Monitoring coral reef and seagrass ecosystems is essential for managing the health of coastal marine environments, requiring detailed information on benthic composition and extent. Traditional in situ monitoring programs are critical for assessing reef health, but are limited in scale, covering only small areas, while conventional remote sensing approaches extend monitoring coverage scale, but provide limited benthic composition detail. Hyperspectral remote sensing has emerged as a solution, offering detailed spectral information capable of discerning diverse benthic classes across increased spatial extents. This review examines recent literature on hyperspectral remote sensing applications for monitoring the composition and spatial extent of coral reefs and seagrass meadows. Research demonstrates that important composition features present distinct spectral signatures, enabling separation of coral from turf algae and seagrass from macroalgae at pixel sizes below 5 m. However, spectral mixing complicates classification at larger pixel sizes, and the use of satellites remains relatively untested. While hyperspectral remote sensing can technically map benthic composition across large spatial scales (>100 km2), several limitations impede widespread adoption, including: lack of management relevant spatial data, restricted technology access, spectral mixing challenges and absence of standardized workflows and spectral libraries. A critical transition is needed from basic mapping exercises toward monitoring functional ecosystem health aspects through derived metrics with direct management relevance. Hyperspectral technology offers strong potential for mapping coral reef and seagrass habitat composition and extent, but achieving routine monitoring value demands continued methodological refinement, improved data accessibility, and enhanced integration into ecological frameworks.

Accurate tree volume and structure are crucial for forest biomass estimation and ecosystem investigations. While terrestrial laser scanning (TLS) offers non-destructive pathways for the detailed three-dimensional tree recon struction, current methods overestimate small branch volumes and often require tree segmentation and leaf-wood separation as a priori. This study introduces and validates RayExtract, a novel method for reconstructing woody volume from TLS data, utilising tools from the RayCloudTools library, to automate the extraction of tree struc tural metrics from point clouds. Our method incorporates two key morphological rules - Self-Similarity and Leonardo's Rule - to aid branch radius and taper calculations. Likewise, it enables rapid and automated plot-scale reconstruction by integrating tree segmentation and woody structure modelling without requiring leaf point classification. In this study, RayExtract demonstrated high accuracy across four high-quality destructive harvest reference sets with concordance correlation coefficient (CCC) values ranging from 0.82 to 0.97 (n=124). To explore algorithm behaviours under different leaf conditions and point densities, we implement a framework using TLS simulation of highly realistic synthetic trees. Results from the simulation framework show consistent high accuracy of total woody volume, with CCC ranging from 0.97 to 0.98 (n=18) across four distinct scan ning configurations. Fine-scale volumetric analysis revealed that incorporating simple morphological rules can effectively inform branch taper and reduce woody volume overestimation, particularly in smaller components. Furthermore, it identifies a limitation in volumetric accuracy in trees exhibiting significant taper in the lower stem. Analysis of RayExtract's computational efficiency demonstrates that runtime and memory usage scale pre dictably with input data size, primarily driven by point count and the associated structural complexity within the point cloud, positioning the algorithm as well suited for large-scale applications. RayExtract represents a significant advancement in forest reconstruction, biomass estimation, and vegetation structural analysis. The method's efficiency, accuracy, and robustness across varied forest conditions mark a substantial improvement in forest structural assessment techniques using laser scanning and have broad implications for improving re gional biomass estimations, and contributing to the calibration and validation of broad-scale remote sensing observations.

Urbanisation is a global phenomenon, with major cities worldwide undergoing rapid transformation driven by economic and population growth. However, the urbanisation process in Brisbane City remains comparatively underexplored relative to other Australian metropolitan areas. This study presents a novel integrated framework that combines Google Earth Engine (GEE), Bayesian weight of evidence (WofE), and Artificial Neural Network–Cellular Automata (ANN–CA) to analyse and forecast spatiotemporal patterns of urban change. The research addresses three objectives: (i) monitoring urban settlements in Brisbane City from 1990 to 2021, (ii) identifying the drivers of urban expansion between 1990 and 2021, (iii) projecting future urban growth to 2030, and (iv) critically assessing the capability, strengths, and weaknesses of the ANN–CA model. Random forest classification using GEE achieved overall, producer, and user accuracies ranging from 0.97 to 1.00. The results revealed a period of stagnation and slight decline during 1990–2000, followed by accelerated expansion from 2000–2021, during which the urban area nearly doubled from 142.29 km2 to 307.41 km2. WofE analysis highlighted key determinants of urban growth, including distance to the central business district, proximity to waterways, roads, and points of interest, as well as topographic variables. ANN–CA simulations underscored the model’s sensitivity to underfitting, overfitting, and imbalanced change rates. Adaptive-period recalibration improved performance, with validation on 2021 dataset yielding 93.69

