Biochar as a carbon dioxide removal solution: An assessment of carbon stability and carbon dioxide removal potential in Sweden

IN

DEGREE PROJECT ENERGY AND ENVIRONMENT, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2020,

Biochar as a carbon dioxide removal solution

An assessment of carbon stability and carbon dioxide removal potential in Sweden

ALESSANDRO CORBO

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ARCHITECTURE AND THE BUILT ENVIRONMENT

i

Biochar as a carbon dioxide removal solution

An assessment of carbon stability and carbon dioxide removal potential in Sweden

ALESSANDRO CORBO

Supervisors

KÅRE GUSTAFSSON JESPER AHRENFELDT Co-supervisor

GIULIA RAVENNI

Nordic Master in Environmental Engineering – Environmental Management Technical University of Denmark

2800 Kgs. Lyngby, Denmark

KTH Royal Institute of Technology

School of Architecture and Built Environment

Department of Sustainable Development, Environmental Science and Engineering SE-100 44 Stockholm, Sweden

ii

Abstract

Biochar is increasingly gaining momentum in the context of climate change mitigation and its production in Sweden could potentially become a large-scale system. Carbon stability in biochar is a crucial factor to assess its the carbon sequestration potential. Currently specific methodologies to assess biochar carbon site-specific stability are missing. This work aims at filling in part this knowledge gap assessing stability for Sweden specific soil conditions. Moreover, this work aims at assessing biomass feedstock availability for biochar production from a system perspective and aims at estimating biochar production and carbon dioxide removal potentials in Sweden. Preliminary carbon stability specific thresholds are provided for soils at 10°C

temperature and, thus, representative for Sweden conditions. Carbon dioxide removal functions are obtained for different feedstock categories (woody, herbaceous, biosolids and animal waste) dependent on pyrolysis conditions (Highest Treatment Temperature), and conditions for maximum carbon removal are assessed.

The need for future analysis in order to validate the presented results is highlighted. Future work should focus on collecting new experimental results of biochar mineralisation based on the requirements presented in this work. An opportunity mapping for biochar production system is provided, focusing on some aspects of the interaction of the former with existing systems (agricultural, energy production and waste management).

From the results of the opportunity mapping, an inventory of the available feedstock for biochar production is presented including woody residues, sewage sludge, manure, garden waste and straw. From the available feedstocks, biochar production and carbon dioxide removal potentials are estimated to range respectively between 0.9 and 1.7 million tbiochar/year and between 2 and 4.2 million tons CO2 sequestered per year (in a 100 years perspective). In terms of carbon dioxide removal potential, biochar production can significantly contribute to the goals set by Sweden in terms of climate change mitigation and emission offsetting for 2030 and 2045, potentially covering all the measures needed from carbon sinks from forest and land. It was found that the most significant contribution derives from the availability of woody residues in Sweden, whose analysis should be prioritised for future assessment of feasibility of biochar large scale production.

Keywords

Climate change mitigation, biomass, pyrolysis, soil

iii

Acknowledgments

I would like to thank my supervisors Kåre Gustafsson, Jesper Ahrenfeldt and Giulia Ravenni for their support and help throughout this thesis. I also want to thank Elias Sebastian Azzi, Cecilia Sundberg and Helena Söderqvist for the help, the valuable feedback and for the interest shown in this work.

I want to thank Stockholm Exergi for the trust given to perform this study. My gratitude goes also to my parents and my sister who have constantly shown their loving support in many uncountable ways.

iv

Table of contents

ABSTRACT ... II ACKNOWLEDGMENTS ... III ABBREVIATIONS ... V

INTRODUCTION ... 1

WHAT IS BIOCHAR? ... 1

CARBON STABILITY ... 1

AIM ... 2

OBJECTIVES ... 2

LIMITATIONS ... 2

METHODS ... 3

PART 1 ... 3

PART 2 ... 7

RESULTS ... 9

CARBON STABILITY ANALYSIS AND CONFIDENCE THRESHOLDS ... 9

CARBON DIOXIDE REMOVAL FUNCTIONS ... 10

UNCERTAINTY ANALYSIS ... 11

SENSITIVITY ANALYSIS ON METHODOLOGICAL APPROACH ... 13

FEEDSTOCK INVENTORY AND POTENTIAL MAPPING ... 16

BIOCHAR AND CARBON DIOXIDE REMOVAL POTENTIALS ... 21

DISCUSSION ... 24

METHODOLOGICAL APPROACH ... 24

LOCATION SPECIFICITY ... 25

CONSIDERATIONS ABOUT UNCERTAINTY AND SENSITIVITY ANALYSIS ... 25

ADDITIONAL EXPERIMENTAL DATA ... 26

CARBON DIOXIDE REMOVAL FUNCTIONS ... 27

FEEDSTOCK INVENTORY AND OPPORTUNITY MAPPING ... 27

BIOCHAR AND CARBON DIOXIDE REMOVAL POTENTIALS ... 28

CONCLUSION ... 29

REFERENCES ...30

APPENDIX ... 33

v

Abbreviations

AW – Animal Waste feedstock BS – Biosolids feedstock HB – Herbaceous feedstock WD – Woody feedstock

HTT – Highest treatment temperature (or pyrolysis temperature) MRT – Mean Residence Time

CDR – Carbon Dioxide Removal BP – Biochar potential

CDRP – Carbon Dioxide Removal Potential

1

Introduction

Biochar has proven to be a highly versatile material with several potential applications in different sectors, including environment, agriculture and energy (Leng et al., 2019). In addition, biochar is increasingly gaining momentum in the context of climate change mitigation. As a matter of facts, it was recently included in the

“Special Report: Global Warming of 1.5°C” (SR1.5) by the Intergovernmental Panel on Climate Change (IPCC) as a promising carbon dioxide removal solution to be implemented in the Agriculture, Forest and other Land Use (AFOLU) sector (IPCC, 2018).

