The Role of Remote Sensing Data and Methods in a Modern Approach to Fertilization in Precision Agriculture

Summary

This review and comparative study synthesises conventional and state-of-the-art machine learning approaches to precision fertilisation, integrating agronomic soil sampling with remote sensing data and spatial interpolation techniques. The authors evaluated eight prediction methods (four conventional, four machine learning) using field data from a Croatian agricultural parcel and demonstrated that hybrid approaches combining remote sensing imagery with machine learning retained superior prediction accuracy and computational efficiency. The work addresses limitations of conventional soil prediction methods—namely lower accuracy, lack of robustness, and high cost—by leveraging open-data satellite missions and UAV-based imagery.

UK applicability

The methodologies for integrating remote sensing with soil prediction are widely applicable to UK arable production, where precision fertilisation can reduce input costs and environmental burden. However, the study was conducted in Croatia; UK practitioners would need to validate these methods against local soil types, climatic conditions, and available satellite data.

Key measures

Prediction accuracy of conventional interpolation methods versus machine learning methods; spatial prediction of P2O5 and K2O concentrations; integration of remote sensing data for hybrid interpolation

Outcomes reported

The study evaluated conventional (ordinary kriging, inverse distance weighted) and modern machine learning methods (random forest, support vector machine, artificial neural networks, decision trees) for predicting soil phosphorous pentoxide (P2O5) and potassium oxide (K2O) concentrations. Remote sensing data from multispectral, thermal, and radar satellite imagery as well as unmanned aerial vehicle (UAV)-based imagery were integrated to improve prediction accuracy in precision fertilisation.

Topic tags