Given the increasing pressures and perturbations on ecosystem due to climatic variability, there has been a developing interest in determining the resilience of ecosystems, particularly given the potential of abrupt and possibly irreversible shifts between alternative ecosystem states. There are numerous conceptual definitions of resilience in environmental systems and, even when resilience is clearly described for a particular ecosystem, it can challenging to quantify a priori to a state change in the ecosystem of interest. Ecosystems that approach transition exhibit generic changes in dynamical behaviour that can be used to signal the approach of a critical transition. When an ecosystem approaches a critical transition, its dynamics “slowdown”, and start exhibiting properties associated to the process critical slowing down (CSD), such as increased recovery time, variance, and autocorrelation. Here we develop a set of analytical methods based on measures of variance and autocorrelation to determine the current state of ecosystem and their likelihood to be at CSD, so to demonstrate how to operationalise what to date has been developed as a theoretical framework. We use wavelets as a measure of identifying changes in the variance term, and autocorrelation was modelled using an Bayesian dynamic linear model. We illustrate this on three case studies; i) on an experimental soil system subjected to dry wet cycles; ii) on an intensely managed ecological system in the Everglades and iii) on extensively managed grassland systems in the UK. We illustrate that although any of the resilience characteristics can be used to define resilience, the identified properties of a resilient response must be described for different contexts.

There is growing interest in understanding resilience of ecosystems because of the potential of abrupt and possibly irreversible shifts between alternative ecosystem states. Tipping points are observed in systems with strong positive feedback, providing early warning signals of potential instability. These points can be detected through metrics like critical slowing down (CSD), such as increased recovery time, variance, and autocorrelation. These indicators have been tested in laboratory experiments and field settings, ignoring trait changes. Here we present a long-term temporal analysis of several large, intensely monitored constructed wetlands, the Everglades Stormwater Treatment Areas (STAs), in which sudden changes in plant community composition have been observed. Using wavelet analysis, significant increases and decreases of variance properties (long-term flow data, water quality and nutrient TP loads) across these systems can indicate when and which STAs are less resilient to perturbations. In this study, continuous wavelet transform (CWT) was used to determine the periodicity of any cyclical activity in the data and to determine changes in autocorrelation and variance as measures of CSD. The change detection methods were used to find significant changes in variations and correlations across the time series. By employing these techniques, we were able to spot substantial shifts in model-observed wavelet correlation and model residual wavelet variance and thereby identify where these systems exhibit CSD. Although our analysis is limited to historical data, the proposed approach has practical value in that it identifies STAs that may be vulnerable to perturbation. The study also presents one of the few studies in which CSD is observed in practice rather than modelled in theory.

Sugarcane is a high-impact crop used in the majority of global sugar production, with India being the second largest global producer. Understanding the timing and length of sugarcane growth stages is critical to improving the sustainability of sugarcane management. Earth observation (EO) data have been shown to be sensitive to the variation in sugarcane growth, but questions remain as to how to reliably extract sugarcane phenology over wide areas so that this information can be used for effective management. This study develops an automated approach to derive sugarcane growth stages using EO data from Landsat-8 and Sentinel-2 satellite data in the Indian state of Andhra Pradesh. The developed method is then evaluated in the State of Telangana. Normalised difference vegetation index (NDVI) EO data from Landsat-8 and Sentinel-2 were pre-processed to filter out clouds and to harmonise sensor response. Pixel-based cloud filtering was selected over filtering by scene in order to increase the temporal frequency of observations. Harmonising data from two different sensors further increased temporal resolution to 3–6 days (70% of sampled fields). To automate seasonal decomposition, harmonised signals were resampled at 14 days, and low-frequency components, related to seasonal growth, were extracted using a fast Fourier transform. The start and end of each season were extracted from the time series using difference of Gaussian and were compared to assessments based on visual observation for both Unit 1 (R2 = 0.72–0.84) and Unit 2 (R2 = 0.78–0.82). A trapezoidal growth model was then used to derive crop growth stages from satellite-measured phenology for better crop management information. Automated assessments of the start and the end of mid-season growth stages were compared to visual observations in Unit 1 (R2 = 0.56–0.72) and Unit 2 (R2 = 0.36–0.79). Outliers were found to result from cloud cover that was not removed by the initial screening as well as multiple crops or harvesting dates within a single field. These results demonstrate that EO time series can be used to automatically determine the growth stages of sugarcane in India over large areas, without the need for prior knowledge of planting and harvest dates, as a tool for improving sustainable production.

