Jadwal Sholat

Memuat jadwal sholat…

Pertanian & Lingkungan editorial

Open AccessOA2022

Statistical and Machine Learning Methods for Crop Yield Prediction in the Context of Precision Agriculture

Random forests achieved R² of 0.85 for corn and 0.94 for soybeans, outperforming multiple linear regression (R² 0.40 and 0.45) in predicting crop yield from high-resolution soil and topographic data. Machine learning shows potential for identifying high-yield areas without existing yield maps.
H. Burdett; C. Wellen· Precision Agriculture· 2022· DOI 10.1007/s11119-022-09897-0

The core problem

Understanding the relationships between crop yield, soil properties, and topographic characteristics is critical for precision agriculture management. This study addresses the need to quantify these relationships using high-resolution data and advanced analytical techniques. Conducted in Southwestern Ontario, Canada—a region where few studies have assessed the impact of precision agriculture and machine learning on soil property-yield relationships—the research compares multiple methods for predicting crop yield at a subfield scale (2 m). The dataset comprises 145,500 observations of corn and soybean yield, along with topographic and soil nutrient characteristics including pH, soil organic matter (OM), cation exchange capacity (CEC), soil test phosphorus, zinc (Zn), potassium (K), elevation, and topographic wetness index. The objective is to identify techniques capable of accurately relating soil properties to crop yields and to evaluate their predictive performance.

Innovation

The study utilized a multiple field dataset from Southwestern Ontario, Canada, containing 145,500 observations of corn and soybean yield, topographic attributes, and soil nutrient properties. Eight attributes were considered: pH, OM, CEC, phosphorus, Zn, K, elevation, and topographic wetness index. Four predictive methods were compared: multiple linear regression (MLR), artificial neural networks (ANN), decision trees, and random forests. The models were trained and evaluated using cross-validation experiments to assess their ability to predict yield from soil and topographic characteristics. Performance was measured using the coefficient of determination (). The high-resolution data allowed analysis at a 2 m subfield scale, enabling detailed assessment of within-field variability.
Introduction
Understanding the relationships between crop yield, soil properties, and topographic characteristics is critical for precision agriculture management. This study addresses the need to quantify these relationships using high-resolution data and advanced analytical techniques. Conducted in Southwestern Ontario, Canada—a region where few studies have assessed the impact of precision agriculture and machine learning on soil property-yield relationships—the research compares multiple methods for predicting crop yield at a subfield scale (2 m). The dataset comprises 145,500 observations of corn and soybean yield, along with topographic and soil nutrient characteristics including pH, soil organic matter (OM), cation exchange capacity (CEC), soil test phosphorus, zinc (Zn), potassium (K), elevation, and topographic wetness index. The objective is to identify techniques capable of accurately relating soil properties to crop yields and to evaluate their predictive performance.

Why it matters

The results indicate that machine learning techniques, particularly random forests, are effective at predicting crop yield from soil and topographic properties at a subfield scale. The high values for random forests suggest that these models can capture complex nonlinear relationships that MLR fails to address. The ability of random forest models to predict yield in fields not used for training implies that they could be applied to fields without existing yield maps, provided those fields share similar soil property-yield relationships. However, the variability in cross-validation success highlights the need for caution when extrapolating to new fields. The identification of significant attributes (e.g., pH, OM, CEC) provides insights for precision agriculture management, though uncertainty in attribute importance warrants further investigation. Overall, the study supports the use of machine learning for precision agriculture applications, particularly for identifying high-yield areas and informing site-specific management.

Who should read this

Pembaca Research Digest dan praktisi rumpun ini.

Opening member content…