Die Bereitstellung von Speisemohren fur den deutschen Markt erfolgt zu einem nicht unerheblichen Anteil aus der Inlandsproduktion im Freiland. Um eine optimale Herbizidwirkung bei minimaler Schadigung der Mohrenpflanze zu erzielen, hat sich ein bis zu dreimaliges Splitting der empfohlenen Hochstmenge bewahrt. Da in den spezialisierten Betrieben teilweise hohe Flachenkonzentrationen erreicht werden, besteht bei einer der Unkrautdichte angepassten Herbizidapplikation ein hohes Potenzial zur Reduzierung des Einsatzes von Pflanzenschutzmitteln.In Streifenversuchen sollte die Wirkung eines teilflachenspezifischen Herbizidsplittings auf die Spatverunkrautung sowie auf verschiedene Ertragsparameter getestet werden. Art und Anzahl der Unkrauter wurden im Fruhjahr mittels manueller Bonituren entlang eines Stichprobengitters erfasst. Danach erfolgte im 3-Blattstadium der Mohren mit einem am Leibniz-Institut fur Agrartechnik (ATB) entwickelten Kamerasensor die kleinraumige Detektion der auf dem Feld vorhandenen Verunkrautung. Die Sensorwerte waren die Grundlage fur das Erstellen von Unkrautverteilungskarten. Anhand dieser Verteilungskarten wurden in dem der Unkrautverteilung angepassten Applikationsstreifen drei Zonen mit Aufwandmengen von 200 L ha-1, 300 L ha-1, und 400 L ha-1 gebildet. In den beidseitig benachbarten Streifen erfolgte ein flacheneinheitliches Splitting mit der betriebsublichen Menge von 400 L ha-1. Die drei Applikationszonen wurden wahrend des Splittings bei der zweiten bzw. dritten Herbizidanwendung beibehalten. An jedem der zwei bzw. drei Spritzzeitpunkte konnten durch das teilflachenspezifische Splitting 16 % (2005) und 20 % (2006) des Pflanzenschutzmittels gegenuber einem flacheneinheitlichen Splitting eingespart werden.Zur Beurteilung der Ertragswirksamkeit wurden in den 2 Applikationsvarianten an jeweils gegenuberliegenden Stichprobenpunkten die Mohren von Hand geerntet und die Ertragsparameter Gewicht sowie Anzahl „gesamt“ bzw. „vermarktungsfahig“ bestimmt. Mit Hilfe der Differenzenmethode (lokale Ertragsparameterwerte teilflachenspezifisch minus einheitlich) erfolgte der statistische Vergleich abhangiger Stichproben mit dem t- Test. Unterstellt man eine Irrtumswahrscheinlichkeit von α = 5 % wurden in 19 der insgesamt 24 Tests keine statistisch gesicherten Unterschiede der Ertragsparameter zwischen den zwei Behandlungsvarianten in den jeweiligen Applikationszonen gefunden. Die Nachverunkrautung war in beiden Varianten sehr gering und damit vernachlassigbar.Stichworter: Applikationskarte, Kamera, Mohren, Prazise Unkrautkontrolle, SensorSite-specific herbicide splitting in field carrots based on camera detected weed infestationAbstractThe production of carrots for the German market comes mainly from domestic production. To ensure the efficiency of chemical weed control and to minimize the damage of the crop a splitting of the recommended dosage up to three times is often practiced. Because of large field areas of the carrot cropping and processing enterprises, the potential to save herbicides by practicing an herbicide application adapted to the weed occurrence is high.The efficiency of a site-specific herbicide splitting on the late weed occurrence as well as on yield parameters was tested in field strip trials. Weed species and abundance were determined manually by raster sampling using a counting frame in spring before spraying. Afterwards in the three leaf growth stages of the carrots the weed coverage level was detected online using a camera sensor developed by the Leibniz Institute for Agricultural Engineering (ATB). Based on weed coverage level maps three application zones (200 L ha-1, 300 L ha-1, 400 L ha-1) were defined. On both sides of the site specific splitting strip a uniform splitting strip (400 L ha-1) was applied. The position of the application zones were the same during site-specific splitting at the second and third herbicide spraying respectively. Compared to a conventional uniform splitting herbicide savings were 16% (2005) und 20% (2006) at each spraying time.To evaluate the efficacy of the site-specific splitting on the yield manually harvesting were performed at opposite points in both treatments. The yield parameters fresh weight and numbers of carrots “total” and “marketable” were determined. Assuming a significance level of α = 5% the difference method for controlled treatment comparison in large scale field trials (t-test) resulted in 19 of the 24 tests in total no differences between the treatments. The late weed occurrence in both treatments was low.Keywords: Application map, camera, carrots, precise weed control, sensor