Coastal cities, such as the Gold Coast region in Australia, are experiencing rapid urbanisation, driven by population growth and the appeal of coastal living. However, this growth poses significant challenges, including environmental preservation and vulnerability to coastal hazards like flooding and erosion. Understanding how urban expansion interacts with shifting coastal boundaries is crucial for sustainable urban planning, particularly in the context of climate change. This study investigates the dynamics of urban growth and coastal development in the Gold Coast over the past three decades. Using temporal clustering, it analyses changes in population density and distance to the coastline, identifies shifts in risk exposure to coastal hazards, and explores the relationship between coastal dynamics and urban development.The findings reveal that urban growth is concentrated in established centres, such as Surfers Paradise and Currumbin, while peripheral areas experience slower development. These patterns are influenced by natural barriers, such as waterways, and socio-economic factors, including access to economic opportunities and tourism. The study emphasises the need for a nuanced, region-specific approach to urban planning that balances growth with environmental sustainability. It also highlights the role of advanced technologies, such as remote sensing, GIS, and geospatial intelligence, including digital twins, in supporting data-driven planning and resilience strategies in coastal environments. This research offers valuable insights for policymakers and urban planners addressing the challenges of coastal urbanisation.

Satellite earth observation (EO) data plays a vital role quantifying vegetation structural and functional metrics across spatio-temporal scales. However, the degree of coupling between satellite derived spectral signals and the rate of photosynthesis, as estimated by Gross Primary Productivity (GPP), both before and after bushfire remain understudied, yet these are a critical part of the global carbon cycle. This study evaluated a combination of passive optical and active LiDAR satellite data to quantify the disturbance and recovery of photosynthesis from a major fire event. The work was completed at the Tumbarumba long-term tall eucalypt flux site following a catastrophic bushfire in December 2019. TROPOMI solar-induced fluorescence (SIF) and Sentinel 2 derived greenness and burn severity metrics (NDVI, EVI, NIRv, and NBR) were investigated, termed 'spectral metrics' herewith. Detailed in-situ observations from leaf-to-canopy scales were utilised to examine variations in vegetation structural-functional parameters. We found the rate of vegetation spectral metrics recovery largely outpaced GPP recovery at the one- and twoyear post-fire mark. Specifically, SIF recovered to 80-90 % compared to pre-fire levels, whereas GPP recovered only 45-50 %. This indicated that separate SIF:GPP functions were required for pre- and post-fire data to account for different recovery trajectories due to changes in canopy structure and species composition. The use of TROPOMI SIF for monitoring canopy productivity at seasonal (monthly) time-scales was advantageous over traditional greenness-based indices, as SIF tracked GPP seasonality both pre- and post-fire. Spaceborne GEDI LiDAR data effectively captured post-fire changes in forest structure, albeit at sparse spatio-temporal sampling intervals, revealing a significant reduction in overstorey vegetation density and a concurrent increase in understorey vegetation density. This contributed to reduced carbon uptake, compared to pre-fire, due to the lower light use efficiency of understorey species, which was verified with in-situ gas exchange measurements. Overall, this study highlights the importance of accounting for disturbance history and the relative abundance of overstorey and understorey vegetation for tracking GPP from satellite platforms. Our results also highlight the crucial role of longitudinal field-based data for calibration and validation of EO data, ultimately enhancing our understanding of forest recovery processes.