There are many examples showing the interest for biochar production from both the private and the public sectors, especially in Scandinavia (Nordic Biochar Network, 2018), and Sweden is a clear example of this. For instance, the Stockholm Biochar project was launched in 2017 and consists in turning garden and park waste from urban centres into biochar, producing heat for the city’s district heating system; the produced biochar is, then, collected by residents or sold to local authorities as a growing media for plants and trees in the green urban areas. The success of this project led other cities and municipalities in Sweden to explore the

possibility to replicate similar systems (Gustafsson, 2017). Another example involves Stockholm Exergi (SE), an energy utility jointly owned by Fortum and Stockholms Stad (Stockholm municipality) which provides services of district heating and cooling to the city of Stockholm and to other businesses. SE, in fact, is currently investigating the possibility of integrating biochar production among its technologies as part of the greater vision to make Stockholm’s district heating become carbon negative (Gustafsson, 2020).

Considering that in Sweden biochar production could potentially evolve to become a large scale system (Gustafsson, 2020), and considering Sweden’s commitment to climate change mitigation measures (Statens offentliga utredningar, 2020), there is the need of investigating the potential of this solution at a national scale, in order to include the former among viable carbon dioxide removal solutions.

What is biochar?

Biochar is a carbon-rich product obtained from the carbonisation of biomass heated at high temperatures under low oxygen concentration through a process known as pyrolysis. The feedstock biomass can include lignocellulosic materials (e.g. wood and wood residues), agricultural waste and organic waste in general (e.g.

animal manure and sewage sludge) (Li et al., 2019) and the pyrolysis process can vary according to the specific process conditions such as heat transfer, maximum temperature and residence time (Li et al., 2019).

Biochar’s potential for climate change mitigation mainly derives from the fact that it contains polycyclic aromatic carbon structures (Bird et al., 2015), which are formed during the pyrolysis process. In fact, these structures make the biochar more recalcitrant to mineralisation (i.e. the conversion of carbonaceous material to carbon dioxide) than the original non-pyrolysed biomass (Lehmann, 2015). Thus, converting biomass into biochar and using it as soil amendment would ensure sequestering and storing carbon for a period ranging from decades up to centuries (IPCC, 2018). In this way the sequestered and stored carbon is prevented from being released back into the atmosphere. In addition, biochar has been proven to have other positive effects in terms of climate change mitigation when used as a soil amendment. In fact, it can lead to decreased N2O and CH4 emissions from soils (Lehmann, 2015) and it can induce a reduction in native soil organic carbon mineralisation (negative priming) (Liu et al., 2018). Moreover, biochar can improve soil fertility by increasing water retention and nutrient uptake, by regulating soil pH and by enhancing cation exchange capacity (Li et al., 2019), reducing the need for mineral fertilisers.

Carbon stability

Biochar equivalents deriving from natural fires or anthropogenically produced biochar have been found in soils dating back to centuries and millennia (Lehmann, 2015), which supports the evidence that part of biochar carbon is recalcitrant to mineralisation for a time ranging from centuries to thousands of years (Fang et al., 2014). Quantifying and predicting biochar carbon stability (or recalcitrance) is crucial in order to estimate its carbon sequestration and, thus, its climate change mitigation potentials to compare the latter to other solutions, and to ensure an optimal use of the biomass resources. However, a mechanistic model which allows to predict mineralisation and persistence based on biochar material properties, environmental

conditions and decomposer dynamics, is currently missing (Lehmann, 2015). A significant part of the current knowledge and research on biochar carbon stability focuses on mineralisation experiments. These are field or laboratory trials in which biochar is incubated and either CO2 or carbon content are measured to test the persistence over time in different conditions. However, these experiments are methodologically constrained, as they constitute a short-term assessment from which long-term mineralisation parameters are to be quantified (Lehmann, 2015).

2 Currently in literature there are different examples of methodologies which provide stability estimates based on biochar material properties which are proxies of carbon stability. A widely accepted methodology

retrieved from Budai et al. (2013) provides values of the unmineralised fraction of carbon in biochar at 100 years (BC+100) based on the H/Corg (i.e. organic carbon) molar ratio of the biochar. H/Corg is an indicator for the aromatic structures in the biochar and, thus, in this case, it is chosen as the proxy indicator for stability.

Only organic C is considered because inorganic C includes carbonates (calcite and dolomite) which are not part of the condensed aromatic structure and are not expected to persist over long-time scales. This method provides stability values (BC+100) based on a statistical analysis on experimental mineralisation results, which are used as conservative estimates of carbon stability. The H/Corg methodology has been assessed as the most suitable method to assess stability for its connection to measured degradation, accessibility and acceptance (Söderqvist, 2019) and it has been adopted by the International Biochar Initiative (IBI) as their reference Stable C Protocol (Lehmann, 2015).

Another methodology is presented by the IPCC who provides a method for estimating the change in mineral soil organic carbon stocks from biochar amendments (IPCC, 2019), in which the factor Fperm represents the fraction of biochar carbon remaining (unmineralised) after 100 years, equivalent to the BC+100 presented by Budai et al. (2013). The IPCC provides estimated stability values (Fperm) based on different temperature of pyrolysis or Highest Treatment Temperature (HTT) ranges, obtained through a statistical analysis on experimental mineralisation results. In this case, the proxy indicator for stability is the HTT of the pyrolysis process, as HTT and the degree of aromatisation of carbon structures in biochar are positively correlated (Bird et al., 2015).

There are similarities between these two approaches: both, for instance, use a selection of specific

mineralisation experiments to perform a statistical analysis and obtain stability estimates. Moreover, it is highlighted that both methodologies are only based on biochar material properties and thus, can be said to only provide generic results. A drawback of both methods is that they are lacking a differentiation for biochar stability dependent on location or site-specific conditions (e.g. environmental conditions). Since the latter have a significant influence on carbon stability (Lehmann, 2015), they necessarily need to be considered when developing methods to estimate carbon stability. Future research should, therefore, focus on

developing location-specific methodologies for biochar carbon stability (Budai, 2020), to be representative especially for places or countries such as Sweden which are likely to produce biochar and use it in soils.

