A sensor-based decision model for precision weed harrowing

Weeds may reduce crop yields significantly if managed improperly. However, excessive herbicide use increases risk of unwanted effects on ecosystems, humans and herbicide resistance development. Weed harrowing is a traditional method to manage weeds mechanically in organic cereals but could also be used in conventional production. The weed control efficacy of weed harrowing can be adjusted by e.g. the angle of the tines. Due to its broadcast nature (both crop and weed plants are disturbed), weed harrowing may have relatively poor selectivity (i.e. small ratio between weed control and crop injury). To improve selectivity, a sensor-based model which takes into account the intra-field variation in weediness and “soil density” in the upper soil layer (draft force of tines), is proposed. The suggested model is a non-linear regression model with three parameters and was based on five field trials in spring barley in SE Norway. The model predicts the optimal weed harrowing intensity (in terms of the tine angle) from the estimated total weed cover and SD per sub-field management unit, as well as a pre-set biological weed threshold (defined as the acceptable total weed cover left untreated). Weed cover and SD were estimated with RGB images (analysed with custom-made machine vision) and an electronic load cell, respectively. With current parameter values, the model should be valid for precision weed harrowing in spring barley in SE Norway. The next step is to test the model, and if successful, adjust it to more cereal species. Weeds may reduce crop yields significantly if managed improperly. However, excessive herbicide use increases risk of unwanted effects on ecosystems, humans and herbicide resistance development. Weed harrowing is a traditional method to manage weeds mechanically in organic cereals but could also be used in conventional production. The weed control efficacy of weed harrowing can be adjusted by e.g. the angle of the tines. Due to its broadcast nature (both crop and weed plants are disturbed), weed harrowing may have relatively poor selectivity (i.e. small ratio between weed control and crop injury). To improve selectivity, a sensor-based model which takes into account the intra-field variation in weediness and “soil density” in the upper soil layer (draft force of tines), is proposed. The suggested model is a non-linear regression model with three parameters and was based on five field trials in spring barley in SE Norway. The model predicts the optimal weed harrowing intensity (in terms of the tine angle) from the estimated total weed cover and SD per sub-field management unit, as well as a pre-set biological weed threshold (defined as the acceptable total weed cover left untreated). Weed cover and SD were estimated with RGB images (analysed with custom-made machine vision) and an electronic load cell, respectively. With current parameter values, the model should be valid for precision weed harrowing in spring barley in SE Norway. The next step is to test the model, and if successful, adjust it to more cereal species.