Biochar / Jun 07 2019

Proposal:Where should we apply Biochar ?

1. Introduction

The heating of plant biomass under low-oxygen conditions generates three co-products, bio-oil, biogas, and biochar. Bio-oil can be stabilized and used as fuel oil or be further refined for various applications and biogas can be used as an energy source during the low-oxygen heating process. On the other hand, biochar application to soil has been practiced to sequester carbon and it has shown the potential to increase crop yields when it is used to improve yield-limiting soil properties.

The complex nature of biochar interactions with soils and crops as well as lack of clear understanding of the mechanics of these interactions has led to reports with conflicting interpretations, even under similar conditions (Bach et al., 2016). In addition, the large amount of missing data in the literature including inconsistent reporting of soil and biochar properties has made the prediction of crop yield response to biochar a very challenging modeling exercise.

We addressed this challenge by developing an extensive informatics workflow for processing and analyzing crop yield response data as well as a large spatial-scale modeling. We used probabilistic graphical models to study the relationships between soil and biochar variables and predict the probability and magnitude of crop yield response to biochar application. These models are inherently capable of handling missing data and accounting for uncertainty associated with observations. In this platform, we combined expert knowledge(for better defining the relationship among the variables) and data (for estimating the parameters of the model) to deal with the complexity of our problem. Probabilistic graphical models are commonly used in cases with incomplete datasets and high uncertainty which makes them a suitable candidate for the current status of biochar science.

2. Methods

2.1. Data

We built our database on top of the raw data previously collected form a meta-analysis (Crane et al., 2013). The original dataset was built upon 40 studies (published up to 2013) with 17 variables and 685 observations (Crane et al., 2013). New peer-reviewed studies from both pot and field studies were found using academic search engines (Google Scholar, Web of Science, Scopus) and the same variables were extracted and added to the database.

Soil organic carbon (SOC), sand, silt, clay content, CEC (Cation Exchange Capacity) and soil pH were extracted from all studies to explain both chemical and physical properties of soil. Biochar carbon, nitrogen, ash content, pH, carbon-to-nitrogen (C:N) ratio, highest production temperature, feedstock, and thermochemical process were also variables extracted to account for both the feedstock properties and production characteristics in assessing biochar type.

2.2. Model Development

A Bayesian Network (BN) was used for modeling the yield response to biochar applications. BN models usually are made of qualitative and quantitative components (hybrid BN). The qualitative component is a graphical model which represents how the variables are statistically dependent on each other; nodes indicate variables and arcs show dependencies (Fig 1). The quantitative component is the conditional probability distribution of a node (specified in the graphical model) on its parents . Taking into account the conditional independence assumption (Markov condition), the joint distribution over all the variables for is equal to :

p(x1,,xn)=i=1np(xipa(xi))p(x_1,\dots ,x_n)=\prod_{i=1}^{n} p(x_i|pa(x_i))

which is the product of conditional distributions defined for each variable. When new evidence is introduced, it propagates through the BN and the posterior probabilities are computed. This is called an inference and it allows for detecting the change in the probabilities of some variables given a value for other variables (Aguilera et al., 2011).

We developed a hybrid Bayesian network model (a model that includes both discrete and continuous variables) within the R environment and by using the 'Bayes Server' software. The final design of the model’s structure resulted in 84 parameters. Given that absolute yield is not readily comparable among studies, the response ratio (RR) was used as the target variable . The is defined as:

RR=ln(YieldBiocharYieldcontrol) RR=\ln \left(\frac{Yield_{Biochar}}{Yield_{control}} \right)

A positive RR indicates a positive yield response to biochar application, whereas a RR of 0 shows no change from control treatment.

We compared our BN model and the only available statistical model for explaining the heterogeneity in yield response to soil and biochar properties (Crane et al., 2013). The authors of that study used a generalized additive model (GAM) with 162 parameters to develop smooth functions mapping of independent variables to RR. In more than 250 iterations, the BN model consistently outperformed the GAM; as the BN model average Model Efficiency (EF) and Mean Absolute Difference (MAD) values were 0.23 and 0.10, respectively, compared to -1.96 and 0.18 for the GAM.

3. Outcomes

The final BN model was then projected onto all cultivated lands in the U.S. based on the 2016 Cropland Data Layer (CDL) with 30m×30m resolution using the Gridded Soil Survey Geographic (gSSURGO) soil database for the five biochar types and two application rates, i.e., 5 and 15 Mg . We developed 20 different high resolution (30m×30m) maps on the U.S scale to present the results of our modeling exercise (Fig 2).

For the first time ever, we produced high resolution maps estimating the probability of crop yield increase following biochar application as well as the expected yield increase for every single farm around the U.S (Example shown in Fig 2).

Before this study, Bach et al., (2016) reported a long list of reasons why biochar may not be economically viable and they were mainly pointing to the fact that we have little knowledge about where to apply biochar. That's why we believe, our newly developed data products could help farmers and policy makers to save hundreds of millions of dollars by carefully allocating resources to where we know biochar is more likely to improve crop yield. However, journal publications are not the most ideal media for us to communicate our outcomes with our audience. Hard-copy or even electronic versions of our data products are not remotely as helpful as an interactive map. We have been contacted by so many people with no or little background in coding/computer science demanding access to our outputs, while we have not been able to respond to these demand.

In a separate effort we tried to designed and develop a Shiny app ( which would let users to make predictions using our model and estimate the probability of yield increase following biochar application at a farm level (Fig 3). However, this effort was unsuccessful due to lack of resources (both monetary and computational) and unreliable hosting services. We believe could be an ideal place and a trustworthy service to share the results of our work with thousands of users ranging from average farmers to policy makers.

4. References

  • Aguilera P A, Fernandez A, Fernandez R, Rumi R and Salmeron A 2011 Environmental Modelling and Software 26 1376–1388
  • Bach M, Wilske B, Breuer L. Current economic obstacles to biochar use in agriculture and climate change mitigation. Carbon Management. 2016 Jul 3;7(3-4):183-90.
  • Crane-Droesch A, Abiven S, Jeffery S and Torn M S 2013 Environmental Research Letters 8.