Pflanzenschutzmitteln. In Streifenversuchen sollte die Wirkung eines teilflachenspezifischen Herbizidsplittings auf die Spatverunkrautung sowie auf verschiedene Ertragsparameter getestet werden. Art und Anzahl der Unkrauter wurden im Fruhjahr mittels manueller Bonituren entlang eines Stichprobengitters erfasst. Danach erfolgte im 3-Blattstadium der Mohren mit einem am Leibniz-Institut fur Agrartechnik (ATB) entwickelten Kamerasensor die kleinraumige Detektion der auf dem Feld vorhandenen Verunkrautung. Die Sensorwerte waren die Grundlage fur das Erstellen von Unkrautverteilungskarten. Anhand dieser Verteilungskarten wurden in dem der Unkrautverteilung angepassten Applikationsstreifen drei Zonen mit Aufwandmengen von 200 L ha -1 , 300 L ha -1 , und 400 L ha -1 gebildet. In den beidseitig benachbarten Streifen erfolgte ein flacheneinheitliches Splitting mit der betriebsublichen Menge von 400 L ha -1 . Die drei Applikationszonen wurden wahrend des Splittings bei der zweiten bzw. dritten Herbizidanwendung beibehalten. An jedem der zwei bzw. drei Spritzzeitpunkte konnten durch das teilflachenspezifische Splitting 16 % (2005) und 20 % (2006) des Pflanzenschutzmittels gegenuber einem flacheneinheitlichen Splitting eingespart werden. Zur Beurteilung der Ertragswirksamkeit wurden in den 2 Applikationsvarianten an jeweils gegenuberliegenden Stichprobenpunkten die Mohren von Hand geerntet und die Ertragsparameter Gewicht sowie Anzahl „gesamt“ bzw. „vermarktungsfahig“ bestimmt. Mit Hilfe der Differenzenmethode (lokale Ertragsparameterwerte teilflachenspezifisch minus einheitlich) erfolgte der statistische Vergleich abhangiger Stichproben mit dem tTest. Unterstellt man eine Irrtumswahrscheinlichkeit von α = 5 % wurden in 19 der insgesamt 24 Tests keine statistisch gesicherten Unterschiede der Ertragsparameter zwischen den zwei Behandlungsvarianten in den jeweiligen Applikationszonen gefunden. Die Nachverunkrautung war in beiden Varianten sehr gering und damit vernachlassigbar. Stichworter: Applikationskarte, Kamera, Mohren, Prazise Unkrautkontrolle, Sensor