The progressive rehabilitation of open-cut coal mines and the demonstration of a sustainable post-mining land use is required prior to mine closure in many parts of the world. There is a general belief that the risk of rehabilitation failure following mine closure due to disturbance events such as fire is minimal, based largely on the assumed resilience of rehabilitated pastures and the assumption that rehabilitated pastures will respond analogous to unmined pastures following fire. However, there is little scientific evidence to support this notion, and additional knowledge gaps on the resilience of rehabilitation age classes and appropriate methods for land managers to measure and demonstrate rehabilitation resilience. We used Sentinel-2 and Landsat-8 time-series and assessed the impact and vegetation response to eight fire events in rehabilitation and five fires in unmined analogues in central and southeast Queensland, Australia. Using the Soil Adjusted Vegetation Index (SAVI), we aimed to compare rehabilitated and unmined areas using three resilience metrics: i) percent impact, ii) recovery time and iii) recovery rate. Compared with unmined pastures, post-mine rehabilitation recorded higher mean impact (52-65 % vs 67-79 % respectively), longer recovery times (38-117 vs 144-245 days respectively) and a slower rate of recovery (2.5-5.7 % vs 0.9-1.7 % per day respectively). Younger age-classes (<10 y/o and 10-15 y/o) recorded reduced resilience compared to mature rehabilitation (>= 16 y/o). We compared three different baseline indices and showed that the choice of baseline index yielded significantly different results for the same fires, indicating the importance of standardised approaches to resilience monitoring.

A very detailed image, such as very high-resolution satellite imagery, is necessary for many applications. By enhancing the image’s detail and reducing its color distortion, we can optimize its use. Image fusion is a method to achieve this goal. This paper proposes a novel method of multi-sensor image fusion called Fusion+. This method enhances spatial resolution, sharpens the image using the kernel of image enhancement, and preserves radiometric consistency using the histogram specification. In addition, the Fusion+ has a shift adjustment between the input images to address the geometric issue, especially for images from multiple sensors. The proposed method can be used for both single-sensor and multi-sensor image fusion. The fused image obtained from the proposed method also increased the image’s detail and was sharper than the other methods. The universal image quality index, peak signal-to-noise ratio, and structural similarity index measure are also sufficient to maintain the radiometric consistency of the original image. The results also showed that the fused image had maximum detail increase, minimum color distortion, and a natural color appearance. Fusion+ obtained an average SSIM score of 0.779, an average PSNR of 29.342, and an average UQI of 0.973 on single-sensor fused images, as well as SSIM averages of 0.556, an average PSNR of 27.896, and an average UQI of 0.947 on multi-sensor fused images. The Fusion+ method demonstrates the potential for improved image fusion quality, which can contribute to improved accuracy and reliability in the analysis of very high-resolution remote sensing data, especially in multi-sensor applications.

Artisanal small-scale mining (ASM) is an environmentally damaging activity in many developing countries, particularly in the wet tropics, yet serves as a crucial economic resource for millions of people. The lack of effective mapping methods hinders quantifying the spatial extent of ASM and management efforts. This study presents a novel approach to integrate multi-spectral and imaging radar datasets within the Google Earth Engine (GEE) platform to map ASGM in a tropical rainforest. We used a case study of gold mining in central Kalimantan and diverse training and validation data sources. The methodology involved pre-processing multispectral and radar imagery, generating and standardizing covariates, applying feature-level data fusion for the Random Forest algorithm in GEE, and training and classifying data with optimized parameters through iterative loops. This approach achieved a classification accuracy of 81% in detecting ASM activities, surpassing the accuracy of a map constructed solely from Sentinel-2 multispectral data by 14%. Through the inclusion of evaluation metrics such as the f(β) score and Matthews Correlation Coefficient (MCC), our approach demonstrates its robustness in accurately identifying target instances, while reducing false positives and addressing imbalanced class sizes by 6.25% and 60%, respectively. Our model’s efficacy underscores its potential to accurately map ASM at larger regional scales (104 – 10⁶ km2) in wet-tropical forests, while being scalable and resource-efficient. Opportunities to further improve this approach by mitigating false-positive errors involve integrating texture filtering with optical and radar data sets. Despite some inherent limitations, our approach overcomes some current challenges of mapping small-scale, but extensive, environmental changes in the wet tropics and thus advances improvements in the continual surveillance, management, and regulation of ASM and other activities that involve selective clearing.

