Therese With Berge
Research Scientist
Attachments
CV (April 2022)Biography
Abstract
Weeds affect crop yield and quality due to competition for resources. In order to reduce the risk of yield losses due to weeds, herbicides or non-chemical measures are applied. Weeds, especially creeping perennial species, are generally distributed in patches within arable fields. Hence, instead of applying control measures uniformly, precision weeding or site-specific weed management (SSWM) is highly recommended. Unmanned aerial vehicle (UAV) imaging is known for wide area coverage and flexible operation frequency, making it a potential solution to generate weed maps at a reasonable cost. Efficient weed mapping algorithms need to be developed together with UAV imagery to facilitate SSWM. Different machine learning (ML) approaches have been developed for image-based weed mapping, either classical ML models or the more up-to-date deep learning (DL) models taking full advantage of parallel computation on a GPU (graphics processing unit). Attention-based transformer DL models, which have seen a recent boom, are expected to overtake classical convolutional neural network (CNN) DL models. This inspired us to develop a transformer DL model for segmenting weeds, cereal crops, and ‘other’ in low-resolution RGB UAV imagery (about 33 mm ground sampling distance, g.s.d.) captured after the cereal crop had turned yellow. Images were acquired during three years in 15 fields with three cereal species (Triticum aestivum, Hordeum vulgare, and Avena sativa) and various weed flora dominated by creeping perennials (mainly Cirsium arvense and Elymus repens). The performance of our transformer model, 1Dtransformer, was evaluated through comparison with a classical DL model, 1DCNN, and two classical ML methods, i.e., random forest (RF) and k-nearest neighbor (KNN). The transformer model showed the best performance with an overall accuracy of 98.694% on pixels set aside for validation. It also agreed best and relatively well with ground reference data on total weed coverage, R2 = 0.598. In this study, we showed the outstanding performance and robustness of a 1Dtransformer model for weed mapping based on UAV imagery for the first time. The model can be used to obtain weed maps in cereals fields known to be infested by perennial weeds. These maps can be used as basis for the generation of prescription maps for SSWM, either pre-harvest, post-harvest, or in the next crop, by applying herbicides or non-chemical measures.
Abstract
Reusing soil can reduce environmental impacts associated with obtaining natural fresh soil during road construction and analogous activities. However, the movement and reuse of soils can spread numerous plant diseases and pests, including propagules of weeds and invasive alien plant species. To avoid the spread of barnyardgrass in reused soil, its seeds must be killed before that soil is spread to new areas. We investigated the possibility of thermal control of barnyardgrass seeds using a prototype of a stationary soil steaming device. One Polish and four Norwegian seed populations were examined for thermal sensitivity. To mimic a natural range in seed moisture content, dried seeds were moistened for 0, 12, 24, or 48 h before steaming. To find effective soil temperatures and whether exposure duration is important, we tested target soil temperatures in the range 60 to 99 C at an exposure duration of 90 s (Experiment 1) and exposure durations of 30, 90, or 180 s with a target temperature of 99 C (Experiment 2). In a third experiment, we tested exposure durations of 90, 180, and 540 s at 99 C (Experiment 3). Obtaining target temperatures was challenging. For target temperatures of 60, 70, 80, and 99 C, the actual temperatures obtained were 59 to 69, 74 to 76, 77 to 83, and 94 to 99 C, respectively. After steaming treatments, seed germination was followed for 28 d in a greenhouse. Maximum soil temperature affected seed germination, but exposure duration did not. Seed premoistening was of influence but varied among temperatures and populations. The relationships between maximum soil temperature and seed germination were described by a common dose–response function. Seed germination was reduced by 50% when the maximum soil temperature reached 62 to 68 C and 90% at 76 to 86 C. For total weed control, 94 C was required in four populations, whereas 79 C was sufficient in one Norwegian population.