Land cover (LC) maps are used extensively for nature conservation and landscape planning, but low spatial resolution and coarse LC schemas typically limit their applicability to large, broadly defined habitats. In order to target smaller and more-specific habitats, LC maps must be developed at high resolution and fine class detail using automated methods that can efficiently scale to large areas of interest. In this work, we present a Machine Learning approach that addresses this challenge. First, we developed a multi-stage semantic segmentation approach that uses Convolutional Neural Networks (CNNs) to classify LC across the Peak District National Park (PDNP, 1439 km2) in the UK using a detailed, hierarchical LC schema. High-level classes were predicted with 95% accuracy and were subsequently used as masks to predict low-level classes with 72% to 92% accuracy. Next, we used these predictions to analyse the degree and distribution of fragmentation of one specific habitat—wet grassland and rush pasture—at the landscape scale in the PDNP. We found that fragmentation varied across areas designated as primary habitat, highlighting the importance of high-resolution LC maps provided by CNN-powered analysis for nature conservation.

Potato black dot is a foliar and tuber blemish disease that has become an increasingly economic problem in recent years. Black dot is caused by the fungus Colletotrichum coccodes and is characterised by silver/brown lesions on the tuber skin leading to lower aesthetic quality of potatoes destined for the pre-pack market. Given the consumers' growing demand for washed and pre-packed potatoes, skin blemish diseases (such as black dot and silver scurf), once considered of minor importance, are now serious challenges for the fresh potato industry. The management of C. coccodes is far from satisfactory at either pre- or postharvest stages: firstly, the disease symptoms have not been consistently described on potato plant foliage; and secondly, black dot disease is often confounded with other tuber blemishes during postharvest storage. Good field managing practices in combination with improved postharvest strategies and an accurate detection support tool can be a useful integrated approach to manage potato black dot disease. This review aims to evaluate and critically discuss different novel approaches for better management and detection of potato black dot disease.

Understanding the relationship between land use and opium production is critical for monitoring the dynamics of poppy cultivation and developing an effective counter narcotics policy in Afghanistan. However, mapping agricultural land accurately and rapidly is challenging, as current methods require resource-intensive and time consuming manual image-interpretation. Deep convolutional neural nets have been shown to greatly reduce the manual effort in mapping agriculture from satellite imagery but require large amounts of densely labelled training data for model training. Here we develop a generalised model using past images and labels from different medium resolution satellite sensors for fully automatic agricultural land classification using the latest medium resolution satellite imagery. The model (FCN-8) is first trained on Disaster Monitoring Constellation (DMC) satellite images from 2007 to 2009. The effect of shape, texture and spectral features on model performance are investigated along with normalisation in order to standardise input medium resolution imagery from DMC, Landsat-5, Landsat-8, and Sentinel-2 for transfer learning between sensors and across years. Textural features make the highest contribution to overall accuracy (∼73%) while the effect of shape is minimal. The model accuracy on new images, with no additional training, is comparable to visual image interpretation (overall > 95%, user accuracy > 91%, producer accuracy > 85%, and frequency weighted intersection over union > 67%). The model is robust and was used to map agriculture from archive images (1990) and can be used in other areas with similar landscapes. The model can be updated by fine tuning using smaller, sparsely labelled datasets in the future. The generalised model was used to map the change in agricultural area in Helmand Province, showing the expansion of agricultural land into former desert areas. Training generalised deep learning models using data from both new and long-term EO programmes, with little or no requirement for fine tuning, is an exciting opportunity for automating image classification across datasets and through time that can improve our understanding of the environment.

