The core problem, according to the review, is complexity. HTC is not a single reaction but a tangled network of coupled processes, including dehydration, decarboxylation, polymerization, and recondensation, all unfolding simultaneously in an aqueous reactor. These transformations reshape the elemental composition, surface functionality, and microstructure of the starting biomass, which means that small changes in operating conditions can produce dramatically different materials. Feedstock heterogeneity compounds the challenge: two batches of the same agricultural waste can differ in moisture, ash content, and lignocellulosic composition, throwing off reproducibility between laboratories and even between runs in the same facility.

Design of experiments, a statistical discipline with roots in agricultural field trials and industrial quality control, offers a structured way to tame this complexity. Rather than varying one factor at a time, DoE frameworks such as factorial designs and response surface methodologies explore many variables simultaneously with statistically efficient experimental layouts. In the HTC context, the key variables include reaction temperature, residence time, the solid-to-liquid ratio in the reactor, system pressure, and the use of catalysts. By quantifying both the main effects of each variable and the interactions between them, DoE generates structured datasets that reveal which parameters matter most and how they combine to shape the final product.

Where DoE alone reaches its limits, machine learning picks up the baton. Classical response surface models fit relatively simple polynomial equations to experimental data, but HTC’s reaction networks are strongly nonlinear, with high-order interactions that polynomials struggle to capture. The review highlights how algorithms including artificial neural networks, ensemble methods, gradient boosting, and Bayesian approaches can learn these nonlinearities directly from data, substantially improving predictions of material-critical outputs. Among the most important targets are mass yield, the higher heating value, fixed carbon content, carbon retention, and textural and chemical descriptors that determine how well a hydrochar performs as an adsorbent or as a precursor for electrochemical materials.

The combination of the two approaches is more than the sum of its parts, the authors argue. DoE provides the statistically sound experimental backbone that ensures the training data for machine learning models actually covers the operating space of interest, while machine learning supplies the predictive muscle to capture relationships that classical models miss. This coupling proves particularly valuable for multi-objective optimization, one of the central tensions in hydrochar engineering. Energy densification, which maximizes heating value and fixed carbon, often works against the retention of oxygen-containing functional groups and the development of porosity, yet those very features determine whether a hydrochar can adsorb pollutants from water or serve as a scaffold for advanced carbon materials.

The choice of optimization target depends entirely on the intended application. Hydrochars destined for use as solid fuels benefit from conditions that push carbon content and calorific value upward, while chars intended for water and gas treatment require carefully preserved surface functionality and pore structure. A third family of applications treats hydrochar as a precursor for advanced carbon materials, where the priorities shift again. Because these objectives frequently conflict, the review emphasizes that DoE-guided machine learning allows researchers to map trade-off surfaces explicitly, identifying operating windows that balance competing demands rather than optimizing a single property in isolation.

Despite the promise, the review is candid about the bottlenecks that currently limit transferability and industrial robustness. Data quality and comparability across studies rank among the most serious. Machine learning models are only as good as the datasets they learn from, and HTC literature contains results generated under widely varying definitions, measurement protocols, and reporting standards. Pooling data across publications therefore risks training models on inconsistent inputs, a problem familiar to anyone who has tried to build predictive models from heterogeneous literature data in other fields of chemistry and materials science.

Interpretability presents a second hurdle. Deep neural networks and complex ensemble models may deliver impressive predictive accuracy, but they often behave as black boxes, offering little physical insight into why a particular combination of temperature, time, and feedstock produces a given outcome. For a field that still seeks mechanistic understanding of its reaction networks, models that cannot be interrogated risk becoming predictive oracles without explanatory power. The review suggests that hybrid strategies, in which DoE structure constrains and interprets machine learning predictions, offer a route to models that are both accurate and scientifically meaningful.

The systematic treatment of biomass variability rounds out the list of open challenges. Because feedstock composition varies with season, geography, and preprocessing, models trained on one biomass type may fail badly when applied to another. Building variability explicitly into experimental designs and model inputs, rather than treating it as noise, emerges as a methodological priority for anyone hoping to move hydrochar production from laboratory curiosity to industrial process.