Plant functional diversity (FD) is a component of biodiversity linking plant functional traits to ecosystem processes (e.g., photosynthesis) and services (e.g., gross primary production). Development of remote sensing capabilities to monitor forest FD across various spatio-temporal scales is critical, especially in view of increasing global climate and anthropogenic pressures. Here, we focus on investigating the capability of unoccupied aerial systems (UAS), acquiring imaging spectroscopy data of high spatial (pixel size <= 0.1 m) and spectral (band-width < 5 nm between 400 and 1000 nm) resolutions, to map two trait-based FD metrics, namely, richness and divergence, of two open sclerophyll forests at the plot-scale (<0.2 km(2)). An emerging scalable kernel-based trait probability density (TPD) approach was implemented to compute spatially explicit metrics of FD at different areal extents and pixel sizes through spatially resampled products. Narrow-band spectral indices were utilized as proxies of selected plant functional traits, including photoprotective zeaxanthin-to-antheraxanthin transformation ratio (VAZ), and foliar pigments of chlorophylls and anthocyanins (C-ab and C-ant). The combination of high-resolution imagery and TPDs presents a suitable alternative to the traditional need for taxonomic information and alleviates pixel-based spectral mixing issues known to affect pixel-based FD metrics. A moving kernel (6 x 6 m) applied to UAS data, allowed to capture fine and medium-scale drivers of functional richness and divergence, including within-crown and complex branching variance, topography, sun aspect, and speciation. For the same kernel size, functional richness computed from coarsened pseudo-airborne products (pixel size of 2 m) was found to be 57-68% of that derived from UAS products. Functional divergence did not portray substantial differences across scales and resolutions, even though this metric further emphasized the complexity of the surveyed open-forest sclerophyll sites. UAS have the potential to become an efficient tool for monitoring FD linked with ecosystem processes at key monitoring sites, and for the validation and support of large-scale but less detailed airborne and satellite products. Finally, this study highlights the sensitivity of FD metrics to variations in scale, resolution, and TPD parametrization suggesting that more research is needed to standardize remote sensing protocols for the quantification of FD across spatial and temporal scales.

Understanding land-use dynamics and patterns and how they respond to management scenarios helps to develop sustainable land use policies. This study simulated future land uses in Brantas River Basin (BRB), East Java, Indonesia using Land Change Modeler (LCM) under policy scenarios. Following identification of twelve important biophysical and socio-economic spatial drivers, the LCM accurately simulated land-use changes over the period 1995–2015, with validation against the 1995 land use map giving an overall accuracy of 85–88 %. Under a business-as-usual scenario, the LCM predicted that during the period 2015 to 2035 continuing deforestation may leave forest cover accounting for only 4 % of the total BRB area, and cause declines in dryland from 45 % to 40 % and in rice-field farming from 24 % to 22 %. The results of that scenario also indicate the prospect of massive urban development, increasing from 18 % of the BRB area in 2015 to 30 % in 2035, with increasing threats to food security and water resources. Both percentage changes and a mean-weighted fractal dimension index suggest increased risks of forest fragmentation and urban aggregation. A spatial planning-influenced scenario, which aims to reduce forest loss and support watershed protection, is predicted to lead to 5.5 % forest cover, 37 % dryland, 29 % rice-field farming and 24 % urban area in 2035. Sensitivity analysis of the LCM showed that the model results were more sensitive to drivers, spatial resolution, and policy scenarios than to uncertainty in model parameters. It is concluded that despite some limitations, the LCM successfully provided'' insights into policy impacts on land-uses in BRB, and the roles of forest and rice-field protection in spatial planning are essential in controlling urban development for watershed sustainability.