Authors
Björn Ringselle Benedikte Watne Oliver Therese With Berge Inger Sundheim Fløistad Liv Berge Lars Olav BrandsæterAbstract
Many herbaceous perennial plant species gain significant competitive advantages from their underground creeping storage and proliferation organs (CR), making them more likely to become successful weeds or invasive plants. To develop efficient control methods against such invasive or weedy creeping perennial plants, it is necessary to identify when the dry weight minimum of their CR (CR DWmin) occurs. Moreover, it is of interest to determine how the timing of CR DWmin differs in species with different light requirements at different light levels. The CR DWmin of Aegopodium podagraria, Elymus repens and Sonchus arvensis were studied in climate chambers under two light levels (100 and 250 μmol m−2 s−1), and Reynoutria japonica, R. sachaliensis and R. × bohemica under one light level (250 μmol m−2 s−1). Under 250 μmol m−2 s−1, the CR DWmin occurred before one fully developed leaf in R. sachaliensis, around 1–2 leaves in A. podagraria and E. repens and around four leaves in S. arvensis, R. japonica and R. × bohemica. In addition to reducing growth in all species, less light resulted in a higher shoot mass fraction in E. repens and S. arvensis, but not A. podagraria; and it delayed the CR DWmin in E. repens, but not S. arvensis. Only 65% of planted A. podagragra rhizomes produced shoots. Beyond the CR DWmin, Reynoutria spp. reinvested in their old CR, while the other species primarily produced new CR. We conclude that A. podagraria, R. sachaliensis and E. repens are vulnerable to control efforts at an earlier developmental stage than S. arvensis, R. japonica and R. × bohemica.
Authors
Ran Nisim Lati Jesper Rasmussen Dionisio Andújar Jose Dorado Therese With Berge Christina Wellhausen Michael Pflanz Henning Nordmeyer Michael Schirrmann Hanan Eizenberg Paul Neve Rasmus Nyholm Jørgensen Svend ChristensenAbstract
The adoption of site-specific weed management (SSWM) technologies by farmers is not aligned with the scientific achievements in this field. While scientists have demonstrated significant success in real-time weed identification, phenotyping and accurate weed mapping by using various sensors and platforms, the integration by farmers of SSWM and weed phenotyping tools into weed management protocols is limited. This gap was therefore a central topic of discussion at the most recent workshop of the SSWM Working Group arranged by the European Weed Research Society (EWRS). This insight paper aims to summarise the presentations and discussions of some of the workshop panels and to highlight different aspects of weed identification and spray application that were thought to hinder SSWM adoption. It also aims to share views and thoughts regarding steps that can be taken to facilitate future implementation of SSWM.
Abstract
This study provides a multi-attribute approach to support decisions by Norwegian crop farmers considering adopting innovative crop protection measures. In modelling choice among pest management strategies, we have accounted for both economic risks, risks to human health and risks to the environment. We used the Simple Multi-Attribute Rating Technique (SMART) to evaluate the results of a field trial comparing four different pest management strategies. In the trial, various pre-crops in year one were followed by two consecutive years of winter wheat. Two treatments had different levels of integrated pest management (IPM). IPM1 was the most innovative treatment and used less pesticides (i.e. herbicides, insecticides and fungicides) than IPM2. The third treatment (‘Worst Case’, WC) used pesticides routinely. The fourth treatment (‘No Plant Protection’, NPP) used no plant protection measures except one reduced dose of herbicide per year on winter wheat. Two main attributes were included in the SMART analysis, an economic indicator and a pesticide load indicator, each of which comprised a number of attributes at a subsidiary level. The results showed that the IPM1 and NPP strategies performed better than IPM2 and the WC strategies. However, the ranking of the pest management practices depended on the weighting of the two main attributes. Although the SMART analysis gave ordinal utility values, permitting only ranking of the alternatives, we were able to transform the results to measure financial differences between the alternatives.