Rivers are dynamic landscape features that change in response to natural and anthropogenic factors through hydrological, geomorphic and ecological processes. The severity and magnitude of human impacts on river system and riparian vegetation has dramatically increased over the last century with the proliferation of valley-spanning dams, intensification of agriculture, urbanization, and more widespread channel engineering. This study aims to determine how changes in geomorphic form and dynamics caused by these human alterations relate to changes in channels and riparian vegetation in the lower Beas and Sutlej Rivers. These rivers are tributaries of the Indus that drain the Western Himalayas but differ in the type and magnitude of geomorphic change in recent decades. Winter season vegetation was analysed over 30 years, revealing increasing trends in vegetated land cover in the valleys of both rivers, consistent with large-scale drivers of change. Greater trends within the active channels indicate upstream drivers are influencing river flow and geomorphology, vegetation growth and human exploitation. The spatial patterns of vegetation change differ between the rivers, emphasizing how upstream human activities (dams and abstraction) control geomorphic and vegetation community response within the landscape context of the river. The increasing area of vegetated land is reinforcing the local evolutionary trajectory of the river planform from wide-braided wandering to single thread meandering. Narrowing of the active channels is altering the balance of resource provision and risk exposure to people. New areas being exploited for agriculture are exposed to greater risk from river erosion, inundation, and sediment deposition. Moreover, the change from braided to meandering planform has concentrated erosion on riverbanks, placing communities and infrastructure at risk. By quantifying and evaluating the spatial variations in vegetation cover around these rivers, we can better understand the interaction of vegetation and geomorphology alongside the impacts of human activity and climate change in these, and many similar, large systems, which can inform sustainable development.

Afghanistan's annual opium survey relies upon time-consuming human interpretation of satellite images to map the area of potential poppy cultivation for statistical sample design. Deep Convolutional Neural Networks (CNNs) have shown ground-breaking performance for image classification tasks by encoding local contextual information, in some cases outperforming trained analysts. In this study, we investigate the development of a CNN to automate the classification of agriculture from medium-resolution satellite imagery as an alternative to manual interpretation. The residual network (ResNet50) CNN architecture was trained and validated for delineating the agricultural area using labelled multi-seasonal Disaster Monitoring Constellation (DMC) satellite imagery (32 m) of Helmand and Kandahar provinces. The effect of input image chip size, training sampling strategy, elevation data, and multi-seasonal imagery were investigated. The best-performing single-year classification used an input chip size of 33 x 33 pixels, a targeted sampling strategy and transfer learning, resulting in high overall accuracy (94%). The inclusion of elevation data marginally lowered performance (93%). Multi-seasonal classification achieved an overall accuracy of 89% using the previous two years' data. Only 25% of the target year's training samples were necessary to update the model to achieve >94% overall accuracy. A data-driven approach to automate agricultural mask production using CNNs is proposed to reduce the burden of human interpretation. The ability to continually update CNN models with new data has the potential to significantly improve automatic classification of vegetation across years.

Earth and Space Science Open Archive PosterOpen AccessYou are viewing the latest version by default [v1]Long-term dynamics of riparian and in-channel vegetation cover on the Beas and Sutlej Rivers, India.AuthorsJohnBealeiDRobertGrabowskiPaulineLokidoriDKimVercruysseDanielSimmsSee all authors John BealeiDCorresponding Author• Submitting AuthorCranfield UniversityiDhttps://orcid.org/0000-0003-4154-8032view email addressThe email was not providedcopy email addressRobert GrabowskiCranfield Universityview email addressThe email was not providedcopy email addressPauline LokidoriDCoventry UniversityiDhttps://orcid.org/0000-0001-8871-0135view email addressThe email was not providedcopy email addressKim VercruysseJoin For Water vzwview email addressThe email was not providedcopy email addressDaniel SimmsCranfield Universityview email addressThe email was not providedcopy email address

<p>Accurate mapping of agricultural area is essential for Afghanistan’s annual opium poppy monitoring programme. Access to labelled data remains the main barrier for utilising deep learning from satellite imagery to automate the process of land cover classification. In this study, we aim to transfer knowledge from historical labelled data of agricultural land, from work on poppy cultivation estimates undertaken between 2007 and 2010, to classify imagery from a range of sensors using deep learning. Fully Convolutional Networks (FCNs) have been used to learn the complex features of agriculture in southern Afghanistan using their inherent spatial and spectral characteristics from satellite imagery. FCNs are trained and validated using labelled Disaster Monitoring Constellation (DMC) data (32 m) to transfer knowledge of agricultural land to classify other imagery, such as Landsat (30 m). The dependency on spatial and spectral characteristics are explored using intensity, Normalised Difference Vegetation Index (NDVI), top of atmosphere reflectance and tasselled cap transformation. The underlying spatial features associated with agriculture are found to play a significant role in agriculture discrimination. High classification performance has been achieved with over 92% overall accuracy and 0.58 intersection over union. The ability to transfer knowledge from historical datasets to new satellite sensors is an exciting prospect for future automated agricultural land discrimination in the United Nations Office on Drugs and Crime annual opium survey.</p>