The estimation of water quality properties through satellite remote sensing relies on (1) the optical characteristics of the water body, (2) the resolutions (spatial, spectral, radiometric and temporal) of the sensor and (3) algorithm(s) applied. More than 80% of global water bodies fall under Case I (open ocean) waters, dominated by scattering and absorption associated with phytoplankton in the water column. Globally, previous studies show significant correlations between satellite-based retrieval methods and field measurements of absorbing and scattering constituents, while limited research from Australian coastal water bodies appears. This study presents a methodology to extract chlorophyll a properties from surface waters from near-coastal environments, within 2 km of coastline, in Tasmania, south-eastern Australia. We use general purpose, global, long-time series, multi-spectral satellite data, as opposed to ocean colour-specific sensor data. This approach may offer globally applicable tools for combining global satellite image archives with in situ field sensors for water quality monitoring. To enable applications from local to global scales, a cloud-based geospatial analysis workflow was developed and tested on several sites. This work represents the initial stage in developing a semi-automated near-coastal water-quality workflow using easily accessed, fully corrected global multi-spectral datasets alongside large-scale computation and delivery capabilities. Our results indicated a strong correlation between the in situ chlorophyll concentration data and blue-green band ratios from the multi-spectral sensor. In line with published research, environment-specific empirical models exhibited the highest correlations between in situ and satellite measurements, underscoring the importance of tailoring models to specific coastal waters. Our findings may provide the basis for developing this workflow for other sites in Australia. We acknowledge the use of general purpose multi-spectral data such as the Sentinel-2 and Landsat Series, their corrections and algorithms may not be as accurate and precise as ocean colour satellites. The data we are using are more readily accessible and also have true global coverage with global historic archives and regular, global collection will continue at least 10 years in the future. Regardless of sensor specifications, the retrieval method relies on localised algorithm calibration and validation using in situ measurements, which demonstrates close-to-realistic outputs. We hope this approach enables future applications to also consider these globally accessible and regularly updated datasets that are suited to coastal environments.

Rangelands, covering half of the global land area, are critically degraded by unsustainable use and climate change. Despite their extensive presence, global assessments of rangeland condition and sustainability are limited. Here we introduce a novel analytical approach that combines satellite big data and statistical modeling to quantify the likelihood of changes in rangeland conditions. These probabilities are then used to assess the effectiveness of management interventions targeting rangeland sustainability. This approach holds global potential, as demonstrated in Mongolia, where the shift to a capitalist economy has led to increased livestock numbers and grazing intensity. From 1986 to 2020, heavy grazing caused a marked decline in Mongolia's rangeland condition. Our evaluation of diverse management strategies, corroborated by local ground observations, further substantiates our approach. Leveraging globally available yet locally detailed satellite data, our proposed condition tracking approach provides a rapid, cost-effective tool for sustainable rangeland management. Rangelands in Mongolia suffered a marked decline in grazing conditions between 1986 and 2020, according to an approach which combines satellite big data with statistical modeling to remotely assess rangeland sustainability and management strategies

Sustainable long-term use of land rehabilitated following mining is required to be resilient to fire and other disturbances. We analysed the vegetation responses to three fires in grassland pasture and open woodland on rehabilitated open-cut coal mine sites in Queensland, Australia. Two fires in central Queensland were controlled burns to manage fuel loads and test the vegetation and landform response, while the third fire, in southeastern Queensland, was an unintended wildfire. We monitored several ecological variables at the study sites for up to five years following the fires and found that vegetation cover, biomass and species richness recovered to pre-fire or unburnt control values within two years. However, one study site experienced lower than average rainfall during the three to five-year post-fire period, resulting in a significant reduction in vegetation cover of between 14 and 31 %, and biomass between 45 and 57 % compared to pre-fire values. Tree and shrub densities changed significantly at two of the sites, reflected in a 635 % increase in stem density of Acacia stenophylla (A.Cunn. ex Benth.) and 82 % mortality of Atriplex nummularia Lindl. subsp. nummularia individuals <2 m in height and 100 % mortality in the 2-5 m height class. The results suggest that rehabilitated pasture systems in central and southern Queensland are resilient to fire in the short-term but are vulnerable to long-term shifts in climate, particularly if a fire precedes a long period of drought. Further resilience work is needed to i) compare rehabilitation recovery with unmined vegetation communities to determine residual risk of future fire impacts, ii) account for seasonality in resilience assessments and iii) understand recovery traits of seed mix combinations sourced from disparate regions.

