Search

Robotic weed control has seen increased research of late with its potential for boosting productivity in agriculture. Majority of works focus on developing robotics for croplands, ignoring the weed management problems facing rangeland stock farmers. Perhaps the greatest obstacle to widespread uptake of robotic weed control is the robust classification of weed species in their natural environment. The unparalleled successes of deep learning make it an ideal candidate for recognising various weed species in the complex rangeland environment. This work contributes the first large, public, multiclass image dataset of weed species from the Australian rangelands; allowing for the development of robust classification methods to make robotic weed control viable. The DeepWeeds dataset consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia. This paper presents a baseline for classification performance on the dataset using the benchmark deep learning models, Inception-v3 and ResNet-50. These models achieved an average classification accuracy of 95.1% and 95.7%, respectively. We also demonstrate real time performance of the ResNet-50 architecture, with an average inference time of 53.4 ms per image. These strong results bode well for future field implementation of robotic weed control methods in the Australian rangelands.

Research science used to inform public policy decisions, herein defined as "Policy-Science", is rarely subjected to rigorous checking, testing and replication. Studies of biomedical and other sciences indicate that a considerable fraction of published peer-reviewed scientific literature, perhaps half, has significant flaws. To demonstrate the potential failings of the present approaches to scientific Quality Control (QC), we describe examples of science associated with perceived threats to the Great Barrier Reef (GBR), Australia. There appears a serious risk of efforts to improve the health of the GBR being directed inefficiently and/or away from the more serious threats. We suggest the need for a new organisation to undertake quality reviews and audits of important scientific results that underpin government spending decisions on the environment. Logically, such a body could also examine policy science in other key areas where governments rely heavily upon scientific results, such as education, health and criminology., (Copyright © 2017 Elsevier Ltd. All rights reserved.)

The coastal hypersaline system of the Great Barrier Reef (GBR) in the dry season, was investigated for the first time using a 3D baroclinic model. In the shallow coastal embayments, salinity increases to c.a. 1‰ above typical offshore salinity (~35.4‰). This salinity increase is due to high evaporation rates and negligible freshwater input. The hypersalinity drifts longshore north-westward due to south-easterly trade winds and may eventually pass capes or headlands, e.g. Cape Cleveland, where the water is considerably deeper (c.a. 15m). Here, a pronounced thermohaline circulation is predicted to occur which flushes the hypersalinity offshore at velocities of up to 0.08m/s. Flushing time of the coastal embayments is around 2-3weeks. During the dry season early summer, the thermohaline circulation reduces and therefore, flushing times are predicted to be slight longer due to the reduced onshore-offshore density gradient compared to that in the dry season winter period., (Copyright © 2016 Elsevier Ltd. All rights reserved.)

Dredging poses a potential risk to tropical ecosystems, especially in turbidity-sensitive environments such as coral reefs, filter feeding communities and seagrasses. There is little detailed observational time-series data on the spatial effects of dredging on turbidity and light and defining likely footprints is a fundamental task for impact prediction, the EIA process, and for designing monitoring projects when dredging is underway. It is also important for public perception of risks associated with dredging. Using an extensive collection of in situ water quality data (73 sites) from three recent large scale capital dredging programs in Australia, and which included extensive pre-dredging baseline data, we describe relationships with distance from dredging for a range of water quality metrics. Using a criterion to define a zone of potential impact of where the water quality value exceeds the 80th percentile of the baseline value for turbidity-based metrics or the 20th percentile for the light based metrics, effects were observed predominantly up to three km from dredging, but in one instance up to nearly 20 km. This upper (~20 km) limit was unusual and caused by a local oceanographic feature of consistent unidirectional flow during the project. Water quality loggers were located along the principal axis of this flow (from 200 m to 30 km) and provided the opportunity to develop a matrix of exposure based on running means calculated across multiple time periods (from hours to one month) and distance from the dredging, and summarized across a broad range of percentile values. This information can be used to more formally develop water quality thresholds for benthic organisms, such as corals, filter-feeders (e.g. sponges) and seagrasses in future laboratory- and field-based studies using environmentally realistic and relevant exposure scenarios, that may be used to further refine distance based analyses of impact, potentially further reducing the size of the dredging footprint.

Maintenance and capital dredging represents a potential risk to tropical environments, especially in turbidity-sensitive environments such as coral reefs. There is little detailed, published observational time-series data that quantifies how dredging affects seawater quality conditions temporally and spatially. This information is needed to test realistic exposure scenarios to better understand the seawater-quality implications of dredging and ultimately to better predict and manage impacts of future projects. Using data from three recent major capital dredging programs in North Western Australia, the extent and duration of natural (baseline) and dredging-related turbidity events are described over periods ranging from hours to weeks. Very close to dredging i.e. <500 m distance, a characteristic features of these particular case studies was high temporal variability. Over several hours suspended sediment concentrations (SSCs) can range from 100-500 mg L-1. Less turbid conditions (10-80 mg L-1) can persist over several days but over longer periods (weeks to months) averages were <10 mg L-1. During turbidity events all benthic light was sometimes extinguished, even in the shallow reefal environment, however a much more common feature was very low light 'caliginous' or daytime twilight periods. Compared to pre-dredging conditions, dredging increased the intensity, duration and frequency of the turbidity events by 10-, 5- and 3-fold respectively (at sites <500 m from dredging). However, when averaged across the entire dredging period of 80-180 weeks, turbidity values only increased by 2-3 fold above pre-dredging levels. Similarly, the upper percentile values (e.g., P99, P95) of seawater quality parameters can be highly elevated over short periods, but converge to values only marginally above baseline states over longer periods. Dredging in these studies altered the overall probability density distribution, increasing the frequency of extreme values. As such, attempts to understand the potential biological impacts must consider impacts across telescoping-time frames and changes to extreme conditions in addition to comparing central tendency (mean/median). An analysis technique to capture the entire range of likely conditions over time-frames from hours to weeks is described using a running means/percentile approach.