The ground breaking performance of fully convolutional neural nets (FCNs) for semantic segmentation tasks has yet to be achieved for landcover classification, partly because a lack of suitable training data. Here the FCN8 model is trained and evaluated in real-world conditions, so called in-the-wild, for the classification of opium poppy and cereal crops at very high resolution (1 m). Densely labelled image samples from 74 Ikonos scenes were taken from 3 years of opium cultivation surveys for Helmand Province, Afghanistan. Models were trained using 1 km2 samples, sub-sampled patches and transfer learning. Overall accuracy was 88% for a FCN8 model transfer-trained on all 3 years of data and complex features were successfully grouped into distinct field parcels from the training data alone. FCNs can be trained end-to-end using variable sized input images for pixel-level classification that combines the spatial and spectral properties of target objects in a single operation. Transfer learning improves classifier performance and can be used to share information between FCNs, demonstrating their potential to significantly improve land cover classification more generally.

The United Nations Office on Drugs and Crime and the US Government make extensive use of remote sensing to quantify and monitor trends in Afghanistan’s illicit opium production. Cultivation figures from their independent annual surveys can vary because of systematic differences in survey methodologies relating to spectral stratification and the addition of a pixel buffer to the agricultural area. We investigated the effect of stratification and buffering on area estimates of opium poppy using SPOT5 imagery covering the main opium cultivation area of Helmand province and sample data of poppy fields interpreted from very high resolution satellite imagery. The effect of resolution was investigated by resampling the original 10 m pixels to 20, 30, and 60 m, representing the range of available imagery. The number of strata 1, 4, 8, 13, 23, 40 and sample fraction 0.2–2% used in the estimate were also investigated. Stratification reduced the confidence interval by improving the precision of estimates. Cultivation estimates of poppy using 40 spectral strata and a sample fraction of 1.1% had a similar precision to direct expansion estimates using a 2% sample fraction. Stratified estimates were more robust to changes in sample size and distribution. The mapping of the agricultural area had a significant effect on poppy cultivation estimates in Afghanistan, where the area of total agricultural production can vary significantly between years. The findings of this research explain differences in cultivation figures of the opium monitoring programmes in Afghanistan and recommendations can be applied to improve resource monitoring in other geographic areas.

Artisanal gold mining (galamsey) and cocoa farming are essential sources of income for local populations in Ghana. Unfortunately the former poses serious threats to the environment and human health, and conflicts with cocoa farming and other livelihoods. Timely and spatially referenced information on the extent of galamsey is needed to understand and limit the negative impacts of Mining. To address this, we use multi-date UK-DMC2 satellite images to map the extent and expansion of galamsey from 2011 to 2015. We map the total area of galamsey in 2013 over the cocoa growing area, Using k-means clustering on a cloud-free 2013 image with strong spectral contrast between galamsey and the surrounding vegetation. We also process a pair of hazy images from 2011 and 2015 with Multivariate Alteration Detection to map the 2011-2015 galamsey expansion in a subset, labelled the change area. We use a set of visually interpreted random sample points to compute bias-corrected area estimates. We also delineate an indicative impact zone of pollution proportional to the density of galamsey, assuming a maximum radius of 10 km. In the cocoa growing area of Ghana, the estimated total area of galamsey in 2013 is 27,839 ha with an impact zone of 551,496 ha. In the change area, galamsey has more than tripled between 2011 and 2015, resulting in 603 ha of direct encroachment into protected forest reserves. Assuming the same growth rate for the rest of the cocoa growing area, the total area of galamsey in 2015 is estimated at 43,879 ha. Galamsey is developing along most of the river network (Offln, Ankobra, Birim, Anum, Tano), with downstream pollution affecting both land and water. (C) 2017 Elsevier B.V. All rights reserved.

Yearly estimates of illicit opium production are key metrics for assessing the effectiveness of counter narcotics policy in Afghanistan. Poor security often prevents access to sample locations and puts pressure on field surveyors, resulting in biased sampling and errors in data recording. Supportive methods using aerial digital photography for improving yield estimates were investigated in the UK in 2004, 2005 and 2010. There were good empirical relationships between NDVI and poppy yield indicators (mature capsule volume and dry capsule yield) for individual fields. The results suggested a good generalised relationship across all sampled fields and years (R2>0.70) during the 3–4 week period including poppy flowering. Regression estimates using this relationship with the imagery counteracted bias in the sample estimate of yield, reduced sample error and enabled the production of detailed maps showing the poppy yield distribution. The application of this approach using VHR satellite imagery was investigated in the context of the annual opium survey in Afghanistan. Initial results indicated the potential for bias correction of yield estimates using a smaller and targeted collection of ground observations as an alternative to random sampling.

