As a result, each single pixel value of the calculated LAI, NDVI, GNDVI, EVI, RVI, TGI, CVI and NGRDI layers and the original 10 m resolution Sentinel-2 bands were associated to one GY value. This processing protocol was carried out at each phenological stage. An overview of few indices calculated is shown in Fig. 3. A total number of 20,124 pixels with their corresponding unique GY values and Sentinel-2 derived data was obtained as data set. The raw data was manually buffered to avoid edge effects of mix pixels as shown in Fig. 4. All the data was matched in a point vector file, it was easier to process and use as data is organized in attribute tables. The processing was carried out on ArcGIS Pro 2.3.0 and a schematic illustration describing the process is shown in Fig. 4. The cut width of the combine harvester was 7.60 m and therefore adequate to 10 m pixels size of Sentinel-2 imagery. Notwithstanding its suitability for Sentinel-2 pixel size, the GY spatial resolution of the combine harvester, as it is a function of the monitoring equipment, the cutting head and the software has some lack of precision, as reported by other authors . All the GPS GY points with values lower to 0.5 t/ha were discarded. Moreover, a portion of the raw data was trimmed using the “tclust” package in R based on GY and Sentinel-2 bands inaccuracies. Trimming allows the removal of the most outlying data which can strongly influence the results; this approach has been introduced by several authors . The number of clusters and portion of data trimmed was optimized and subsequently applied to remove outliers . The general process is sh own in Fig. 5. Regarding question two we observed that multi-date data improves the capacity to estimate within-field grain yield variability. In all three data feeding instances , R2 was higher and RMSE lower when using multi-date Sentinel-2 data . Yet, the full temporal resolution of Sentinel-2 images could not be exploited as the availability of images for the regions studied was limited throughout the three seasons due to cloud cover. Despite of the limitations, several images throughout the season could be analysed following the phenological stages stem elongation and heading/anthesis. Interestingly, LAI assessed at stem elongation and heading/anthesis predicted better GY than VIs and Sentinel-2 bands. This is coherent with the fact that LAI is a good indicator of potential canopy photosynthesis, potted blueberries with these reproductive stages being crucial in defining yield . In the case of the VIs and eventually of the single bands they may reflect later ; particularly in terms of stay green and the onset of crop senescence during grain filling .

Decrease in Stay Green, as response for example of water stress or lack of mineral nutrients effects may affect GY through an accelerated reduction of photosynthetic assimilates during the grain filling , which might be too late for a reasoned management . Moreover, LAI and VIs in wheat are frequently weakly correlated, which means their performance predicting yield are not necessarily comparable . In this sense, for single phenological stages, the efficiency of the models was relatively high. Hence the estimation of GY before the whole season is completed, around March to May can help farmers to guide management decisions and ensure a better harvest. Several authors have matched Sentinel-2 images with phenology to estimate wheat GY at various resolutions that allows an improved opportunity for precision farming management. The technical progress of Sentinel-2 and the increasing development of modelling approached for GY estimation now makes precision farming affordable and cost-effective and thus, almost operational . Regarding the third question we observed that RF clearly outperformed BR, SVM and multilinear regression . In contrast with this, other authors observed an improved performance of BR in comparison with RF regression, when using VIs. Nonetheless, we can argue that the findings of this article show RF as an improved modelling approach when using LAI to estimate within field grain yield, as discussed previously. In contrast with other authors who did not find improvements between simple linear regressions and ML , we found that the relationship between crop yield and reflectance is complex enough for ML approaches and results in improvements in within-field yield predictions. RF is less prone to outliers, and hence one would expect yield estimation performance to increase, as we have corroborated here. Besides, RF algorithm is powerful in handling both linear and non-linear relationship as wheat yield and spectral information might have some non-linearity . In general, we observed that the modelling approach is central for an effective GY estimation. We can argue that the modelling is the most important factor, followed by the sensing date and the processing of the spectral data . In an equivalent study developed in the UK, RF was also used and compared to simple regression but no other machine learning approaches were presented . The results here obtained confirm the effectiveness of RF regression to estimate within-field GY variability using Sentinel-2 imagery for the case of Spain. Nonetheless, a further research and comparison of ML approaches could be developed aiming to standardize methodologies for specific regions and crop-specific cases for this almost operational precision agriculture technology. The ability of RF to cope with multivariate relationships between data of different types and resolutions is a key advantage over methods such as linear regression, which can only address univariate relationships.

