The research team, led by Augusto Cesar Laviola de Oliveira and Natalia dos Santos Renato, built their analysis around Thompson’s model, a semi-empirical formulation dating to 1972 that describes how grain dries in thick layers under low airflow and near-ambient temperatures. The model divides a silo’s grain column into thin virtual layers and, hour by hour, balances the energy and moisture exchanged between the drying air and the beans. Heat exchange is treated as adiabatic, and equilibrium moisture is calculated with the modified Halsey equation, while psychrometric properties of the air follow standards from the American Society of Agricultural Engineers. The algorithm iterates until the relative humidity of the air and grain converge within a tolerance of half a percent, using a secant-method approach that the authors found robust across the vast majority of conditions.

To ground the simulation in reality, the team adopted a standardized storage silo, about 18 meters in diameter and nearly 13 meters tall, holding 2,675 metric tons of soybeans at an initial moisture content of 20 percent wet basis. Drying was considered complete when the topmost layer reached the target of 13 percent, the level needed to prevent fungal contamination during storage. Real weather drove every simulation: dry bulb temperature, relative humidity, and atmospheric pressure recorded at INMET automatic weather stations across four harvests, from 2012/13 to 2021/22. After preprocessing and convergence checks, the database contained 55,798 valid simulation samples spanning the Brazilian territory from January through August of each crop year.

The results reveal how dramatically climate shapes the energy bill. Electricity demand, which powers the fans that push air through the grain, proved relatively stable, ranging from about 72,800 to 97,900 kilowatt-hours for a standard silo load depending on the day simulated. Thermal energy, supplied by burning liquefied petroleum gas when ambient air is too humid to finish the job, swung far more wildly, from roughly 46,000 to 156,000 kilowatt-hours. On the cheapest drying days the air was warm and dry; on the most expensive, cold and saturated. Multiple regression confirmed that every variable mattered: rising temperature cut both electrical and thermal demand, rising humidity raised them, and atmospheric pressure pushed electricity up while nudging gas consumption down, all with p-values near zero.

Geography compounds the climate effect. Using inverse-distance-weighted interpolation in QGIS, the team converted point simulations into continuous national maps and extracted statistics for all 27 states. The Midwest, home to Mato Grosso, Brazil’s largest soybean producer, accounted for 42 to 44 percent of national demand, with Mato Grosso alone requiring between 1.3 and 2.7 terawatt-hours per harvest. The South followed closely, where winter humidity pushes demand to its national peak: Rio Grande do Sul recorded the highest specific demand, up to 130 kilowatt-hours per ton of grain. In the Northeast, by contrast, hot dry air does much of the work for free, with Piauí posting the lowest specific demand at under 54 kilowatt-hours per ton.

Statistical analysis of variance added a temporal dimension. For most states, energy demand fluctuated significantly from month to month but remained stable across harvests, suggesting that a single average year can represent long-term planning. The South showed a steady climb in demand from January to August as autumn and winter arrived, while states in the Center-West and North trended downward over the same period. Only Acre, Amazonas, and Rio Grande do Norte showed significant differences between harvests, hinting at shifting climate patterns in those regions.

The study then asked whether the crop could pay its own energy bill. Soybean hulls, about 5 percent of grain mass and normally sold as animal feed, carry a calorific value that, when briquetted and burned in a Rankine cycle at 37 percent efficiency, can generate roughly 1.18 megawatt-hours of electricity per ton. Nationally, that generation could technically cover the entire electrical demand of the drying sector. Coverage varies sharply by state: Piauí and Bahia could produce more than 100 percent of their needs, while Rio Grande do Sul and Paraná would fall short at 42 to 88 percent, and Mato Grosso would land between 61 and 128 percent depending on the year.

The carbon arithmetic is equally compelling. Accounting for carbon dioxide released by grain respiration during drying, by LPG combustion, and by the electricity drawn from the grid, then crediting the emissions avoided by hull-based cogeneration, the national balance ranges from a net sequestration of about 595,000 tons of carbon dioxide to net emissions of about 778,000 tons. In favorable states and years, drying soybeans could actually remove carbon from the atmosphere; in others, it remains a modest source. The economics stay manageable either way: energy costs run from about 4.37 dollars per ton in Piauí to 14.73 dollars in Acre, or between 0.99 and 3.79 percent of the soybean’s market value per 60-kilogram bag.

Perhaps the most forward-looking contribution is the artificial intelligence layer. The team trained feed-forward neural networks on the 55,798 Thompson simulations, using temperature, humidity, and pressure as inputs to predict electrical, thermal, and total energy demand. With 90 neurons in the hidden layers for the electrical and thermal models and 30 for total energy, the networks achieved coefficients of determination above 0.997, with root mean square errors as low as 0.27 percent for electricity. In independent validation, deviations from the deterministic model stayed within roughly 2.4 percent. Because the neural surrogate runs in a fraction of the time and avoids the convergence failures that occasionally plague Thompson’s formulation, it opens the door to real-time decision support, letting operators adjust fan and burner settings as weather shifts.

The authors are careful to note the framework’s boundaries: by holding silo geometry, fan efficiency, and airflow constant, they isolated climate as the driver of demand, leaving operational optimization for future sensitivity studies. But the template travels well. Any country or crop with representative weather and production data could rerun the same pipeline, from deterministic physics to geospatial statistics to machine learning. For Brazil, whose electricity matrix remains nearly 62 percent hydroelectric and increasingly exposed to climate volatility, the message is clear: the humble act of drying grain is a national-scale energy enterprise, and the residue left behind after cleaning the crop may be the cheapest, cleanest way to power it.

Subject of Research: Energy demand modeling of low-temperature soybean drying in Brazil

Article Title: Analysis and modeling of energy demand in the low-temperature drying process of soybean in Brazil

Article References: Oliveira, A. C. L. D., Oliveira, V. H. L. D., Silva, L. C. D., Moraes, C. A., & Renato, N. D. S. (2026). Analysis and modeling of energy demand in the low-temperature drying process of soybean in Brazil. Cleaner Engineering and Technology, 34, Article 101313. https://doi.org/10.1016/j.clet.2026.101313

Image Credits: AI Generated

DOI: 10.1016/j.clet.2026.101313

Keywords: soybean drying, low-temperature drying, energy demand, Thompson model, artificial neural networks, geospatial analysis, biomass energy, soybean hulls, carbon emissions, Brazil, agribusiness, energy efficiency