Taken together, the review sketches a roadmap for moving HTC from empirical tuning toward data-informed, property-targeted engineering. By pairing statistically rigorous experimental design with machine learning capable of capturing high-order nonlinearities, researchers can begin to specify the hydrochar they need and work backward to the operating conditions that produce it. The authors outline methodological directions for developing more reliable hybrid DoE-ML strategies, with the goal of accelerating rational design of hydrochar materials at scale. If those strategies mature, the humble wet-biomass reactor could become a predictable manufacturing platform for carbon materials tailored to energy, environmental, and electrochemical applications alike.

Subject of Research: Design of experiments-guided machine learning for predicting hydrochar properties in hydrothermal carbonization

Article Title: Toward predictable hydrochar properties at scale: a critical review of design of experiments-guided machine learning in hydrothermal carbonization

Article References: El Ouadrhiri, F., & Lahkimi, A. (2026). Toward predictable hydrochar properties at scale: a critical review of design of experiments-guided machine learning in hydrothermal carbonization. Frontiers in Chemical Engineering, 8, Article 1883652. https://doi.org/10.3389/fceng.2026.1883652

Image Credits: AI Generated

DOI: 10.3389/fceng.2026.1883652

Keywords: hydrothermal carbonization, hydrochar, machine learning, design of experiments, biomass conversion, response surface methodology, neural networks, multi-objective optimization, higher heating value, adsorption, carbon materials, data-driven modeling

The research team, led by Sunyong Park and colleagues including Sunhwa Ryu, assembled a literature-derived dataset of 552 thermally treated biomass samples, spanning torrefaction and pyrolysis experiments drawn from dozens of published studies. Each record contained ultimate analysis, proximate analysis, higher heating value, and the process conditions—treatment temperature and residence time—that produced the fuel. Rather than feeding these numbers raw into an algorithm, the researchers invested heavily in preprocessing. Because both ultimate and proximate analyses are compositional data whose variables must sum to 100 percent, they applied closure normalization separately to each block and computed centered log-ratio transforms, a technique from compositional statistics that suppresses the spurious correlations that fixed-sum data otherwise generate.

Feature engineering went further. The team built process-aware descriptors—squared temperatures, logarithmic residence times, temperature–time interaction terms, and a combined severity index—to capture the nonlinear way thermal treatment intensity drives devolatilization, oxygen removal, and carbon enrichment. Composition-aware descriptors such as H/C and O/C ratios encoded the degree of carbonization, while proximate-derived quantities like the volatile-matter-to-fixed-carbon ratio captured the balance between volatile release and char formation. Critically, the authors also built in safeguards against information leakage and circular reasoning: auxiliary features were constructed only from training data, and a Dulong-equation-derived heating value surrogate was deliberately excluded from the final heating value models to ensure the algorithm was not simply copying a textbook formula hidden inside its own inputs.

After outlier removal using the interquartile range rule—which cut 234 records, or 42.4 percent of the dataset, leaving 318 cleaned samples—the team trained and compared linear, bagging, boosting, and tree-ensemble models, including Ridge regression, partial least squares, Random Forest, Extra Trees, XGBoost, and CatBoost. Each target variable got its own single-output model, selected by validation performance and tuned with grid search. To avoid the common trap of reporting a flattering single data split, the researchers re-evaluated every selected model across five repeated random train-validation-test partitions and reported the mean and standard deviation of test-set performance.

The results reveal a striking asymmetry between the two prediction directions. Predicting proximate properties and heating value from elemental composition—the ultimate-to-proximate direction—worked well. CatBoost models achieved repeated test-set R-squared values of 0.756 for volatile matter and 0.751 for higher heating value, with XGBoost reaching 0.702 for fixed carbon and Extra Trees 0.607 for ash. In plain terms, knowing what elements a treated biomass contains, along with how hot and how long it was treated, is enough to estimate its volatile content and energy density with useful accuracy. The heating value model predicted values with a mean absolute error of roughly 0.75 megajoules per kilogram—impressive for a screening tool that never lights a flame.