Am Leibniz-Institut fur Agrartechnik wurde in Zusammenarbeit mit den Firmen SYMACON Bildverarbeitung GmbH Barleben und Muller-Elektronik GmbH & Co. KG Salzkotten ein sensorgesteuertes Feldspritzensystem zur Applikation von Pflanzenschutzmitteln entwickelt. Bei engstehenden Reihenkulturen wie Getreide erfolgt die Unkrauterfassung in der Fahrspur. Systembedingt wird nicht zwischen Unkraut und Kulturpflanze unterschieden. Dadurch ist ein langer Einsatzzeitraum des Systems gegeben, zumal eine Unterscheidung von Kulturpflanze und Unkraut je nach Uberlappungsgrad im fortgeschrittenen Wachstumsstadium auserst schwierig bzw. nicht mehr moglich ist. Situationsbedingt konnen als Regelparameter die Unkrautanzahl (fruhe Wachstumsstadien des Unkrautes) oder der Unkrautdeckungsgrad (spate Wachstumsstadien) verwendet werden. Auf Grund der mechanischen Tragheit der Feldspritze beim Erreichen des Sollwertes der Applikationsmenge (Druckanpassung innerhalb des Regelbereiches der Dusen) ist ein gewisser Abstand zwischen dem Sensor und dem Spritzbalken notwendig. Daher ist der Kamerasensor am Frontdreipunkt des Traktors positioniert. Feldspritzen mit einer schnellen kontinuierlichen Regelung der Applikationsmenge, die eine Sensorpositionierung am Spritzbalken erlauben, werden zurzeit nicht produziert. Im Unterschied zu anderen am Markt verfugbaren bzw. in Entwicklung befindlichen sensorgestutzten Applikationssystemen, wo Dusen bzw. Teilbreiten an- und abgeschaltet werden, wird entsprechend einer definierten Regelfunktion wahrend der Fahrt die Applikationsmenge entsprechend dem Sensorsignal variiert. In Feldbereichen, in denen die Verunkrautung einen bestimmten Schwellenwert ubersteigt, wird mit der betriebsublichen Menge gespritzt, wahrend in Bereichen ohne Unkraut bis zu 50 % reduziert wird. Zwischen 50 % bis 100 % erfolgt eine proportionale Anpassung der Applikationsmenge. Im Gegensatz zu einer Dusenbzw. Teilbreitenschaltung „an/aus“ bleiben bei einer kontinuierlichen Regelung keine Feldbereiche unbehandelt. Die Gefahr einer ungehinderten Unkrautentwicklung bis hin zur Samenbildung ist nicht gegeben. Im Herbst 2009 wurde mit dem sensorgesteuerten Spritzensystem in einem 26 ha grosen Wintergerstenfeld eines Landwirtschaftsbetriebes eine Herbizidapplikation durchgefuhrt. Die Mitteleinsparung belief sich auf 20 %. Stichworter: Multispektralkamera, prazise Unkrautkontrolle, Unkrautsensor, variable Aufwandmengen Online variable rate herbicide application using a camera sensor In cooperation with the two companies SYMACON Bildverarbeitung GmbH Barleben and Muller-Elektronik GmbH & Co. KG Salzkotten a sensor controlled field sprayer for precise plant protection was developed at the Leibniz-institute for Agricultural Engineering. In narrow seeded field crops like cereals, the weed detection is done within the tramlines. Because of the way the system operates, there is no discrimination between cultivated crops and weeds. In later growth stages, crop and weed are overlapping and discrimination within the plant stand becomes difficult or even impossible. Situational, the weed number (early growth stages) or the weed coverage level (late growth stages) can be used as parameter to control the field sprayer. The sensor is positioned at the front three point linkage of the tractor. Field sprayers with a fast adaption of the application rate are not yet commercially available. In contrast to other “on/off” sensor controlled application systems which are on the market or under development, the application rate in the presented technology is in- or decreased according to the sensor signal. At heavy weed infested sites of the field, the customary application rate is sprayed. It is reduced up to 50 % at sites with low weed pressure. Between the volume of 50 % and 100 %, the application rate is proportional adapted to the weed infestation. In contrast to the “on/off” switching of single nozzles or boom sections, no sites are left unsprayed using the variable rate approach. There is no risk of unobstructed weed development or seed setting. In autumn 2009, a 26 ha winter barley field on a farm was sprayed against weeds with the sensor controlled field sprayer. The product Falkon® was used with the dosage of 1 l/ha. The spray volume was varied between 100 and 200 l/ha. The product savings were on average 20 %. Keywords: Multispectral camera, precise weed control, sensor-controlled sprayer, weed sensor