Aim

The aim of this work is dual: the first goal is to assess biochar carbon stability and to develop preliminary estimates for biochar carbon stability in location-specific (Sweden) soil conditions. Moreover, this work aims at assessing the potentials for biochar production and carbon sequestration in Sweden.

Objectives

The research objectives of this work are:

- To obtain preliminary confidence thresholds of carbon stability for biochar for Sweden specific soil conditions

- To assess appropriate conditions for maximum carbon sequestration, considering pyrolysis parameters, carbon stability and biomass feedstock

- To assess potential feedstock availability for biochar production in Sweden in the present conditions - To quantify biochar production and carbon sequestration potentials for selected available feedstocks

in the present conditions Limitations

The focus of this work is the use of biochar in soil and thus, literature review and data collection were focused around this specific use. Results and conclusions are to be related to this specific use of biochar as all other uses were disregarded in the stability analysis. More specifically, the use of biochar in soil is to be intended as limited to agricultural sector. Additionally, the concept of location-specificity used in this work is limited to Sweden.

3

Methods

The methods used in this study are comprised of two parts: the first part (see section Part 1) describes the procedure used to obtain biochar carbon stability thresholds and closely follows the methodology provided by the IPCC (2019) described in the introduction. The followed steps in this part consisted of:

- Preliminary operations on the dataset (processing, cleaning, and selection)

- Calculation of necessary data (BC+100) from mineralisation experimental results retrieved from the dataset

- Data preparation (division in classes) and statistical analysis to obtain a 95% confidence interval (i.e.

the chosen stability thresholds) - Uncertainty and sensitivity analyses

The second part (see section Part 2) focused on feedstock availability, biochar production and carbon sequestration potentials. The followed steps in this part consisted of:

- Development of Carbon Dioxide Removal (CDR) functions

- Inventory of specific feedstocks and opportunity mapping for biochar production - Estimation of biochar production and carbon dioxide removal potentials (BP and CDRP) Part 1

Dataset

A comprehensive set of biochar mineralisation experiments in soil was retrieved by Lehmann (2015) which includes results from various biochar samples (feedstock, production process), different experimental settings (type of study, temperature of mineralisation, period of assessment, soil type) and different methods of calculation of mineralisation. This dataset was chosen for the analysis mainly for the following reasons:

- It is the most comprehensive collection of biochar mineralisation experiments including experiments performed until 2015.

- It includes additional information about pyrolysis process and experimental conditions (e.g. Highest Treatment Temperature, residence time, feedstock, soil types).

- It presents mineralisation data, namely the Mean Residence Time (MRT), adjusted to 10°C soil temperature.

The last point, in particular, is utterly significant for the purpose of this work. In fact, using data adjusted to that specific soil temperature allows to consider location specificity for Sweden in the stability analysis, which is part of the aim of this work. More on the concept and methods behind temperature adjustment can be found in later sections of this work (see Temperature adjustment). The choice of a reference soil temperature of 10°C was justified by an assessment on environmental conditions of soils in Sweden. In fact, most of Swedish soils are classified as Cryic (USDA-NRCS, 1997) according to the soil temperature regime

classification, with a mean annual temperature lower than 8°C (Rossiter, n.d.) and only a smaller fraction of soils is classified as either Mesic or Pergelic (USDA-NRCS, 1997), with temperatures ranges 8-15°C and <

0°C, respectively (USDA, 2015). Thus, it was assumed that such reference temperature (10°C) could relatively well represent most of Swedish soils. This assumption seems to be valid even considering results from

models predicting soil temperature increase due to climate change provided by Jungqvist et al. (2014).

Data processing, cleaning and selection

As the initial dataset was retrieved in a format that could not be directly analysed, it was necessary to process it in order for it to be suitable for data and statistical analyses. Then, a process of data cleaning was

performed comparing each processed data with the original source, in order to avoid potential errors.

The criteria for selection of suitable data for the purpose of the study were based on the criteria presented in the analyses performed by Budai et al. (2013), Lehmann (2015) and the IPCC (2019). In fact, these three analyses select their data to perform an analysis on biochar carbon stability based on duration of the

experiment and on the method used to obtain mineralisation parameters. The criteria used in this work were the following:

- Experimental duration greater than 1 year

Experimental duration significantly affects the estimated biochar mineralisation parameters, as Wang et al. (2016) concluded from their meta-analysis on 128 experiments, which showed a decreasing pattern of experimental biochar decomposition rates by increasing experimental

4 duration. This result is consistent with basic theory and experimental evidence, as the decomposition during the initial period (first year) of incubation is more likely to affect the pyrolysis co-products of the char (Kuzyakov, 2019). Thus, experiments shorter than one year generally underestimate carbon stability in biochar, as the measured mineralisation does not reflect the behaviour of the more stable carbon structures in the biochar. This justifies the criterion of experimental duration greater than one year used for the selection of data.

- Double or triple exponential model fitting

During biochar incubation experiments, either CO2 fluxes or the remaining carbon in soil are measured through isotope labelling which allow to estimate the evolution of biochar carbon over time. Then, exponential decay models are generally used to fit a mineralisation curve from the measurements to obtain mineralisation parameters such as the Mean Residence Time (MRT) or the decay rate (k) (Leng et al., 2019). Single exponential models assume a single pool of carbon in biochar with a single decay rate (Leng et al., 2019), whereas multiple exponential decay models (double, triple or infinite) assume that biochar is composed of different pools of carbon, with

different decay rates (Leng et al., 2019). Experiments which used double or triple exponential models to fit the mineralisation curve were selected for this work, and mineralisation data obtained with single exponential model were disregarded. This is because it is proven that double or triple

exponential curves provide a better fit than single exponential ones (Kuzyakov, 2019), and the latter ones do not represent well the real mineralisation behaviour especially on the long term. Infinite pool models provide adequate results but they are rare due to the high requirements of experimental measurements (Kuzyakov, 2019, Leng et al., 2019) and were not available in the original dataset.