Going the other way proved harder. When the models tried to infer elemental composition from proximate data, performance dropped and became far more sensitive to how the data were partitioned. Carbon, hydrogen, oxygen, and heating value all landed in a moderate range, with repeated R-squared values between roughly 0.41 and 0.57, and large standard deviations betraying instability across splits. The reason is mathematical as much as chemical: proximate analysis compresses the fuel into three broad fractions, and many different elemental combinations can produce nearly identical volatile, fixed-carbon, and ash profiles. The inverse problem is simply underdetermined. Sulphur fared worst of all, with a negative mean R-squared of −0.085, meaning the model performed worse than naively guessing the dataset average. The authors are refreshingly blunt about this: sulphur, present at very low and heavily skewed concentrations, is flagged as unreliable and excluded from quantitative decision-making, and nitrogen is marked as a low-confidence screening estimate only.

The team also stress-tested their own pipeline. A filtering sensitivity analysis compared no outlier removal, gentle filtering, and the strict final setting, and found the effects were target-dependent rather than uniformly beneficial. Volatile matter and heating value barely changed across scenarios, but ash prediction improved substantially with stricter filtering—its R-squared rising from 0.44 to 0.64—because extreme ash values from heterogeneous feedstocks were dragging the model off course. Meanwhile, several inverse-direction targets actually got worse under strict filtering, suggesting that aggressive pruning had removed compositional variability the models needed. This nuanced outcome matters: it demonstrates the reported performance was not an artifact of quietly discarding nearly half the data, while honestly acknowledging that the final models cover a narrower slice of the biomass universe than the raw literature does.

The benchmark against human-devised formulas may be the study’s most compelling result. On the same held-out test set, the Park et al. empirical equations for predicting proximate properties from elements managed R-squared values of 0.72 for volatile matter and 0.40 for ash, while the machine-learning models reached 0.89 and 0.64 respectively. The gap widened dramatically for heating value: the century-old Dulong equation produced a negative R-squared of −1.01 on this thermally treated dataset, whereas the Dulong-free machine-learning model scored 0.87. In the inverse direction, the Shen et al. correlations from 2010 essentially failed on modern torrefied biomass, with carbon prediction collapsing to an R-squared of −0.54, while the machine-learning framework achieved 0.62. The lesson is that fixed-form equations derived decades ago on different feedstocks simply do not transfer to the complex, treatment-dependent fuels now entering the market, whereas flexible nonlinear models trained on relevant data do.

Interpreting the models with SHAP analysis added physical plausibility to the statistics. For heating value prediction, carbon- and oxygen-related descriptors—including closure-aware carbon terms and O/C variables—dominated the feature importance, exactly matching the established chemistry that carbon enrichment and deoxygenation raise energy density during torrefaction. Severity-weighted terms and temperature–residence-time interactions also contributed, indicating the models had genuinely learned treatment-intensity effects rather than memorizing feedstock labels. For ash, the influential features were hydrogen, nitrogen, and sulphur descriptors, which the authors interpret not as causal drivers of ash but as statistical fingerprints of feedstock type and inorganic-rich sample groups—a candid reminder that SHAP values describe model behavior, not experimental mechanism.

To make the framework usable beyond the modeling team, the researchers wrapped it in a lightweight Streamlit graphical interface that accepts either ultimate or proximate inputs plus process conditions and returns cross-predictions with heating value estimates, input validation, compositional closure checks, and warnings when users venture outside the training domain. Reliability flags are baked in: sulphur outputs are explicitly marked unusable for quantitative decisions. The authors are careful to position the tool as a screening aid, not a laboratory replacement—regulatory evaluation of low-concentration elements still demands real analysis, and the literature-derived dataset carries inherent heterogeneity in feedstock, reactor design, and analytical protocol. But for traders, plant operators, and researchers needing a fast preliminary read on whether a biomass batch is worth pursuing, the message is clear: a well-engineered machine-learning model, honest about its own uncertainty, can now deliver in seconds what once required a destructive, instrument-intensive afternoon—and it can do it more accurately than the empirical formulas the field has relied on for decades.