To our knowledge, hitherto advanced models using ML approaches to estimate within-field GY in cereals have focused on demonstrations of single machine learning models and VIs or simple regressions . Overall, there are a handful of articles dealing with GPS combine harvester and Sentinel-2 images and none have assessed several ML approaches, spectral indices and biophysical parameters retrieved from RTM. Hence, the results here presented assessing several ML approaches and Sentinel- 2 data processing approaches are a novelty for this growing research field. In this sense, we believe that the methodology here developed, and the results obtained can contribute to precision farming. First, the ML approaches analysed present RF as an efficient model to monitor within field wheat GY. The use of ML to retrieve GY has been considered one of the most important areas develop associated with remote sensing and agriculture . On top of that, based on the results, we can argue that retrieving biophysical parameters with RTM that can be linked to crop traits and subsequently used for ML models is a step forward in precision agriculture. The usage of RTMs with ML regression algorithms is opening up a powerful and promising field of vegetation properties retrieval from EOS data , but has yet to be fully explored in precision agriculture. LAI, leaf structure parameters, leaf chlorophyll content, leaf carotenoid content, leaf anthocyanin content or leaf equivalent water thickness are, among others, traits of agricultural crops that can be retrieved from RTM . The availability of biophysical processors in software such as SNAP of the European Space Agency used here opens a door to the applicability of these parameters. In this sense, with increased computing power available in the next years, RTM inversion can provide an elegant alternative to estimate fertilizer requirements in precision agriculture . Furthermore, RTM have shown the potential of Sentinel-2 data to map disease-incidence dynamics in agriculture plots . Regarding irrigation, cropland canopy water content thematic layers have been developed with RTM and Sentinel-2 imagery using several parameters such as water thickness or dry matter content, among others. In addition to this, the results obtained in this study with single image acquisition and its matching with phenological stage can help to locate low-yielding spots in croplands and subsequently apply precise farming decisions. In this sense, square plastic pot the performance of LAI and other biophysical parameters of interest for precision agriculture retrieved from RTM could contribute to find physiological or agronomic explanations and apply reasoned managements. Our study also has some limitations to overcome such as its applicability in other crops and in different agroclimates.

The aquaculture industry accounts for 46% of production of seafood per annum and is seen as a strategically vital sector for sustainable Blue Growth . However, in the EU, including the UK, production is stagnating. Reasons for this include: conflicts with other marine users, licensing issues, the carrying capacity of the marine environment and a lack of social license to operate . There is increasing awareness that site-scale opposition to aquaculture operations by local communities and communities of interest, is causing conflict . Through the concept of social license to operate, this study takes an exploratory, in-depth approach to understanding community perceptions of fin-fish aquaculture in Scotland.The term social license to operate first emerged from the mining industry, coined by James Cooney.The use of the concept has now expanded from the mining sector, and has been applied across other extractive sectors such as forestry and oil and gas, as well as environmental governance practises across both terrestrial and marine environments . Many definitions of SLO have since arisen, but one that encompasses the breadth of SLO comes from Moffat and Zhang, who describe it as “the ongoing acceptance and approval of a project by local community members and other stakeholders”. SLO is granted by communities , however who makes up community and the influences on communities are often context specific. The extent to which communities can grant social license is related to the economic power that both communities and industry have, mediated by the states and governments, and under the influences of media and cultural norms . Legitimacy, credibility and trust have been discussed as central elements of SLO . With these concepts often used to model a framework of social license, where legitimacy forms the gatekeeper of SLO and the three concepts build upon each other to create more stable SLO . These components provide a good overview of the multitude of components that have been suggested to be part of SLO. Reputation, transparency and permission are posited as vital for SLO but these arguably form a part of the wider legitimacy of an industry . Mutual understanding has also been suggested as an important part of SLO, but this is also an essential component in relationships with high levels of trust . Further critiques of three component models, and SLO in general, highlight that the concepts must be examined in the context of power dynamics, the role that governing state play in mediating SLO and communities access to knowledge . There is argument that due to the overlap of SLO with a number of acceptability concepts, SLO loses its utility, however the focus upon local and community scales continues to make SLO an important concept to consider . The ambiguity of the term allows for greater power for communities by letting them define it, versus say corporate social responsibility, which is most often defined by industries . The focus that it shines upon local scales is also vital in making sure that local contexts and values are explored and considered . Finer scale focus also highlights specific action situations, elucidating the roles various bodies play in such situations. SLO is arguably a useful framing to apply to the aquaculture industry, as it arises from a need to measure private industry and community relations, whilst also approaching it from a community perspective that acknowledges the contextual and historical factors that can influence these relationships. It has been suggested that applying SLO to the aquaculture industry maybe more complex than its original application in the mining sector , as often seas are seen as a common pool resource, with inherent access rights and with little private property rights . Often aquaculture sites require the creation of private space through ownership or leasing of the seabed. This has serious ramifications for local communities, especially in spaces that are used by indigenous peoples, whose claim to the land and coastline are of cultural significance .