Selected data are visible in the Appendix (Table 13).

Calculation of BC+100

The BC+100 (or BC100) represents the carbon remaining in biochar after 100 years in soil (Budai et al., 2013) which corresponds to the Fpermp factor presented in the IPCC (2019). The equations to calculate BC+100 for a generic multiple exponential decay model are showed as an example, with an explanation of each parameter:

BC(t) = ∑ 𝑐_𝑖 𝑒^{−𝑘𝑖 𝑡}

𝑖

= ∑ 𝑐_𝑖 𝑒^{−1/𝑀𝑅𝑇𝑖 𝑡}

𝑖

(1)

𝐵𝐶+100= ∑ 𝑐𝑖 𝑒^{−𝑘𝑖 100}

𝑖

= ∑ 𝑐𝑖 𝑒^{−1/𝑀𝑅𝑇𝑖 100}

𝑖

(2)

BC(t) [Cremaining/Ctotinitial]= fraction of remaining carbon in biochar at time t

ci [gCi/gCtotinitial] = fraction of carbon in biochar belonging to pool i (carbon pool size)

ki [1/year]= decay rate of carbon pool i = 1/MRTi , where MRT is the Mean Residence Time [years]

t [years]= time

Eq. 1 shows how to calculate the unmineralised carbon at any given time t, whereas Eq. 2 is specific for unmineralised carbon at time t = 100 years. In order to calculate the unmineralised biochar carbon at 100 years, as many ci and MRTi are needed as the pools assumed in the fitting. For instance, for a double exponential fitting of the original experiment, c1, c2 and MRT1, MRT2 values are needed. However, Lehmann (2015) only reports one mineralisation parameter of the original experiment, defined as the “persistence as reported” (see Appendix - Table 13). Considering the order of magnitude, it was assumed in this work that the reported parameter is the MRT of the most stable pool of biochar carbon. For the same reason, it was also assumed for the purpose of this work that the MRT adjusted at 10°C corresponded to the MRT of the most stable pool adjusted to that temperature. In order to calculate the BC+100 for the whole dataset, a two-pool exponential decay (Eq.3) was used in this work, considering a labile and a stable pool of carbon.

5 𝐵𝐶+100= 𝑐1 𝑒^{−1/𝑀𝑅𝑇1 100}+ 𝑐2 𝑒^{−1/𝑀𝑅𝑇2 100} (3)

In Eq. 3 c1, MRT1 and c2, MRT2 are the parameters related to the labile pool and the stable pool of biochar carbon, respectively. As mentioned previously the MRT adjusted to 10°C shown from Lehmann (2015) was assumed in this study to be the MRT2. Bird et al. (2015) show the relationship between the pyrolysis temperature (HTT) and the size of three carbon pools (labile, semilabile and stable) in different biochar samples, providing empirical fitting curves. These curves were retrieved using a web plot digitiser tool (Rohatgi, 2019). According to the definition provided by Bird et al. (2015), the labile pool is composed of anhydrosugars and methoxylated phenols; the semilabile one is composed of polycyclic aromatic compounds with ring number lower than 7; stable pool is composed of polycyclic aromatic compounds with ring number higher than 7. This presented model was simplified for the purpose of this work, assuming no distinction between semilabile and stable pools. The reason behind this choice is that the distinction between semilabile and stable pool suggested by Bird et al. (2015) is only based on the number of the polycyclic rings and the related expected difference in minerasability by microbes in soils. In the dataset not all experiments were performed with microbial inoculant and thus, considering a third semilabile pool, highly degradable by microbes for all the experiments would have led to unrealistic assumptions about the experimental mineralisation of the original sample. In addition, the majority of experiments used a double exponential fitting model, thus, assuming three pools would have meant assuming additional mineralisation parameters adding potential uncertainty to the results. Assuming only labile and stable pool in this work, was considered also, more consistent with the approach and results of pool sizes obtained by Wang et al. (2016). In this way, c2 could be estimated using the empirical curve showed in the Appendix (Figure 12), whereas c1 was obtained as the remaining fraction of carbon (c1 = 1-c2). The value of MRT1 was assumed to be 108 days (roughly 0.3 years) consistently with the findings in the meta-analysis performed by Wang et al. (2016).

Division in pyrolysis temperature classes

The dataset was divided in four classes, according to the pyrolysis temperature or highest treatment

temperature (HTT). The choice of classes was done in order to qualitatively separate biochar categories based on the expected carbon structures as presented in the analysis Bird et al. (2015). The different classes are presented in Table 1 and are visually represented in the Appendix (Figure 14).

Table 1 – HTT classes

HTT range (°C) Expected carbon

structure (Bird et al., 2015)

<350 Labile– Semilabile

[350, 450) Semilabile

[450, 600) Semilabile – Stable

>= 600 Stable

The expected carbon structures mentioned in Table 1 are expressed as the dominant carbon pools and correspond to a qualitatively differentiation of expected stability. In fact, increasing temperatures lead to a higher aromatisation and, consequently, to potentially higher stability of biochar. The first class (< 350 °C) is dominated by semilabile and labile pools, and, thus, biochar in this class can be expected to be the least stable. The following classes show respectively as dominant pools semilabile, semilabile-stable and stable, with the last class having the highest expected stability.