A disposable instrument for measuring water turbidity in rivers and coastal oceans is described. It transmits turbidity measurements and position data via a satellite uplink to a processing server. The primary purpose of the instrument is to help document changes in sediment runoff from river catchments in North Queensland, Australia. The 'river drifter' is released into a flooded river and drifts downstream to the ocean, measuring turbidity at regular intervals. Deployment in the Herbert River showed a downstream increase in turbidity, and thus suspended sediment concentration, while for the Johnstone River there was a rapid reduction in turbidity where the river entered the sea. Potential stranding along river banks is a limitation of the instrument. However, it has proved possible for drifters to routinely collect data along 80 km of the Herbert River. One drifter deployed in the Fly River, Papua New Guinea, travelled almost 200 km before stranding., (Copyright © 2014 Elsevier Ltd. All rights reserved.)

Water turbidity and suspended sediment concentration (SSC) are commonly used as part of marine monitoring and water quality plans. Current management plans utilise threshold SSC values derived from mean-annual turbidity concentrations. Little published work documents typical ranges of turbidity for reefs within open coastal waters. Here, time-series turbidity measurements from 61 sites in the Great Barrier Reef (GBR) and Moreton Bay, Australia, are presented as turbidity exceedance curves and derivatives. This contributes to the understanding of turbidity and SSC in the context of environmental management in open-coastal reef environments. Exceedance results indicate strong spatial and temporal variability in water turbidity across inter/intraregional scales. The highest turbidity across 61 sites, at 50% exceedance (T50) is 15.3 NTU and at 90% exceedance (T90) 4.1 NTU. Mean/median turbidity comparisons show strong differences between the two, consistent with a strongly skewed turbidity regime. Results may contribute towards promoting refinement of water quality management protocols., (Copyright © 2013 Elsevier Ltd. All rights reserved.)

A statistical model for predicting monthly Dengue Hemorrhagic Fever (DHF) cases from the city of Makassar is developed and tested. The model uses past and present DHF cases, climate and meteorological observations as inputs. These inputs are selected using a stepwise regression method to predict future DHF cases. The model is tested independently and its skill assessed using two skill measures. Using the selected variables as inputs, the model is capable of predicting a moderately-severe epidemic at lead times of up to six months. The most important input variable in the prediction is the present number of DHF cases followed by the relative humidity three to four months previously. A prediction 1-6 months in advance is sufficient to initiate various activities to combat DHF epidemic. The model is suitable for warning and easily becomes an operational tool due to its simplicity in data requirement and computational effort.

Sediment accumulation rate is a frequently required parameter in environmental and management studies, in particular near coral reefs where sediment accumulation can potentially cause severe impact. However, opportunities to obtain accurate sediment accumulation measurements are often limited by a lack of adequate instrumentation, in particular for high temporal resolution monitoring. For instance the traditional use of sediment traps, as the most widespread technique, offers poor temporal resolution (commonly of weeks) besides having significant hydrodynamic shortcomings. Therefore, a new optical backscatter sediment accumulation sensor (SAS) was developed to continuously measure in situ short-term sediment accumulation in sensitive riverine and coastal environments, enabling high temporal and vertical resolution (order of 1 h and with a deposited thickness resolution in the order of 20 microm respectively). This allows investigations of various parameters that influence accumulation: tides, current, waves, rain, or anthropogenic activity such as sediment dumping. This paper briefly describes the SAS and presents three field applications on nearshore coral reefs at Ishigaki Island (Japan), Lihir Island (Papua New Guinea), and Magnetic Island (Australia).

Coastal development activities can cause local increases in turbidity and sedimentation. This study characterises the spatial and temporal variability of turbidity near an inshore fringing coral reef in the central Great Barrier Reef, under a wide range of natural conditions. Based on the observed natural variability, we outline a risk management scheme to minimise the impact of construction-related turbidity increases. Comparison of control and impact sites proved unusable for real-time management of turbidity risks. Instead, we suggest using one standard deviation from ambient conditions as a possible conservative upper limit of an acceptable projected increase in turbidity. In addition, the use of regional weather forecast as a proxy for natural turbidity is assessed. This approach is simple and cheap but also has limitations in very rough conditions, when an anthropogenic turbidity increase could prove fatal to corals that are already stressed under natural conditions.

An extensive sediment transport survey took place at Lihir Island (Papua New Guinea), where mining operations involve disposal of waste rocks and soil in nearshore waters. To investigate the potential impact of these practices over neighbouring fringing reefs, turbidity and sediment accumulation were measured continuously for extended periods. Turbidity records provided a map of observed impact zones based on turbidity thresholds. The main zoning features were (a) that an extreme turbidity gradient persists between the inner harbour (turbidity levels of 100-1000 mg l(-1)) and the adjacent reefs (turbidity levels in the order of 10 mg l(-1)), and (b) that observed zones conform with pre-operations impact predictions. Accumulation measurements unveiled no significant sediment accumulation over fringing coral reefs. This study contributes to the understanding of the potential impact of sediment discharge to nearshore waters.