Abstract
Ornamental jewelweed (Impatiens glandulifera Royle) is an alien invasive plant in Europe. This annual plant often grows in riparian habitats where herbicides are prohibited. Several studies have reported the negative effect on ecosystem and ecosystem services by this species. However, limited research is published on control measures and the aim of our study was to explore use of hot water and cutting to control I. glandulifera. A lab experiment showed that the lethal water temperature for seed was between 45 and 50 C. In a pot experiment with seeds in soil, emergence of I. glandulifera was reduced by 78% and 93% compared with the untreated control with volumes of hot water (80 C) of 7.2 and 14.5 L m−2, respectively. When treatments were conducted on relatively tall plants (almost 60 cm) in late June, hot water gave significantly better control than cutting. Compared with an untreated control, I. glandulifera cover was reduced by 97% and 79% after hot water and cutting, respectively. Application of hot water to smaller (<40 cm) and less developed plants (BBCH 12–13) in early June and cutting of plants with visible flower buds (mid-July) led to no significant difference in cover. Compared with an untreated control, I. glandulifera cover was reduced by 99% (cut below first node) and 91% (hot water and cut above first node). When relatively tall plants (almost 60 cm) were treated, hot water use was high (31.1 L m−2) and required twice as many work hours (4.8 min m−2) as cutting (2.4 min m−2). When smaller plants (<40 cm) were targeted, work hours and hot water use were reduced to 2.1 min m−2 and 13.7 L m−2, respectively.
Authors
Björn Ringselle Therese With Berge Daniel Stout Tor Arvid Breland Paul E. Hatcher Espen Haugland Matthias Koesling Kjell Mangerud Tor Lunnan Lars Olav BrandsæterAbstract
Docks (Rumex spp.) are a considerable problem in grassland production worldwide. We investigated how different cultural management techniques affected dock populations during grassland renewal: (I) renewal time, (II) companion crop, (III) false seedbed, (IV) taproot cutting (V), plough skimmer and (VI) ploughing depth. Three factorial split-split plot experiments were carried out in Norway in 2007–2008 (three locations), 2008–2009 (one location) and 2009 (one location). After grassland renewal, more dock plants emerged from seeds than from roots. Summer renewal resulted in more dock seed and root plants than spring renewal. Adding a spring barley companion crop to the grassland crop often reduced dock density and biomass. A false seedbed resulted in 71% fewer dock seed plants following summer renewal, but tended to increase the number of dock plants after spring renewal. In some instances, taproot cutting resulted in less dock biomass, but the effect was weak and inconsistent, and if ploughing was shallow (16 cm) or omitted, it instead increased dock root plant emergence. Fewer root plants emerged after deep ploughing (24 cm) compared to shallow ploughing, and a plough skimmer tended to reduce the number further. We conclude that a competitive companion crop can assist in controlling both dock seed and root plants, but it is more important that the renewal time is favourable to the main crop. Taproot cutting in conjunction with ploughing is not an effective way to reduce dock root plants, but ploughing is more effective if it is deep and a skimmer is used.
Academic – Robotic in-row weed control in vegetables
Trygve Utstumo, Frode Urdal, Anders Brevik, ...
Authors
Trygve Utstumo Frode Urdal Anders Brevik Jarle Dørum Jan Netland Øyvind Overskeid Therese With Berge Jan Tommy GravdahlAbstract
Vegetables and other row-crops represent a large share of the agricultural production. There is a large variation in crop species, and a limited availability in specialized herbicides. The robot presented here utilizes systematic growing techniques to navigate and operate in the field. By the use of machine vision it separates seeded vegetable crops from weed. Each weed within the row is treated with individual herbicide droplets, without affecting the crop. This results in a significant reduction in herbicide use, and allows for the use of herbicides that would otherwise harm the crop. The robot is tailored to this purpose with cost, maintainability, efficient operation and robustness in mind. The three-wheeled design is unconventional, and the design maintains maneuverability and stability with the benefit of reduced weight, complexity and cost. Indoor pot trials with four weed species demonstrated that the Drop-on-Demand system (DoD) could control the weeds with as little as 7.6 μg glyphosate or 0.15 μg iodosulfuron per plant. The results also highlight the importance of liquid characteristics for droplet stability and leaf retention properties. The common herbicide glyphosate had no effect unless mixed with suitable additives. A field trial with the robot was performed in a carrot field, and all the weeds were effectively controlled with the DoD system applying 5.3 μg of glyphosate per droplet. The robot and DoD system represent a paradigm shift to the environmental impact and health risks of weed control, while providing a valuable tool to the producers.
