A Farmer-Oriented Decision Support System for Site-Specific Weed Management in Maize

The transition towards more sustainable cropping systems requires substantial reductions in herbicide use without compromising crop productivity. Although site-specific weed management (SSWM) is widely recognised as a promising approach, its adoption remains limited due to technical complexity, cost, and limited transferability of existing solutions. To address these barriers, we developed SWIM (Smart Weed Impact Management), a user-oriented decision support system that transforms low-cost UAV-RGB imagery into operational herbicide prescription maps for maize fields. Unlike most image-based approaches focused solely on weed mapping, SWIM directly links weed infestation to expected yield loss and economic intervention thresholds, enabling management decisions based on agronomic and economic criteria. SWIM integrates three sequential modules: (i) weed detection, (ii) estimation of yield loss due to weed competition, and (iii) generation of prescription maps for Patch Spraying and Variable Rate Application (PSA and VRA). The system is implemented in two versions: a Simple Method (SM), based on minimal user expertise and rapid site- specific calibration, and an Improved Method (IM), which requires the evaluation of spatial crop density and a more extensive calibration activity to improve accuracy. SWIM was tested in a maize field located in Central Italy using UAV-acquired RGB imagery. Weed green cover was estimated through a simple image-based approach in which maize green cover was subtracted from total green cover derived from aerial images, providing a low-cost alternative to more complex weed- recognition techniques. The resulting weed green cover estimates were subsequently used to quantify competitive pressure and identify economically justified intervention thresholds. Both methods showed strong performance in weed detection, with root mean square errors (RMSE) below 0.10 and concordance correlation coefficients (CCC) exceeding 0.91. However, the IM more effectively captured within-field spatial variability and provided more robust estimation of the economic intervention thresholds. Consequently, the resulting prescription maps identified substantial opportunities for reducing herbicide applications, with estimated savings of 44% (for SM) and 28% (for IM) compared with conventional uniform application. Beyond its technical performance, SWIM introduces a practical framework that actively involves farmers in defining acceptable economic risk levels while avoiding dependence on expensive sensors, proprietary software, or advanced artificial intelligence models. This combination of accessibility, transparency, and agronomic relevance makes the system particularly suitable for supporting the broader adoption of precision agriculture in real farming conditions. The proposed DSS provides a practical framework for improving herbicide-use efficiency through SSWM while maintaining a strong focus on operational feasibility. By combining low-cost sensing technologies with economically based decision rules, SWIM may contribute to the wider adoption of precision weed management in maize production systems. In addition, its farmer-oriented design can help bridge the gap between digital innovation and practical field management, promoting a more effective integration of precision agriculture technologies into routine farming decisions.