95% confidence interval calculation

The 95% confidence interval was calculated as it allows to have more information about the precision of an estimate, which in this case is the mean of the BC+100 for each of the temperature classes. The choice of using HTT as the proxy indicator in this work influenced the statistical analysis on the data: in fact, the non- linear relationship between HTT and BC+100 (IPCC, 2019) did not allow to do use a linear regression to obtain the confidence interval. It was, therefore, decided to perform a bootstrap analysis on different classes of samples to obtain the confidence intervals, instead of a linear or polynomial regression. The bootstrap is a widely applicable statistical tool belonging to the simulation methods (Navidi, 2011) which was also used by the IPCC for its stability analysis (IPCC, 2019). When a sample is drawn from a non-normal distribution,

6 confidence intervals can be obtained through bootstrap methods. As the probability distribution of the sample of BC+100 is not known, a non-parametric bootstrap was used. The underlying assumption was that each class represents a statistical population of BC+100 with an unknown probability distribution function and the experiments in each class are a sample of the population.

These steps were followed to obtain the 95% confidence interval of the mean BC+100 for each of the temperature categories mentioned previously (Navidi, 2011):

- Drawing bootstrap samples (10000) from the data, with the same size of original sample and drawn with replacement.

- Computing the sample mean for each bootstrap sample

- Computing the 2.5% and 97.5% percentiles (x.025, x.975) which denote the lower and upper values of the 95% confidence interval.

Similarly to what shown by Budai et al. (2013) the x.025 was be taken as the stability threshold for the specific HTT class. The choice of the lower value of the 95% confidence interval as the value for carbon stability is justified by its statistical meaning: in fact, being X.025 the lower value of the interval, there is 2.5% confidence that the true mean of the BC+100 will be lower than X.025 and a 97.5% confidence that it will be higher than X.025. This same approach can be considered as a conservative approach to estimate carbon stability, based on the available experimental data.

Uncertainty analysis

A qualitative uncertainty analysis was performed focusing on the temperature adjustment and on the methodological assumption related to the carbon pools (size, number and mineralisation parameters) and BC+100 calculation. The analysis consisted in classifying uncertainty according to three dimensions as defined by Warmink et al. (2010):

“(1) the location, which is where the uncertainty manifests itself [...], (2) the level, which is where the uncertainty manifests itself along the (continuous) spectrum between deterministic knowledge and total ignorance, and (3) the nature of the uncertainty.[…] five possible locations of uncertainty (are): (a) context uncertainty, (b) input uncertainty, (c) model uncertainty, which consists of model structure uncertainty and model technical uncertainty, (d) parameter uncertainty, and (e) uncertainty in the model outcomes. The levels of uncertainty range from statistical uncertainty and scenario uncertainty through recognized

ignorance to total ignorance. For the last dimension, they distinguish between epistemic uncertainty (due to a lack of knowledge) and variability uncertainty (due to the variability in the behaviour of the natural, social, economic or technical system)”.

Sensitivity analysis on methodological approach

A sensitivity analysis was performed in order to assess the sensitivity of the results to the methodological approach, focusing on temperature adjustment and the choice of the model used to estimate BC+100 (single, double or triple exponential pool and mineralisation rates of pools). Additional bootstrap analyses were performed on the dataset using different approaches and were compared to the reference method (REF) used in this work.

In the first approach (A1), the same HTT classification as the reference was used, and BC+100 values were estimated through a double exponential decay model. The sizes of the labile and stable pools were obtained according to the HTT, in the same way as described for the reference approach. The MRT of the labile pool was set to 108 days, as for the reference, whereas the ‘Persistence as reported’ value was taken as the MRT of the stable pool. As presented previously, this factor represents the MRT obtained in the original experiment, related to the original temperature of mineralisation and not adjusted to 10°C. Therefore, this approach allowed to assess the sensitivity of the results to the adjustment of the mineralisation temperature.

In the second approach (A2), the HTT classes were kept as the reference. However, to estimate the BC+100 values, a single exponential decay was used, assuming that 100% of the biochar carbon belonged to the stable pool. Thus, only a single MRT was needed and it was chosen to use the values adjusted to 10°C. In A3 and A4, a triple exponential decay was used including the semilabile pool which was disregarded and merged with the stabile pool for the reference analysis. The MRT of the labile pool was set to 108 days, and for the stable pool the adjusted MRT was used, as for the reference. The semilabile MRT was set equal to 10 years and 90 years, respectively for A3 and A4, to assess the sensitivity to different mineralisation rates of the semilable pool.

7 The sizes of the pools were determined using the empirical relation shown by Bird et al. (2015) (see Appendix - Figure 13).

An additional approach was assessed using a different division of classes and a bootstrap analysis was performed on each HTT with at least one experimental data. This allowed to assess the sensitivity to the choice of classes.

Part 2

CDR functions

The purpose of biochar production can be different as presented by Lehmann (2015): in this work, the focus was to assess potential production in the context of carbon sequestration for climate change mitigation and therefore, carbon dioxide removal (CDR) functions were developed. These functions represent the equivalent amount of CO2 (mass) that is originally sequestered by the feedstock biomass and stored in the form of biochar organic carbon in a 100 years’ timeframe per unit mass of feedstock biomass. The functions include a yield factor to account for the mass of biochar produced from the feedstock undergoing the pyrolysis process, a factor that represents the amount of initial organic carbon content of biochar (FC), a factor to account for the estimated mineralisation of organic biochar carbon (BC100) and a factor to convert mass of organic carbon to the equivalent mass of CO2 (CF).