Abstract
The success of precision agriculture relies largely on our ability to identify how the plants’ growth limiting factors vary in time and space. In the field, several stress factors may occur simultaneously, and it is thus crucial to be able to identify the key limitation, in order to decide upon the correct contra-action, e.g., herbicide application. We performed a pot experiment, in which spring wheat was exposed to water shortage, nitrogen deficiency, weed competition (Sinapis alba L.) and fungal infection (Blumeria graminis f. sp. tritici) in a complete, factorial design. A range of sensor measurements were taken every third day from the two-leaf stage until booting of the wheat (BBCH 12 to 40). Already during the first 10 days after stress induction (DAS), both fluorescence measurements and spectral vegetation indices were able to differentiate between non-stressed and stressed wheat plants exposed to water shortage, weed competition or fungal infection. This meant that water shortage and fungal infection could be detected prior to visible symptoms. Nitrogen shortage was detected on the 11–20 DAS. Differentiation of more than one stress factors with the same index was difficult.
Authors
JC. Streibig J. Rasmussen D. Andùjar C. Andreasen Therese With Berge D. Chachalis T. Dittmann Gerhard Gerhardsen T. M. Giselsson P. Hamouz C. Jaeger-Hansen K. Jensen R.N. Jørgensen M. Keller M. Laursen H.S Midtiby J. Nielsen S. Muller H. Nordmeyer G. Peteinatos A Papadopoulos J. Svensgaard M. Weis S. ChristensenAbstract
No abstract has been registered
Abstract
Arable weeds are generally distributed in patches, while herbicides are applied uniformly. Herbicides can be saved if only the patches are sprayed, i.e. patch spraying (PS). Bottlenecks for cost-effective PS are weed monitoring technology and valid technology-based decision rules for PS (thresholds). The novel machine vision algorithm Weedcer has been developed as an efficient weed monitoring tool for PS. Weedcer estimates the proportions of young weed leaves and cereal leaves in high resolution red–green–blue images. We conducted field trials to test relative weed cover (RWC) and relative mayweed cover (RMC) estimated by Weedcer as decision rules for PS. RWC is the total weed cover divided by the total plant cover and RMC is the mayweed cover divided by the total plant cover. The main criterion for evaluation and basis of these thresholds was the measured grain yield. Images (about 0.06-m2) were acquired with a GPS guided autonomous field robot in spring, the normal time for spraying seed-propagated broadleaf weeds in both winter – and spring cereals in Norway. Three map-based trials (weed monitoring and spraying in two separate operations) showed that mean RWC per management unit (12.0 × 12.5-m) was generally adequate. In winter wheat heavily infested with scentless mayweed (Tripleurospermum inodorum (L.) Sch.Bip.) and/or scented mayweed (Matricaria recutita L.), the mean RMC per management unit was more adequate. Progress during the project allowed three additional trials conducted in real-time (weed monitoring and spraying in the same operation). These were conducted with the robot in spring cereals, and showed that a weighted moving average of RWC per image was adequate. The sprayed and unsprayed management units in these trials were minimum 3.0 × 3.0-m and 0.5 × 3.0-m, respectively. Results indicated that the Weedcer-based thresholds should be lower in wheat (Triticum aestivum) than in barley (Hordeum vulgare).
Abstract
Lack of automatic weed detection tools has hampered the adoption of site-specific weed control in cereals. An initial object-oriented algorithm for the automatic detection of broad-leaved weeds in cereals developed by SINTEF ICT (Oslo, Norway) was evaluated. The algorithm ("WeedFinder") estimates total density and cover of broad-leaved weed seedlings in cereal fields from near-ground red-green-blue images. The ability of "WeedFinder" to predict 'spray'/'no spray' decisions according to a previously suggested spray decision model for spring cereals was tested with images from two wheat fields sown with the normal row spacing of the region, 0.125 m. Applying the decision model as a simple look-up table, "WeedFinder" gave correct spray decisions in 65-85% of the test images. With discriminant analysis, corresponding mean rates were 84-90%. Future versions of "WeedFinder" must be more accurate and accommodate weed species recognition.