Subject of Research: Machine-learning cross-prediction of ultimate and proximate properties of thermally treated biomass for cleaner fuel assessment

Article Title: Machine-Learning-Based Cross-Prediction of Ultimate and Proximate Properties for Cleaner Biomass Fuel Assessment

Article References: Park, S., Lee, J., Yang, J., Kim, S., Hwang, K., & Ryu, S. (2026). Machine-Learning-Based Cross-Prediction of Ultimate and Proximate Properties for Cleaner Biomass Fuel Assessment. Cleaner Engineering and Technology, Article 101333. https://doi.org/10.1016/j.clet.2026.101333

Image Credits: AI Generated

DOI: 10.1016/j.clet.2026.101333

Keywords: biomass, machine learning, torrefaction, proximate analysis, ultimate analysis, higher heating value, fuel quality, compositional data, SHAP, XGBoost, CatBoost, cleaner energy

The motivation is both environmental and economic. In Canada, heat production, electricity generation, and transportation fuels account for roughly 73 percent of greenhouse gas emissions, and the country has pledged to cut total emissions by 40 to 45 percent by 2030 and reach net zero by 2050. Saskatchewan, one of the provinces most dependent on coal-fired electricity, is also the nation’s agricultural powerhouse: more than 40 percent of Canada’s total cropland lies within its borders, and the country produces an estimated 82.4 million metric tons of agricultural residues each year. Converting even a fraction of that stream into clean-burning fuel would simultaneously reduce waste disposal problems, curb open burning of stubble, and displace fossil fuels in power plants.

Raw agricultural residues, however, make poor fuels on their own. They are bulky, hygroscopic, oxygen-rich, and energy-dilute, with heating values in the study’s raw samples ranging only from 17.6 to 18.5 megajoules per kilogram. Their high moisture content saps combustion efficiency, and their tendency to absorb water makes long-term storage a nightmare of mold and decay. Torrefaction addresses these weaknesses directly. The process heats biomass to between 200 and 350 degrees Celsius for anywhere from half an hour to two hours, driving off moisture and volatile compounds and leaving behind a darker, drier, carbon-enriched solid sometimes called bio-coal. In this study, the team used oxidative torrefaction—running the reaction in air rather than an inert nitrogen atmosphere—a choice that dramatically lowers cost and complexity at industrial scale.

The researchers torrefied all four feedstocks at 250, 275, and 300 degrees Celsius for 60 minutes in sealed stainless-steel reactors. The transformation was striking. Elemental carbon content, which started between 44.6 and 46.9 weight percent across the raw materials, climbed to between 65.2 and 69.8 weight percent after treatment at 300 degrees Celsius. Meanwhile, oxygen and hydrogen were stripped away through dehydration, depolymerization, and decarboxylation reactions that release water, acetic acid, carbon monoxide, carbon dioxide, and light hydrocarbons. On a Van Krevelen diagram, which plots atomic hydrogen-to-carbon against oxygen-to-carbon ratios, the torrefied samples marched steadily toward the coal corner of the chart. After treatment at 300 degrees Celsius, the oxygen-to-carbon ratio fell by more than 71 percent for every feedstock, and the hydrogen-to-carbon ratio dropped by roughly half.

The heating values told an equally compelling story. Torrefaction at 250 degrees Celsius already pushed the fuels into the 23.8 to 25.4 megajoule per kilogram range, and the most severe treatment lifted hemp straw to 29.0 megajoules per kilogram—the highest value in the study. That figure places torrefied hemp straw squarely in the territory of sub-bituminous coals, meaning it could potentially be co-fired in existing coal infrastructure with minimal modification. Moisture content, meanwhile, plummeted by 35.7 to 82.4 percent depending on the feedstock, because the heat destroys the hydroxyl groups that normally form hydrogen bonds with water. The result is a hydrophobic fuel that resists reabsorbing moisture during storage, grinds more easily, and burns with less smoke and water vapor, losing less energy up the stack.