𝐶𝐷𝑅 = 𝑌𝑖𝑒𝑙𝑑 ∙ 𝐹𝐶 ∙ 𝐵𝐶100 ∙ 𝐶𝐹 (4)

CDR [gCO2/gBiomass] = carbon dioxide removal in a 100 years’ timeframe Yield [gBiochar/gBiomass] = biochar yield from pyrolysis process

FC [gC(0)/gBiochar] = initial organic carbon fraction of biochar

BC100 [gC(100)/gC(0)] = fraction of initial carbon remaining at time t = 100 years CF (Conversion Factor) [gCO2/gC] = 44/12

The BC100 is the stability stepwise function obtained in the first part of this work (lower limits of the 95%

confidence intervals for each HTT class), and it provides the unmineralised carbon in biochar after 100 years as a function of pyrolysis temperature. The yield functions were obtained from the meta-analysis performed by Li et al. (2019) on 154 different independent studies. In their analysis Li et al. (2019) show that pyrolysis temperature (HTT) affects biochar yield and the carbon content biochar (FC) and empirical fitting functions for these factors based on the HTT were provided. These functions were retrieved from the source using a web plot digitiser tool (Rohatgi, 2019), obtaining stepwise values of yield and FC. These values were obtained for HTT ranging from 250 to 650°C, consistently with the range of HTT available from the dataset and used to assess the stability function BC+100. In this step the following assumptions were made: firstly, the FC factor only included organic carbon, disregarding inorganic carbon fraction (e.g. carbonates). Secondly, the yield was assumed to be expressed in dry matter (kg biochar dry/kg feedstock dry).

Thus, the final CDR function resulted in:

𝐶𝐷𝑅(𝐻𝑇𝑇) = 𝑌𝑖𝑒𝑙𝑑(𝐻𝑇𝑇) ∙ 𝐹𝐶(𝐻𝑇𝑇) ∙ 𝐵𝐶100(𝐻𝑇𝑇) ∙ 𝐶𝐹 (5)

One function was obtained for each of the biomass category defined by Li et al. (2019) (Animal Waste, Woody, Herbaceous and Biosolids), since the yield and the FC functions were specific for each biomass type.

It is pointed out that, on the other hand, BC100 function was not feedstock specific. It can be seen that the final CDR functions are HTT dependent, which allows to assess how CDR is related to the pyrolysis process.

Feedstock inventory

The feedstock inventory was limited to certain categories of residual biomass from other sectors (see Table 2) and each feedstock was assigned to a specific category as presented by Li et al. (2019). The choice of only considering residual instead of virgin biomass is motivated by the fact that, following what concluded by the analysis of Roberts et al. (2010), the former ensures higher climate change mitigation benefits and economic viability in a life-cycle perspective. The specific categories and sectors were chosen as they were expected to have already existing flows of residual organic matter belonging to the categories defined by Li et al. (2019).

8 Table 2 - Biomass categories and production sectors

Biomass category Categorisation* (Li et al.,

2019) Production sector

Wet residues (manure) AW Agriculture

Woody residues WD Forest

Sludge BS Wastewater treatment

Garden waste HB Waste management

Straw HB Agriculture

*AW = Animal waste, WD = Woody, BS = Biosolids, HB = Herbaceous.

For each biomass category it was necessary to find the production at national level for Sweden: this was done using primarily web database and literature research where official sources related to each sector were preferred. However, whenever data could not be found from official sources, data from peer-reviewed literature was used using keywords such as: agricultural residues, forestry residues, waste management, straw, manure, slash, stumps, sewage sludge, garden waste.

Preliminary opportunity mapping

In order to assess the realistic availability of each feedstock for biochar production it was necessary to perform a preliminary opportunity mapping. As described by Lehmann (2015) opportunity mapping for biochar production system requires taking a system perspective in order to assess biochar biophysical and socioeconomic system fit. Lehmann (2015) categorises the overall integrated system in three phases (biomass, conversion and use phases) and for each of those specific environmental and socio-economic factors are identified. Moreover, the interaction with the other systems is highlighted in the different phases.

In the specific case of this work the analysis was limited to certain aspects taken from Lehmann (2015) which are summarised as follows:

- Biomass phase:

o Systems: waste, farming (agriculture) and forestry.

o Environmental and socio-economic factors: feedstock production, price or economic value of feedstock, transport & storage of feedstock, current uses of feedstock.

- Conversion phase (i.e. pyrolysis process):

o Environmental and socio-economic factors: suitability of feedstock.

- Use phase (of produced biochar):

o Systems: climate (C sequestration), farming (agriculture)

o Environmental and socio-economic factors: regulations and standards.

The needed information to assess each aspect of the considered system were retrieved through an extensive literature review. Missing information was retrieved through experts’ advice and experience. In addition, each of the mentioned factors together with other feedstock specific factors were categorised as being either a constraint or an opportunity to the implementation of a biochar production system.

For this part of the work all retrieved information were assumed to be representative for present and near future conditions of the system: for instance, present feedstock production was considered and future changes or trends were not addressed; the same can be said for all the assessed components of the system.

The final assessment on the availability of feedstock was the result of a qualitative assessment on the integrated systems.

Estimation of biochar production and carbon dioxide removal potentials

Biochar Potential (BP) was defined as the amount of biochar that can be produced from a defined amount of feedstock which is feedstock category specific and was calculated as shown in Eq. 6. The Carbon Dioxide Removal Potential (CDRP) was defined as the amount of total sequestered CO2 from a certain amount of biomass feedstock and was calculated as shown in Eq. 7. Also the CDRP is feedstock category specific.

𝐵𝑃 = 𝐹𝑒𝑒𝑑𝑠𝑡𝑜𝑐𝑘 ∙ 𝑌𝑖𝑒𝑙𝑑(𝐻𝑇𝑇) (6)

9 CDRP = 𝐵𝑃 ∙ 𝐹𝐶(𝐻𝑇𝑇) ∙ 𝐵𝐶100(𝐻𝑇𝑇) ∙ 𝐶𝐹 = 𝐹𝑒𝑒𝑑𝑠𝑡𝑜𝑐𝑘 ∙ 𝐶𝐷𝑅(𝐻𝑇𝑇) (7)

Feedstock [kt/year] = amount of yearly available feedstock BP = biochar potential [ktbiochar/year]

CDRP = carbon dioxide potential [ktCO2/year]

For the Yield(HTT) and the CDR(HTT), the same assumptions as for previous section were considered.