Artisanal and small-scale mining (ASM) significantly influences the socio-economic development of many low-to-middle-income countries, albeit sometimes at the expense of environmental and human health. Characterized by its labor-intensive extraction from confined (<5 ha) or peripheral mineral reserves, congregated ASM practices can rival the spatial footprint of industrial mines. The unregulated and informal nature of many ASM activities presents monitoring challenges that remote sensing (RS) methods aim to address. While local-scale ASM mapping has seen success, scaling these methods to regional or global levels remains unclear. We review literature on mapping ASM to determine: (1) if studies represent the global distribution and diversity of ASM activities, (2) how ASM's unique characteristics influence the choice of RS methods, and (3) which RS approaches are the most accurate and cost-effective. We found current studies disproportionately focused on ASM regions in Africa, which highlights the need to extend the research to other regions with unique ASM characteristics, such as coal and sand mining in India and China. The selection of RS approaches is heavily influenced by local ASM contexts, the scale of analysis, and resource constraints such as funding for high-resolution imagery and validation data availability. We argue that accurate regional-scale ASM mapping (>100,000 km2) requires innovative combinations of data and methods to overcome data management and storage challenges. Local community participation, including miners, is vital for on-ground mapping and monitoring capacity. We outline a research agenda needed to develop a range of approaches for mapping and monitoring ASM in under-studied regions. By synthesizing effective methods, we provide a foundation for generating accurate and comprehensive spatial data, addressing the issues of inaccurate and incomplete data that global ASM platforms aim to resolve. This spatial data can guide policymakers, NGOs, and businesses in making informed decisions and targeted interventions to improve ASM sector safety, sustainability, and efficiency. Leveraging cloud-based geoprocessing platforms, with regularly updated global satellite image archives, combined with crowd-sourced on-ground information offers a potential solution for sustained regional-scale monitoring.

Coral reefs underpin the environmental, social, and economic fabrics of much of the world's tropical coast. Yet, the fine-scale distribution and composition of coral reefs have never been reported consistently across the planet. Here, we present new area estimates enabled by global geomorphic zone and benthic substrate maps at 5 m pixel resolution. We revise global coral reef estimates to 348,361 km2 of shallow coral reefs and 80,213 km2 (46,237–106,319 km2, 95% confidence interval) of coral habitat. The mapping used more than 1.5 million training samples supported by 480+ data contributions to deploy a coral reef classification of over 100 trillion pixels from the Sentinel-2 satellites and the Planet Dove CubeSat constellation. The publicly available maps are accessible via the Allen Coral Atlas and Google Earth Engine and are already being used by thousands of people to improve the conservation, management, and research of coral reef ecosystems.

The commencement of the United Nations Decade on Ecosystem Restoration has highlighted the urgent need to improve restoration science and fast‐track ecological outcomes. The application of remote sensing for monitoring purposes has increased over the past two decades providing a variety of image datasets and derived products suitable to map and measure ecosystem properties (e.g. vegetation species, community composition, and structural dimensions such as height and cover). However, the operational use of remote sensing data and derived products for ecosystem restoration monitoring in research, industry, and government has been relatively limited and underutilized. In this paper, we use the Society for Ecological Restoration (SER) ecological recovery wheel (ERW) to assess the current capacity of drone‐airborne‐satellite remote sensing datasets to measure each of the SER's recommended attributes and sub‐attributes for terrestrial restoration projects. Based on our combined expertise in the areas of ecological monitoring and remote sensing, a total of 11 out of 18 sub‐attributes received the highest feasibility score and show strong potential for remote sensing assessments; while sub‐attributes such as gene flows, all trophic levels and chemical and physical substrates have a reduced capacity for monitoring. We argue that in the coming decade, ecologists can combine remote sensing with the ERW to monitor restoration recovery and reference ecosystems for improved restoration outcomes at the local, regional, and landscape scales. The ERW approach can be adapted as a monitoring framework for projects to utilize the benefits of remote sensing and inform management through scalable, operational, and meaningful outcomes.

Context Coastal and estuarine finfish species are responding to human-induced climate change by altering their distributions. In tropical regions, the species mostly affected by warming have limited acclimation capacity or live close to their upper thermal limits. Consequently, coastal fish assemblages may dramatically contract in range, experience declining population abundance or local extinction. Aim Here we use two different predictive modelling techniques that cope with non-linear empirical relationships between responses and environmental predictors to investigate distribution change. Methods The habitat-suitability models we use are the maximum entropy model (MaxEnt) and the generalised additive model (GAM). We built the models for the period 2004–2019 with environmental data relevant to coastal systems. We incorporated climate change at current conditions, near future (2015–2054) and distant future (2055–2100) from CMIP6 climate models. Key results We identified bathymetry and sea-surface temperature to be key variables explaining the current and future distribution of coastal finfish and elasmobranchs of the Great Barrier Reef coast in central Queensland. Conclusions We showed how the distributions of valuable fisheries species will change under future warming conditions. Implications The objective is to inform fisheries management supporting the restructure of existing fisheries or the development of new resources for the dual purposes of conservation and food security.