To provide a rigorous comparison, the team also ran all four residues through steam explosion at the Canadian Feed Research Centre in Saskatoon. In that process, biomass is exposed to pressurized steam at roughly 300 pounds per square inch for one minute and then subjected to explosive decompression that physically ruptures the rigid fiber structure. Steam explosion did deliver some benefits: it altered lignin structure, increased cellulose crystallinity, and improved the material’s suitability for pelletization. But its chemical impact proved modest. Heating values rose only slightly, from 17.6 to 18.5 megajoules per kilogram in raw samples to between 18.2 and 19.6 megajoules per kilogram after treatment, and elemental composition barely shifted. The contrast illuminates the fundamental difference between the two technologies: steam explosion is primarily a physical restructuring of the biomass, while torrefaction is an irreversible chemical conversion that fundamentally rewrites the fuel’s composition.

Spectroscopic analysis confirmed the chemistry. Fourier transform infrared spectra of torrefied samples showed weakened peaks associated with cellulose and hemicellulose, indicating decomposition of those polymers, and a flattened hydroxyl stretching band consistent with moisture loss. A new peak emerged in the aromatic carbon-hydrogen bending region, signaling that lignin-derived aromatic rings had recombined and formed during heating. The steam-exploded samples, by contrast, looked nearly identical to their raw counterparts in the spectra. Compositional analysis of the raw materials showed the four residues shared similar architectures—hemicellulose contents of 26 to 35 percent and comparable cellulose levels in three of the four—which suggests the findings could generalize across a wide swath of prairie biomass.

The study also confronted an awkward truth about biomass combustion: ash. X-ray fluorescence analysis revealed that potassium and sodium, the alkali metals responsible for fouling boiler heat-exchange surfaces and agglomerating fluidized beds, were abundant in most samples. Predictive indices showed that nearly all torrefied fuels carried a high probability of fouling and slagging, with raw wheat straw scoring an alkali index of 1.93 kilograms per gigajoule—well above the 0.34 threshold at which fouling becomes almost certain. The researchers note that water or acid leaching before torrefaction could strip out these troublesome alkalis, and that adjusting or doping the bed material in fluidized-bed combustors offers another mitigation route. Hemp straw emerged as the cleanest candidate, with the lowest potassium content and a low fouling probability even after treatment.

The economics of scaling up shaped the final analysis. Solid yields declined as torrefaction temperature rose, since devolatilization sacrifices mass to concentrate energy—hemp straw lost 14.6 weight percent of yield moving from 250 to 275 degrees Celsius. Energy yields at 250 degrees Celsius clustered between 62.6 and 64.9 percent but fell as temperatures climbed, creating a classic trade-off between fuel quality and fuel quantity. To find the sweet spot, the team calculated an energy-mass co-benefit index, which peaked at 275 degrees Celsius for canola straw, hemp straw, and wheat straw, and at 300 degrees Celsius for oat hulls. That optimization framework, the authors argue, gives industry a practical target rather than a simple push toward maximum severity.

Challenges remain before torrefied prairie straw feeds a power plant. Torrefied biomass is harder to pelletize than raw material, requiring binders that add cost, and industrial-scale reactors still struggle with uniform heat distribution and process control. Large-scale life-cycle assessments, product standardization, and supportive policy all lag behind the science. Yet the study’s verdict is clear: oxidative torrefaction transforms abundant, cheap agricultural waste into a hydrophobic, coal-like fuel with heating values up to 29 megajoules per kilogram, outperforming steam explosion on every major fuel metric. For a province sitting on tens of millions of tons of residue while burning coal for electricity, the path from field waste to clean power may run through a furnace set at 275 degrees Celsius.

Subject of Research: Upgrading agricultural residues into solid biofuels through oxidative torrefaction and steam explosion pretreatment

Article Title: Development of biofuels from agricultural residues via torrefaction and steam explosion

Article References: Wattan, R., Zamiri, M. A., Newkirk, R., Dalai, A. K., & Acharya, B. (2026). Development of biofuels from agricultural residues via torrefaction and steam explosion. Results in Engineering, 32, Article 113252. https://doi.org/10.1016/j.rineng.2026.113252

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113252

Keywords: torrefaction, steam explosion, biofuels, agricultural residues, lignocellulosic biomass, higher heating value, canola straw, hemp straw, wheat straw, oat hulls, fouling and slagging, renewable energy