 = 0.8):
E = 1 + 0.0125 x F
where E = emission (kg N20-N ) and F = fertilizer application rate (kg N ha-1 y-1).
This relationship was based on only 20 experiments, with measurements covering a full year; its global
applicability is highly uncertain.
Example of empirical model development / application
(Bowman et al. 1996)
Simple to use and
transparent
but are not sensitive for
a range of soil, climate
and farm management
practices.
8. Process-based models attempt to simulate the underlying biogeochemical
processes
Activity data
Emission Factor (EF)
GHGemission|ΔCstocks
(Li, 2000)
9. Site Validation Sensitivity tests Scale up
Process-based models can be more responsive to the effects
of soil properties, climate and management
(Li et al., 2013; Lugato et al., 2010)
10. Examples of common process models:
Events and management practices
such as fire, grazing, cultivation,
residue management, and organic
matter or fertilizer additions are
modeled.
Set of farming management
practices such as crop rotation,
tillage, residue management,
fertilization, manure amendment,
irrigation, flooding, grazing, etc.
• N2O, NOx, CH4, and CO2.
• Nitrate leaching loss.
• Soil carbon sequestration
• Crop development and
biomass yields.
Soil management, crop management.
• Soil carbon.
• N2O, NOx, CH4, and CO2.
• Nitrogen losses.
• Soil carbon sequestration
• Crop development and
biomass yields.
11. Examples of common process models:
Choosing a model
Does it perform well for my target scope (GHG emissions/Soil C Stocks)?
Has it been tested in similar conditions of mine (Country/Region /AgManag)?
Did it have a good performance/validation?
What was the main issues related to calibration and validations (data and parametrization)?
12. Find the appropriated process model: Literature review is one of
the first step
(Smith et al, 1997)
13. Set and run the model
• Climate
• Soil Properties
• Cropping / Livestock system and management
14. Step 1: The model is first run in “default” mode (DEF)
Step 2: Then run in calibration mode (CAL) using values for soil parameters that gave the closest fit to
measurements.
Step 3: Validation using another set
of data - determine the confidence
level in the model.
Calibrate and validate the model using measured values/data
(Rafique et al., 2011)
15. (Cui & Wang, 2019)
Step 1: The model is first run in “default” mode (Original)
Step 2: Then run in calibration mode (Modified) using
values for soil parameters that gave the closest fit to
measurements.
Calibrate and validate the model using measured values/data
16. R-squared (r2)
how close the data points around the
fitted regression line.
Root Mean Square Error (RMSE)
SD of the residuals (prediction errors) – indicates
how close the observed data points are to the
model’s predicted values.
0 – 100%
The Higher, the Better fit
(% of the variability of simulations
explained)
0 – 100%
The Lower, the Better fit.
(how accurate is the model)
does not provide a formal hypothesis test for
this relationship (F-test determines statistically
significant)
Model performance evaluation (Statistical analysis)
22. Availability of data to satisfy the input requirements of
models and understanding process driven GHG
emissions and C stocks.
Limitation/Challenge
Model’s parametrization for tropical conditions
23. Conclusion
• Statistical models are almost always simpler, more transparent, and easier to use
than process models.
• Therefore, there is less risk of obtaining a spurious prediction from a statistical
model than from a process model.
• In contrast, process models attempt to represent all processes affecting
environmental flows and stocks, providing more flexibility in modeling different land
use scenarios.
• Limited biophysical data present a challenge for validation of both statistical and
process models, limiting the regions and management practices for which models
are suitable for predictions
• Field experiments and model parameterization/calibration are crucial and
necessary to improve understanding and predictions of environmental processes.
Editor's Notes
Empirical models have the advantage of being relatively simple to use and can be developed to use only readily available data. However, they are only applicable over the range of soil, climate and other properties considered in developing the model.
The IPCC has developed procedures for estimating changes in GHG emissions: default, or Tier I, emission factors are very general and can result in substantial errors under certain conditions. Tier II emission factors for agricultural systems, which are based on country-specific measurements and thus provide more accurate emission estimates, however, because the Tier II emission factors were developed under certain climatic conditions, they may not apply for climate change projections.
Much research published since the adoption of the present default emission factors (1.25% of N applied as
synthetic and organic fertilizers, crop residues etc, and 2% of the N deposited by grazing animals) strongly
suggests that seasonal weather fluctuations and management variables (e.g. the timing of irrigation), and crop
type in a given region, have a large impact on fluxes. This poses the question of whether an analysis of existing
data may lead to modified emission factors.
The IPCC has developed procedures for estimating changes in GHG emissions: default, or Tier I, emission factors are very general and can result in substantial errors under certain conditions. Tier II emission factors for agricultural systems, which are based on country-specific measurements and thus provide more accurate emission estimates, however, because the Tier II emission factors were developed under certain climatic conditions, they may not apply for climate change projections.
Much research published since the adoption of the present default emission factors (1.25% of N applied as
synthetic and organic fertilizers, crop residues etc, and 2% of the N deposited by grazing animals) strongly
suggests that seasonal weather fluctuations and management variables (e.g. the timing of irrigation), and crop
type in a given region, have a large impact on fluxes. This poses the question of whether an analysis of existing
data may lead to modified emission factors.
Simulation models offer the opportunity to study long- term trends in soil C changes based on mathematical representation of nutrient-cycling process in soil–plant– atmosphere systems (Paustian et al., 1997; Skjemstad et al., 2004). Moreover, predictive modelling exercises can provide significant insights into complex ecosystem dynamics such as the conversion of forest to pasture in the Brazilian Amazon (Cerri et al., 2003, 2004).
Process-based models can be used to predict the impact of various agricultural management practices on net GHG emissions by analysing the interactions between management practices, primary drivers (climate, soil type, crop type, etc.), and biogeochemical reactions.