As part of research efforts aimed at improving crop production efficiency, specialists from the Siberian Research Institute of Agriculture, Biologization and Digitalization SFSCA RAS are conducting comprehensive studies to assess the impact of plant biophysical parameters—specifically, leaf area and chlorophyll activity—on potato yield formation.
The research is conducted in field experiments where various potato disease control strategies are studied. Throughout the entire growing season, systematic collection of multiparameter data is carried out, including:
- recording the phenological stages of potato development (sprouts, budding, flowering, biological maturity);
- development and prevalence of major potato diseases;
- morphometric indicators of plants;
- yield measurement, including the number and weight of tubers from each experimental plot;
- monitoring of meteorological conditions (daily air temperature, precipitation, sum of active temperatures).
Particular attention is paid to the application of modern digital technologies. Aerial photography is regularly conducted using a quadcopter equipped with a multispectral camera. The obtained images enable the calculation of key vegetation indices characterizing the state of vegetation. Based on these data, the leaf area index and chlorophyll activity are assessed, which serve as important indicators of the biological productivity of plants.
The collected field and remote sensing data are integrated into a unified database for subsequent analysis using modern machine learning and artificial intelligence methods. A comprehensive approach is employed to build predictive models of potato yield, which includes testing various machine learning algorithms: from classical regression methods to modern ensemble algorithms (Random Forest, Gradient Boosting) and neural networks. Particular attention is paid to the methods capable of effectively processing multivariate time series and accounting for nonlinear relationships between factors. The models are trained on historical data using cross-validation techniques to ensure their reliability and resistance to overfitting. The goal of the modeling is to establish quantitative relationships between the spectral characteristics of crops at different stages of vegetation, agroclimatic conditions, and the final yield.
The scientific significance of the work lies in the development of a scientifically grounded approach that combines ground-based measurements, remote sensing data, and machine learning for precise modeling of potato yield.