For this part of the work it was assumed that: the production facilities are adapted to the specific biomass feedstock; the biomass feedstocks are not mixed, assuming an ideal condition in which each feedstock is separately converted into biochar; As Feedstock, the available biomass estimated from the opportunity mapping was taken; The specific HTT used to assess yield and CDR values were chosen according to the results from the analysis on CDR functions. As presented previously the yield is assumed to be related to dry matter, thus it was necessary to consider dry matter feedstock. Moreover, in order to obtain an uncertainty range for the selected HTT the lower and upper limits of the 95% confidence intervals of the yield and FC functions were used, as retrieved from Li et al. (2019).

Results

Carbon stability analysis and confidence thresholds

Table 3 shows the results of the carbon stability analysis of the different classes, namely the BC+100 sample mean and the lower and upper 95% confidence interval limits obtained with the bootstrap methodology.

Additionally, the number of samples in each class is reported. The lower limits are highlighted in Table 3 as they represent the chosen stability reference thresholds. The same results together with the BC+100 calculated for each sample, are visually represented in Figure 1 with different colours highlighting the different HTT classes. It can be observed that the values are all increasing with the temperature classes as expected, considering that increasing temperature leads to higher stability. This is also explained considering that samples at higher temperatures show a wider range and higher values of adjusted MRT of the stable pool (see Figure 9) resulting in higher BC+100 values. There is a sharp increase (roughly 100%) for the values (mean and 95% confidence interval limits) between the first class and the second class, whereas the increase is less substantial between the second and third, and between the third and fourth classes. This can be related to the fact that, the only samples belonging to the first class show a HTT of 250°C which, as assumed, have a size of the stable pool very low (roughly 50%, see Figure 12) compared to those of the second class (roughly 80%, see Figure 12) . The sharp increase in the stability between these two classes is therefore related to the high variation in the size of stable pool.

Table 3 - Results of stability analysis with chosen stability thresholds Temperature

(HTT) classes Mean BC+100 Lower 95%

confidence limit (chosen stability thresholds)

Upper 95%

confidence limit Number of experiments

<350 0.33 0.3 0.37 6

[350, 450) 0.64 0.6 0.68 16

[450, 600) 0.75 0.72 0.79 24

>= 600 0.82 0.78 0.86 12

10 Figure 1 - Carbon stability analysis

Carbon Dioxide Removal functions

The CDR stepwise functions for each feedstock are represented graphically in Figure 2. It is clearly visible how all feedstock categories follow the same pattern up to 350°C which consists in an initial slight decrease in the amount (grams) of CO2 sequestered per gram of feedstock between 250°C and 300°C, followed by a

significant increase between 300 and 350°C. The increase in the sequestration potential between 300 and 350°C is explained by the stability threshold difference which occurs from the first to the second HTT class.

Three feedstock categories from 350 up to 650°C show a similar pattern, with an overall decreasing trend with less important relative increases when changing HTT classes (450°C and 600°C): this overall behaviour can be explained by the decrease in yield which dominates the overall CDR function (see Figure 10). On the other hand, animal waste feedstock shows, after a slight decrease between 350 and 400°C, a rising trend, which can be justified by the AW yield function, which after 400°C is relatively stable compared to the yield of other feedstocks (see Figure 10). Herbaceous feedstocks show the highest CDR at HTT lower than 550°C whereas thereafter it is Animal Waste which shows the highest potential among feedstocks. The feedstock showing the lowest potential overall is Biosolids, due to its significantly lower carbon content (FC) in the assessed HTT range (see Figure 11).

Analysing the CDR functions obtained it was possible to assess which conditions show maximum carbon sequestration: according to these results, for HB, WD and BS there seems to be an absolute maximum in CDR at 350°C, whereas AW shows an absolute maximum at 650°C. Considering classes of HTT it can be concluded that to maximise CDR for the above mentioned three feedstocks (HB, WD and BS), biochar should

11 be produced in the temperature range going from 350 to 450°C. On the other hand, biochar from AW should be produced at the highest range of the assessed temperatures.

Figure 2 - CDR functions (AW = Animal waste, WD = Woody, BS = Biosolids, HB = Herbaceous)

Uncertainty analysis Temperature adjustment

It is necessary to highlight that the 10°C adjusted MRT values were not calculated in this work, but were retrieved directly from Lehmann (2015). From the selected dataset (Appendix - Table 13), it can be noted that none of the included experiments are performed at an incubation temperature of 10°C, as all show am original mineralisation temperature above this value. It is stated from Lehmann that MRTs were adjusted, using Q10 values which are factors generally related to soil respiration and which represent the increase in soil respiration resulting from an increase in soil temperature of 10°C (Meyer et al., 2018). This same factor (Q10 ) is generally used in literature to describe the difference in biochar mineralisation with a 10°C increase in temperature (Fang et al., 2014, Lehmann, 2015, Budai et al., 2013). Lehmann (2015) states that the Q10

which were used were retrieved from an empirical function represented in Figure 3, however, it is not clear which values were used, how they were related to original incubation temperature and how the Q10 factors are related to the MRT.

12 Figure 3 - Q10 vs Temperature of mineralisation – Adapted from Lehmann (2015)

From the equations presented by Fang et al. (2014), whose work is cited by Lehmann (2015), it was attempted to find the relation between the Q10 and the MRT.

𝐵𝐶(𝑡) = 𝑐₁ 𝑒^{−𝑟(𝑇) 𝑘1 𝑡}+ 𝑐₂ 𝑒^{−𝑟(𝑇) 𝑘2 𝑡} (8)

Eq.8 represents the double exponential decay model used to quantify carbon mineralisation of biochar as presented previously including a factor r(T), introduced by Fang et al. (2014), which is a temperature adjustment function, relating the mineralisation rates (k1 and k2) at a certain temperature (T2) to the rates at a reference temperature (T1). The r(T), according to Fang et al. (2014) is estimated using a non-linear least- squares curve fitting on experimental data of mineralisation at the different temperatures (T1 and T2). It is assumed by Fang that both recalcitrant and labile pools show the same sensitivity to temperature changes in terms of mineralisation (Fang et al., 2014). Eq.9 is taken directly from Fang et al. (2014) and shows the relationship between T2, T1 and Q10.