The image-interpretation of opium poppy crops from very high resolution satellite imagery forms part of the annual Afghanistan opium surveys conducted by the United Nations Office on Drugs and Crime and the United States Government. We tested the effect of generalization of field delineations on the final estimates of poppy cultivation using survey data from Helmand province in 2009 and an area frame sampling approach. The sample data was reinterpreted from pan-sharpened IKONOS scenes using two increasing levels of generalization consistent with observed practice. Samples were also generated from manual labelling of image segmentation and from a digital object classification. Generalization was found to bias the cultivation estimate between 6.6% and 13.9%, which is greater than the sample error for the highest level. Object classification of image-segmented samples increased the cultivation estimate by 30.2% because of systematic labelling error. Manual labelling of image-segmented samples gave a similar estimate to the original interpretation. The research demonstrates that small changes in poppy interpretation can result in systematic differences in final estimates that are not included within confidence intervals. Segmented parcels were similar to manually digitized fields and could provide increased consistency in field delineation at a reduced cost. The results are significant for Afghanistan’s opium monitoring programmes and other surveys where sample data are collected by remote sensing.

Yearly estimates of illicit opium production are key metrics for assessing the effectiveness of the counter-narcotics policy in Afghanistan. Poor security often prevents access to sample locations and puts pressure on field surveyors, resulting in biased sampling and errors in data recording. Supportive methods using aerial digital photography for improving yield estimates were investigated in the UK in 2004, 2005, and 2010. There were good empirical relationships between normalized difference vegetation index and poppy yield indicators (mature capsule volume and dry capsule yield) for individual fields. The results suggested a good generalized relationship across all sampled fields and years (R-2 > 0.70) during the 3-4 week period including poppy flowering. Regression estimates using this relationship with the imagery counteracted bias in the sample estimate of yield, reduced sample error, and enabled the production of detailed maps showing the poppy yield distribution. The application of this approach using very-high-resolution satellite imagery was investigated in the context of the annual opium survey in Afghanistan. Initial results indicated the potential for bias correction of yield estimates using a smaller and targeted collection of ground observations as an alternative to random sampling.

We investigated and developed a prototype crop information system integrating 250 m Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) data with other available remotely sensed imagery, field data, and knowledge as part of a wider project monitoring opium and cereal crops. NDVI profiles exhibited large geographical variations in timing, height, shape, and number of peaks, with characteristics determined by underlying crop mixes, growth cycles, and agricultural practices. MODIS pixels were typically bigger than the field sizes, but profiles were indicators of crop phenology as the growth stages of the main first-cycle crops (opium poppy and cereals) were in phase. Profiles were used to investigate crop rotations, areas of newly exploited agriculture, localized variation in land management, and environmental factors such as water availability and disease. Near-real-time tracking of the current years' profile provided forecasts of crop growth stages, early warning of drought, and mapping of affected areas. Derived data products and bulletins provided timely crop information to the UK Government and other international stakeholders to assist the development of counter-narcotic policy, plan activity, and measure progress. Results show the potential for transferring these techniques to other agricultural systems.

Soil and landform information is needed for a wide range of applications but available data are often inaccessible, incomplete, or out of date. Within the European FP7 project 'e-SOTER', following recognised harmonisation principles, we developed an XML schema to serve as an exchange format for soil and terrain data derived from e-SOTER methodologies (SoTerML). It encompasses existing SOTER database conceptual modelling as well as the WRB (World Reference Base of soil resources) and the FAO soil data structures and classifications, therefore covering major soil and terrain databases such as the European Soil Database (ESD). The flexibility of the modelling achieved is demonstrated from legacy data integrated in the new scheme and made available using an OGC Web Feature Service. Along with the description of SoTerML, the paper aims at pointing out the modelling approach and the modelling principles used for soil and terrain observations, which extracted from our proposal, could prove useful for emerging initiatives towards defining an exchange standard framework for the soil theme.