Post-Harvest Loss Remains a Major Challenge
Perishable agricultural products can deteriorate rapidly after harvest. Changes in temperature, humidity, respiration, water loss, and tissue condition can accelerate quality decline, particularly when storage conditions are poorly controlled.
Post-harvest losses affect more than individual farmers. They can reduce production efficiency, weaken supply-chain sustainability, affect food availability, and lower the economic value of agricultural commodities.
Technological developments in smart agriculture have created new opportunities to address this problem. Sensors can continuously collect information about environmental conditions, while digital systems can process the data to support decisions.
One technology attracting increasing attention is the Digital Twin. In simple terms, a Digital Twin is a digital representation of a physical object or system that can be updated using real-world data. While Digital Twin applications have been widely explored in areas such as manufacturing and production systems, their use for predicting the quality of agricultural products during storage remains comparatively limited.
Rustan's work addresses this gap by using a predictive Digital Twin approach that does more than monitor storage conditions. The model estimates how product quality is likely to change in the following observation period.
How the Predictive Model Was Developed
Rustan used an experimental quantitative approach involving 50 units of perishable agricultural products. The samples were selected purposively according to characteristics such as size, maturity, initial physical condition, and susceptibility to post-harvest damage.
The products were stored under controlled conditions for 12 days. Observations were conducted from day 0 through day 12, producing 13 observation periods.
Five main parameters were recorded:
- Storage temperature
- Relative humidity
- Product weight
- Product color
- Product texture
The experiment generated 650 observation points from the 50 samples and 3,250 individual measurements across the five parameters.
The predictive model used historical changes in product quality to estimate subsequent changes. Rather than relying on a highly complex artificial intelligence architecture, the approach calculated average daily quality changes from the observed data.
Model performance was assessed using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), which indicate how closely predicted values match actual observations.
Model Accurately Predicted Quality Deterioration
The observations showed a clear decline in product quality during storage.
Storage temperature ranged from 24.6°C to 29.8°C, while relative humidity ranged from 68.5% to 84.7% RH.
Over the 12-day storage period:
- Weight loss increased from 0% to 18.6%.
- Discoloration increased from 0% to 42.3%.
- The damage rate increased from 0% to 28%.
The predictive model estimated an average weight-loss progression of approximately 1.55% per day.
Its predictions remained close to the actual observations. For example, on day three, the model predicted weight loss of 4.6%, compared with an actual value of 4.8%. By day six, the predicted value was 9.3%, compared with 9.7% measured in the experiment.
On day nine, the model predicted 13.9% compared with an actual 14.2%, while on day 12 it predicted 18.2%, compared with an actual 18.6%.
The model recorded an MAE of 0.32%, an RMSE of 0.36, and a reported predictive fit or accuracy of 98.3%. These results indicate that the simple trend-based approach was able to closely follow the observed pattern of quality deterioration under the tested conditions.
From Monitoring to Predictive Post-Harvest Management
The importance of the model lies in its potential to shift post-harvest management from simply observing deterioration to anticipating it.
A conventional monitoring system can tell users that temperature, humidity, or product quality has changed. A predictive Digital Twin can use accumulated data to estimate what may happen next.
For agricultural supply chains, such information could support decisions about how long products should remain in storage, when they should be distributed, and when they should be marketed.
Rustan of Universitas Sulawesi Tenggara describes the Digital Twin approach as both a digital representation of the physical product system and a predictive model. The findings suggest that even a relatively simple historical-trend model can provide useful decision support, particularly in agricultural environments where access to sophisticated digital infrastructure may be limited.
The approach could therefore contribute to smart agriculture by connecting real-world storage observations with data-based decisions. In practical terms, better predictions could help reduce unnecessary storage time, improve shelf-life estimation, and minimize economic losses caused by product deterioration.
Limitations and Future Development
The research also identifies important limitations. The experiment involved only 50 sample units and specific storage conditions, meaning the findings cannot automatically be generalized to every agricultural commodity or storage environment.
The predictive model was based primarily on historical quality trends and did not incorporate more complex biological factors, including microorganisms, chemical composition, or extreme environmental changes.
Future development could expand the number and diversity of samples, test different agricultural commodities and storage conditions, and integrate real-time sensors and Internet of Things (IoT) technologies. More adaptive artificial intelligence models and additional chemical and microbiological indicators could also improve prediction capabilities.
Author Profile
Rustan is affiliated with Universitas Sulawesi Tenggara and is the author of the 2026 article on Predictive Digital Twin Modeling for Quality Preservation of Perishable Agricultural Products during Storage. The article does not specify Rustan's academic degree or a more detailed field-of-expertise profile, so these details are not added here.
0 Komentar