In early growth stages of weeds there was a linear dependency between number of weeds and coverage level detected by camera vision. If older weeds occurred in the population the two parameters number of weeds and coverage level became uncorrelated. In four year field trials the percentage of young weeds on the camera detected abundance were highest, while the oldest one caused the highest percentage on the coverage level. The age structure of the weed population is often heterogeneous. If weed control with variable rates according to the camera detected abundance is carried out, miss dosages could be occur. Older weeds with low abundance would be under treated with low application rates in contrast to a variable rate dosage algorithm using the coverage level of weeds. If threshold based weed control decisions with zero rates according to weed abundance are used, this could lead to untreated areas with single old weeds causing a high competition against the crop. Under a heterogeneous weed population a threshold based weed control should not be carried out. A variable rate application of herbicides should then be done according to coverage level of weeds instead of using their abundance.

An important step towards application of variable herbicide rates is the development of online sensors for detecting weeds. Field trials were conducted with a sensor-controlled field sprayer from the year 2000 to 2003 in cereals and pea. The aim was to quantify the influence of the real-time application of variable herbicide rates on the amount of herbicides used, grain yield and late weed infestation. In 13 field trials average herbicide savings of 24.6% were achieved (cereals: 22.8%, pea: 27.9%), and compared with conventional application on average no yield reduction was caused by sensor-based herbicide application. When late weed infestation was checked shortly before harvest, no differences in weed density were observed between areas with reduced and areas with standard dosage.

Within a spring barley field in two successive years variable rates of herbicides were applied in real time with a sensor operated field sprayer. Grid sampling was conducted along a 24 x 50 in sampling grid. The two weed species Stellaria media L. and Polygonum convolvulus L. were most abundant. The average Jaccard number of 0.5 showed that between the two years the species composition at the sampling points was different. The relative frequency of the two main species at the sampling points was for black bindweed 26 % (2002), 63 % (2003) and for common chickweed 59 % (2002) and 19 % (2003). The dispersion of the weeds between the two years was different. The analysis of the spatial covariance between the sensor values of 2002 and 2003 showed that there was no spatial dependence of the weed occurrence between the two years.

Effect of sensor based herbicide application in grain cereals and peas Field trials for herbicide application in real time were conducted with a sensor operated field sprayer in the last 4 years in grain cereals and pea. Grid sampling for the estimation of dispersion and abundance took place before spraying. For assessment of the effectiveness of the operational method, the late weed occurrence was recorded at the same sampling points before harvesting. In spite of herbicide savings a higher late weed occurrence could not be obtained. The abundance of the weeds was not dependent from the application rate at the sampling points. There were neither harvest difficulties nor green residues within the harvested grains.

A method of site-specific real-time application of herbicides in narrow row crops like cereals and legumes are described. Weed detection is done within the tramlines by an optoelectronic sensor. There is no discrimination between weed species, because an functional dependency between the weed counts in total and the calculated species dependent yield loss were found. For an experimental field (Triticale) there was a good correlation between the calculated species dependent yield loss and the estimated species-independent yield loss. In an other field trial in winter rye were site-specific real-time application of herbicides practised there was an herbicide saving of 30%. At further test in grey peas 22% of the herbicide could be saved. Yield differences didn't appear.

Because of the spatial variability of the occurrence of pest and diseases as well as weeds an adjustment of the spraying liquid amount on the real intensity of infection makes sense. The development of sensors for detecting the nitrogen demand, the crop density and the weed occurrence, require spraying equipment which guarantee an precise variable rate application. At present control systems of sprayers are working well only in case of an constant application of the plant protection agent.Experiments in crop stands under practical conditions were done with a sensor as well as hard- and software equipped sprayer for detecting and spraying variable rates in real-time. Conclusions arising there from according to dosage faults and operation limits are drawn. The required and the real spraying liquid amount were compared. The required precision of dosage was not satisfied in the case of detection length of 5 meters. Depending on the working speed of the sprayer it became more precise from a length of 30 meters upwards if the spatial variability of the weeds is more on an larger scale. Within an winter rye field there were deviations from the required dosage more than 20 % on 27 % of the 5 meter long field blocks.From our investigations the following conclusions for use of the variable rate sprayers could be drawn:A variability of the required spraying liquid amount from 0 % to 100 %.Fast variation of the spraying liquid amount.Simultaneous variable rate application of more than one plant protection agent.