Abstract
A possible cost-effective real-time patch spraying implementation against seed-propagated broad-leaved weeds in cereals is a camera mounted in front of the tractor taking images at feasible distances in the direction of travel, on-board image analysis software and entire boom switched on and off. To assess this implementation, manual weed counts (0.25 m(2) quadrats) in a 1.5 m x 2 m grid, were used to simulate camera outputs. Each quadrat was classified into 'spray' and 'not spray' decisions based on a threshold model, and the resulting map defined the 'ground truth'. Subsequently, 'on/off' spraying at larger control areas where sizes were given by the boom width and image distance, and spraying decision controlled by weed status at the single quadrat simulating the camera's view, were simulated. These coarser maps were compared with 'ground truth', to estimate mapping error (area above threshold not sprayed), spraying error (area below threshold sprayed), total error (sum of mapping and spraying error) and the herbicide reduction. Three levels of the threshold model were tested. Results were used to fit models that predict errors from boom width and image distance. Size of control area did not on average affect the magnitude of the simulated herbicide reductions, but the bigger the control area the higher the risk that the simulated herbicide reduction deviate from the reduction in 'ground truth'. Mean simulated herbicide reductions were 42-59%, depending on threshold level. Only minor differences due to threshold level were seen for mean mapping and spraying errors at given spraying resolutions. Using original threshold level and image distance 2 m, predicted total errors for boom widths 2 m, 6 m, 20 m and 40 m would be 6%, 10%, 16% and 17%, respectively. Results indicate that control area should not exceed about 10 m 2 if acceptable total error is maximum 10%.
Authors
Marian Malte Weigel Therese With Berge Jukka Salonen Timo Lötjönen Bärbel Gerowitt Lars Olav BrandsæterAbstract
Controlling creeping perennial weeds is challenging throughout all farming systems. The present study distinguished and explored three different methods to control them non-chemically: disturbance with inversion, disturbance without inversion, and competition. Focusing on Cirsium arvense, Elymus repens, and Sonchus arvensis, we conducted a field study (2019–2021) at three northern European sites in Germany, Finland, and Norway. We investigated the effects of the control methods ploughing (inversion disturbance), root cutting (non-inversion disturbance), and cover crops (competition) alone. Root cutting was conducted using a prototype machine developed by “Kverneland”. Eight treatments were tested in factorial designs adapted for each site. Control methods were applied solely and combined. Response variables after treatments were aboveground weed biomass and grain yield of spring cereals. The control method of ploughing was most effective in reducing weed biomass compared to root cutting or cover crops. However, compared to the untreated control, a pronounced additive effect of root cutting and cover crops occurred, reducing weed biomass (−57.5%) similar to ploughing (−66%). Pooled over sites, the response was species-specific, with each species showing a distinct reaction to both control methods. C. arvense was most susceptible to root cutting, followed by E. repens, while S. arvensis showed no susceptibility. Crop yield losses were prevented compared to untreated plots by ploughing (+60.57%) and root cutting (+30%), but not by cover crops. We conclude that the combination of non-inversion disturbance and competition is a promising strategy to reduce the reliance on herbicides or inversion tillage in the management of perennial weeds.
Authors
Therese With BergeAbstract
No abstract has been registered
Abstract
Weed harrowing is commonly used to manage weeds in organic farming but is also applied in conventional farming to replace herbicides. Due to its whole-field application, weed harrowing after crop emergence has relatively poor selectivity and may cause crop damage. Weediness generally varies within a field. Therefore, there is a potential to improve the selectivity and consider the within-field variation in weediness. This paper describes a decision model for precision post-emergence weed harrowing in cereals based on experimental data in spring barley and nonlinear regression analysis. The model predicts the optimal weed harrowing intensity in terms of the tine angle of the harrow for a given weediness (in terms of percentage weed cover), a given draft force of tines, and the biological weed damage threshold (in terms of percentage weed cover). Weed cover was measured with near-ground RGB images analyzed with a machine vision algorithm based on deep learning techniques. The draft force of tines was estimated with an electronic load cell. The proposed model is the first that uses a weed damage threshold in addition to site-specific values of weed cover and soil hardness to predict the site-specific optimal weed harrow tine angle. Future field trials should validate the suggested model.