The determination of key phenological growth stages of banana plantations, such as flower emergence and plant establishment, is difficult due to the asynchronous growth habit of banana plants. Identifying phenological events assists growers in determining plant maturity, and harvest timing and guides the application of time-specific crop inputs. Currently, phenological monitoring requires repeated manual observations of individual plants’ growth stages, which is highly laborious, time-inefficient, and requires the handling and integration of large field-based data sets. The ability of growers to accurately forecast yield is also compounded by the asynchronous growth of banana plants. Satellite remote sensing has proved effective in monitoring spatial and temporal crop phenology in many broadacre crops. However, for banana crops, very high spatial and temporal resolution imagery is required to enable individual plant level monitoring. Unoccupied aerial vehicle (UAV)-based sensing technologies provide a cost-effective solution, with the potential to derive information on health, yield, and growth in a timely, consistent, and quantifiable manner. Our research explores the ability of UAV-derived data to track temporal phenological changes of individual banana plants from follower establishment to harvest. Individual plant crowns were delineated using object-based image analysis, with calculations of canopy height and canopy area producing strong correlations against corresponding ground-based measures of these parameters (R2 of 0.77 and 0.69 respectively). A temporal profile of canopy reflectance and plant morphology for 15 selected banana plants were derived from UAV-captured multispectral data over 21 UAV campaigns. The temporal profile was validated against ground-based determinations of key phenological growth stages. Derived measures of minimum plant height provided the strongest correlations to plant establishment and harvest, whilst interpolated maxima of normalised difference vegetation index (NDVI) best indicated flower emergence. For pre-harvest yield forecasting, the Enhanced Vegetation Index 2 provided the strongest relationship (R2 = 0.77) from imagery captured near flower emergence. These findings demonstrate that UAV-based multitemporal crop monitoring of individual banana plants can be used to determine key growing stages of banana plants and offer pre-harvest yield forecasts.

Convolutional Neural Networks (CNN) consist of various hyper-parameters which need to be specified or can be altered when defining a deep learning architecture. There are numerous studies which have tested different types of networks (e.g. U-Net, DeepLabv3+) or created new architectures, benchmarked against well-known test datasets. However, there is a lack of real-world mapping applications demonstrating the effects of changing network hyper-parameters on model performance for land use and land cover (LULC) semantic segmentation. In this paper, we analysed the effects on training time and classification accuracy by altering parameters such as the number of initial convolutional filters, kernel size, network depth, kernel initialiser and activation functions, loss and loss optimiser functions, and learning rate. We achieved this using a well-known top performing architecture, the U-Net, in conjunction with LULC training data and two multispectral aerial images from North Queensland, Australia. A 2018 image was used to train and test CNN models with different parameters and a 2015 image was used for assessing the optimised parameters. We found more complex models with a larger number of filters and larger kernel size produce classifications of higher accuracy but take longer to train. Using an accuracy-time ranking formula, we found using 56 initial filters with kernel size of 5 × 5 provide the best compromise between training time and accuracy. When fully training a model using these parameters and testing on the 2015 image, we achieved a kappa score of 0.84. This compares to the original U-Net parameters which achieved a kappa score of 0.73.

This study introduces a prototype end-to-end Simulator software tool for simulating two-dimensional satellite multispectral imagery for a variety of satellite instrument models in aquatic environments. Using case studies, the impact of variable sensor configurations on the performance of value-added products for challenging applications, such as coral reefs and cyanobacterial algal blooms, is assessed. This demonstrates how decisions regarding satellite sensor design, driven by cost constraints, directly influence the quality of value-added remote sensing products. Furthermore, the Simulator is used to identify situations where retrieval algorithms require further parameterization before application to unsimulated satellite data, where error sources cannot always be identified or isolated. The application of the Simulator can verify whether a given instrument design meets the performance requirements of end-users before build and launch, critically allowing for the justification of the cost and specifications for planned and future sensors. It is hoped that the Simulator will enable engineers and scientists to understand important design trade-offs in phase 0/A studies easily, quickly, reliably, and accurately in future Earth observation satellites and systems.