𝑄10(𝑎𝑣𝑒𝑟𝑎𝑔𝑒𝑑) = 𝑟(𝑇)(10/(𝑇2−𝑇1)) (9)

Considering, then, only the recalcitrant pool, it was assumed in this work that:

𝑀𝑅𝑇(𝑇1) = 1/𝑘 (10)

𝑀𝑅𝑇(𝑇2) = 1/𝑘𝑟(𝑇)

(11)

𝑀𝑅𝑇(𝑇1)/𝑀𝑅𝑇(𝑇2) = 𝑟(𝑇) = 𝑄10(𝑎𝑣𝑒𝑟𝑎𝑔𝑒𝑑)((𝑇2−𝑇1)/10)= 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (12)

Thus, the ratio between the MRT at a given temperature and the MRT at the reference temperature should be constant and be related to Q10 as presented in Eq. 12. The ratio between MRT (10°C) and the original MRTs, was calculated for each sample in the original dataset, before the selection for the stability analysis: it was assessed that the ratio is constant for experiments performed at the same incubation temperature as presented in Table 4 and in the Appendix (Figure 16, Figure 17, Figure 18, Figure 19).

13 Table 4 – Average value of the ratios between adjusted MRT at 10°C and original MRTs according to original temperature of mineralisation

T(°C) MRT(10°C)/MRT(T)

19 1.92

20 1.79

21 1.91

25 2.03

27 2.09

30 2.16

32 2.25

Following Eq.12 assuming, for instance, T1 = 10°C and T2 = 20°C, Q10 should be equal to the ratio between the MRT at 10°C and that at 20°C. However, this value does not match with the Q10 obtained substituting 10°C in the Equation presented by Lehmann (2015) in Figure 3. This approach was tested on the other mineralisation temperatures unsuccessfully. Thus, it was assessed that there is a relationship between adjusted and the MRTs at the original mineralisation temperature, however, the theoretical rationale and the procedure used are not clear from the source and should be investigated further.

The lack of knowledge about the validity of input data is considered as a source of uncertainty for this

analysis. The location of uncertainty was identified as input data; the level is recognised ignorance since it is related to the relations and mechanisms being studied (in this case biochar temperature sensitivity and models used to assess it); the nature is lack of knowledge (or epistemic) which could be filled with more clarity and transparency in how calculations were made.

Size, number and mineralisation rates of carbon pools

An additional source of uncertainty was the choice of methodology used to estimate the BC+100: in fact, the input data information is not complete, and it was necessary to take different methodological assumptions to proceed in the analysis, namely the number, size and mineralisation parameters of each carbon pool. The location of the uncertainty was classified as input data, as missing information in the latter led to different methodological assumptions; the level is recognised ignorance as it is uncertainty about the relations and mechanisms (fitting method and the results of the fitting of the mineralisation curve). The nature is lack of knowledge (or epistemic) which could be filled with more transparency and completeness in input data.

Sensitivity analysis on methodological approach

The results of this analysis are represented in Figure 4 and Figure 5 and presented in Table 5, where the 95%

confidence interval lower limit of the two assessed approaches are compared to the reference. In Table 5 the two highest relative changes compared to REF for each class are highlighted.

14 Figure 4 – A1 vs. Reference

Figure 5 – A2, A3, A4 vs Reference

15 Table 5 - Results of sensitivity analysis - Lower 95% confidence limit of BC+100 mean value (Lower limit), % change compared to reference (% change)

Results show overall that A1 led to lower results for all the classes which was expectable, since not adjusted data present all a lower BC+100 compared to the adjusted counterparts because of the methodological approach used for the adjustment (see section Temperature adjustment). The higher results were expectable for A2 as considering the entire carbon content to be stable means disregarding that part of the carbon (the labile or semilabile fractions) will mineralise faster over time. This is clearly visible when comparing the limit of the first class (HTT < 350), where A2 result is approximately two times higher than REF (+106% relative change): such significant difference derives from the fact that the estimated labile pool in that class ranges from 50 to 20% of the total carbon. It is also noted that the percentage change of A2 compared to REF is decreasing while temperature increases, which derives from the fact that the size of the labile pool decreases with temperature. Similarly, overall lower stability results were expected for A3 and A4 since in those approaches part of the stabile pool used in A2 is assumed to be semilabile with a much lower MRT compared to that of the stable pool, which results in less carbon unmineralised in a 100 years’ time frame.

It is noted that for the first two classes the results are primarily sensitive to the different methodological choices of pool numbers and mineralisability. On the other hand, the last two classes are mostly sensitive to methodological approach of pool number and mineralisability as well as temperature adjustment. Moreover, for all approaches it is visible that the relative changes are decreasing proceeding to upper HTT classes and the upper class shows the least sensitivity to the different approaches.

Regarding the last approach (see Appendix - Figure 15), it is noted that results are significantly dependent on the choice of classes and especially on the number of samples in each class. In fact, classes with less samples tend to show 95% confidence limits which are closer to the extreme values of the samples, which is

reasonable considering that the statistical tool is based on the process of resampling from existing data. This approach led to a higher variability in the lower 95% confidence limits, which makes the results considerably sensitive to number of samples and choices of classes. This highlights the influence on the available

experimental data on such statistical analyses.

Classes REF A1 A2 A3 A4

Lower

limit Lower

limit % change Lower

limit % change Lower

limit % change Lower

limit % change

<350 0.3 0.17 -43.33 0.62 106.67 0.17 -43.33 0.02 -92.67 [350,

450) 0.6 0.41 -31.67 0.7 16.67 0.35 -41.67 0.13 -78.33

[450,

600) 0.72 0.57 -20.83 0.78 8.33 0.6 -16.67 0.49 -31.94

>= 600 0.78 0.64 -17.95 0.83 6.41 0.73 -6.41 0.7 -10.26