Effects of weeds on yield and determination of economic thresholds for site-specific weed control using sensor technology

摘要

Weeds can cause high yield losses. Knowledge about the weeds occurring, their distribution within fields and their effects on the crop yield is important to achieve effective weed control. The critical period for weed control (CPWC) and the economic threshold (ET) are important key concepts and management tools in weed control. While the former helps to time weed control in crops of low competitiveness, the latter provides a decision aid to determine whether weed control is necessary. This decision is generally taken at the field level.Weeds have been found to be distributed heterogeneously within fields. Site-specific weed control (SSWC) addresses this sub-field variation by determining weed distribution as input, by taking control decisions in the decision component and by providing control measures as output at high spatial resolution. Sensor systems for automated weed recognition were identified as prerequisite for SSWC since costs for scouting are too high. While experiences with SSWC using sensor data as input are still scarce, studies showed that considerable herbicide savings could be achieved with SSWC.ETs can serve as thresholds for the decision component in SSWC systems. However, the commonly used ETs were suggested decades ago and have not been updated to changing conditions since. The same is the case for the CPWC in maize in Germany. In addition, the approaches to determine the CPWC are usually not based on economic considerations, which are highly relevant to farmers. Thus, the objectives of this thesis are:1. To test different models and to provide a straightforward approach to integrate economical aspects in the concept of the CPWC for two weed control strategies: Herbicide based (Germany) and hoeing based (Benin); 2. To determine the effect of weeds on yield and to calculate ETs under current conditions which can be used for SSWC; 3. To evaluate the use of bi-spectral cameras and shape-based classification algorithms for weed detection in SSWC; and 4. To determine changes in weed frequencies, herbicide use and yield over the last 20 years in southwestern Germany. Datasets in maize from Germany and Benin served as input for the CPWC analyses. The log-logistic model was found to provide a similar fit as the commonly used models but its parameters are biologically meaningful. For Germany, analyses using a full cost model revealed that farmers should aim at applying herbicides early before the 4-leaf stage of maize.In Benin, where weed control is mainly done by hoeing, analyses showed that one well- timed weeding operation around the 10-leaf stage could already be cost-effective. A second weeding operation at a later stage would assure profit.The precision experimental design (PED) was employed to determine the effect of weeds, soil properties and herbicides on crop yield in three winter wheat trials. In this design, large field trials’ geodata of weed distribution, herbicide application, soil properties and yield are used to model the effects of the former three on yield. Galium aparine, other broadleaved weeds and Alopecurus myosuroides reduced yield by 17.5, 1.2 and 12.4 kg ha-1 plant-1 m2 determined by weed counts. The determined thresholds for SSWC with independently applied herbicides were 4, 48 and 12 plants m-2, respectively. Bi-spectral camera based weed–yield estimates were difficult to interpret showing that this technology still needs to be improved. However, large weed patches were correctly identified.ETs derived of field trials’ data carried out at several sites over 13 years in the framework of the ’Gemeinschaftsversuche Baden-Württemberg’ were 9.2-9.8 and 4.5-8.9 % absolute weed coverage for winter wheat and winter barley and 3.7% to 5.5% relative weed coverage for maize. Overall, the weed frequencies in winter cereals were found to be more stable than the weed frequencies in maize during the observation period. In maize, a frequency increase of thermophilic species was found. Trends of considerable yield increases of 0.16, 0.08 and 0.2 t ha-1 for winter wheat, winter barely and maize, respectively, were estimated if weeds were successfully controlled.In order to evaluate the use of bi-spectral cameras and shapebased classification algorithms for weed detection in SSWC, herbicides were applied site-specifically using weed densities determined by bi-spectral camera technology in a winter wheat and maize field. Threshold values were employed for decision taking. Using this approach herbicide savings between 58 and 83 % could be achieved. Such reductions in herbicide use would meet the demand of society to minimize the release of plant protection products in the environment. Misclassification occurred if weeds overlapped with crop plants and crop leaf tips were frequently misclassified as grass weeds. Improvements in equipment, especially between the interfaces of camera, classification algorithms, decision component and sprayer are advisable for further trials.In conclusion, the derived ETs can be easily implemented in a straightforward SSWC system or can serve as decision aid for farmers in winter wheat and winter barley. Further model testing and adjusting would be necessary. For maize, the use of ETs at the field level is not suggested by this study, however the need for early weed control is clearly demonstrated. Bi-spectral camera technology combined with classification algorithms to detect weeds is promising for research use and for SSWC, but still requires some technical improvements.