Plant surface to be protected against fungal diseases differs on small-scale in he- terogeneous cereal crops. Measurements of several years in winter cereals showed a linear relationship between the parameter plant height and plant surface when ears are fully emerged. Small-scale information about the actual plant height within the cereal stand can be obtained by indirect measurement with the "pen- dulum-meter", a mechanical sensor to detect biomass. There was a high linear correlation between the deviation angle of the "pendulum-meter" and measure- ments of plant height along transects within heterogeneous cereal stands. To ob- tain plant surface per unit area to adjust application rates knowledge about stem density is necessary. But correlation between the deviation angle of the "pendu- lum-meter" and measurements of stem density at the transect sampling points was weak.

Various scientific publications regarding site specific weed regulation show that 25 % to 70% of the amount of herbicides commonly needed could be saved depending on the technical method used. Weed monitoring by walking to detect the distribution of weeds within the field is timeconsuming and expensive. Therefore an economic use of site specific weed control can only be achieved if simple sensoric detection methods combine weed recording and herbicide application in one working cycle. To replace manual weed recording, an optoelectronic "green sensor" was tested. First results from the use of the optoelectronic sensor between the rows of a corn field and in the tram lines in cereal fields yielded a linear correlation coefficient between the manual and sensoric counts of 0,6 and 0,9.With a machine combination (tractor, plant protection equipment) field tests were carried out with different crops on 130 hectares of land. On the average, 20 % of the herbicides could be guaranteeing successful treatment at the same time saved.

Different works regarding site specific weed regulation show that 25% to 70% of the amount of herbicides could be saved in dependence of the used technical method. Weed monitoring by walking to detect the distribution within the field is time consuming and expensive. An economic use of site specific weed control shall therefore expect only if simple sensoric detection methods make weed recording and herbicide application possible in one working cycle. Instead of manual weed recording an optoelectronic "green sensor" was tested.The represented results show a good correlation between the weed counts in total and the sum of the yield losses due to certain weed species. An economic assessment of a herbicide application by use of the prevented yield loss has therefore not been carried out by consideration of individual weed species but by weeds in total. Treatment thresholds for site specific spraying on the fields were established by "intrinsically nonlinear regression" using weeds in total.First results for the use of the optoelectronic sensor within rows of a corn field and within the tram lines in cereal fields yielded a linear coefficient of correlation between the manual and sensoric counts of 0,6 to 0,9.The signal from the sensor in connection with the equipment combination (tractor, field sprayer) proved as promising fora real time regulation of the application rate.

The site specific herbicide application after damage thresholds requires the area proportional examination of the weed distribution. However; weeds counts cause costs which lie over the obtained savings. Simple methods must therefore be developed. An alternative method for the weed recognition is counting weeds with a optoelectronic sensor.The damage to expect of the weed infestation had been calculated from the competition determined experimentally of the weeds. The correlations between the weeds frequency and the yield loss shall be used for the derivation of application decisions for the herbicide application.Correlation coefficients were found out between the hand and sensor count of 0,6 to 0,9 in field tests at different cereal species and maize.

Weed density in slow growing crops can be detected by aerial photographs because the weeds are more developed than the crop, The methodology of taking aerial photographs is state of the art. The interpretation of the pictures must be investigated. Results of experiments show the relationship between the red and infrared reflection (Normalized Vegetation Index NVI) and the altitude of the aircraft as well as the number of image pixels.