The expansive scale and inaccessible nature of archaeological landscapes in the Australian Arid Zone provide a significant challenge for cultural heritage management. These areas, such as the Strzelecki Desert, are also home to a complex mosaic of stakeholders including petroleum companies, pastoralists, Traditional Owners, and National Parks who require large-scale, high-quality affordable mapping of archaeological landscapes in order to make informed management decisions. This project will undertake an extensive trial of high-resolution remote sensing methods to map key geomorphic features to inform our understanding of archaeological potential, chronology, and depositional history.

The use of remote sensing data to map archaeological and geomorphic features of interest in arid zone archaeology in Australia is rare but has shown great potential (Law et al. 2017, Law et al. 2019, Law et al. 2021). The use of machine learning approaches to automatically detect geomorphic features in other locations based on elevation (ie. Orengo et al. 2020) or multispectral (ie. Orengo and Petrie 2017) data has shown significant potential. This project will form part of Associate Professor Ian Moffat’s Australian Research Council-funded Future Fellowship working in the Simpson Desert and the Coongie Lakes in collaboration with the Yandruwandha Yawarrawarrka and Wangkangurru Yarluyandi Traditional Owners. In addition to working with A/Prof Moffat, Oliver Hatswell will receive high-level supervision from Dr Jarrad Kowlessar from Flinders University and Professor Hector Orengo from the Catalan Institute of Classical Archaeology who are international experts in the fields of archaeological remote sensing and machine learning.

In P2-38, simulated imagery and a low-cost emulation system were utilized to replicate HyperScout 2 AI onboard processing for fire smoke detection. This project aims to adjust and test all onboard imagery processing tasks, developed as part of P2-38, and verify them for their error-proof deployment on the Kanyini/HyperScout 2 system. Project activities include functional and performance verifications on four systems – (a) Virtual System, (b) Emulation system, (c) Distributed FlatSat system and (d) FlatSat system with EM. Expected outcomes are adjusted CPU and VPU scripts for onboard fire smoke detection that will have been systematically tested to run successfully onboard of Kanyini, ensuring that all onboard function is working correctly under all possible configurations, scenarios and constraints.

Once growers and agronomist have a reliable forecast of disease risk they need to know which farms and paddocks are most at risk. This project will enable growers to manage the impacts of grain crop diseases more profitably using new detection and mapping capabilities that detect infection prior to visual symptoms that can be scouted by an agronomist. It will do so by coupling satellite-based analytics with on-farm data and mechanistic (process-based) models of disease epidemiology developed by Australian grain crop pathologists.

This project will test the feasibility of developing scalable analytics products capable of mapping weeds via satellite imagery. Two key analytics products are envisaged from this project: one to map priority weeds in fallow, and another to map weeds in-crop in high value use-cases.

Data Farming will work with technical experts in hyperspectral imagery and earth observation analytics at the SmartSat CRC, along with technology providers in each GRDC region capable of providing high quality ground-based validation data on weed species distributions, densities, growth stages, and other key parameters.

The project will establish national linkages and leverage additional agronomic and crop protection expertise from Elders and Syngenta Australia respectively.

Crop mapping is an important research area in agriculture monitoring, since an accurate map of agricultural land provides a comprehensive understanding of crop distribution, yield estimation, and the factors influencing crop production.

Over the past years, the launch of Earth Observation satellites with diverse sensors has produced abundant satellite image time series containing valuable agricultural information. However, effectively leveraging the data for accurate crop mapping still faces the following problems: a) The optical satellite image time series suffer from data gaps caused by haze or cloud shadow, severely affecting the performance of subsequent classification or segmentation task; b) Current crop mapping methods attempt to make more comprehensive use of information from different sensors by fusing them trivially, but they ignore the data corruption existed in part of modalities. c) Current deep neural networks for semantic segmentation often exhibit over-confidence in their predictions due to overfitting, resulting in unreliable decision making, such as improper agricultural subsidy allocation.

To address these problems, the corresponding solutions are proposed: a) Explore a diffusion model for cloud removal in optical satellite time series. The model leverages both spatial and temporal information for cleaner, high-quality images and improved crop mapping performance; b) Develop an adaptive attention-based fusion method with a quality-aware module, which dynamically highlights the most relevant high-quality features from different modalities to obtain more accurate and robust predictions; c) Design a probability calibration method to adaptively calibrate the prediction of each pixel, which is a crucial step towards achieving trustworthy predictions in practical applications.

Through these solutions, the research aims to enhance the utilization of satellite image time series and develop more effective and robust algorithms to improve the performance of crop mapping.

Many satellite data users lack the expertise and infrastructure to efficiently and effectively access, pre-process, and utilize the growing volume of space-based data for local, regional, and national decision-making (Barker et al., 2022; Chue Hong et al., 2021). This represents a significant obstacle to realizing the full potential of and achieving successful utilization of space-based imagery data. In response to this challenge, various countries and international organizations have expressed a strong desire for support from the Committee on Earth Observation Satellites (CEOS), a pioneering advocate of the CEOS Analysis Ready Data (ARD) initiative (Sazonov et al., 2019).

The CEOS ARD initiative focuses on streamlining access to and processing of satellite data, transforming it into CEOS Analysis Ready Data for Land (CARD4L) products. These products are specifically designed to facilitate seamless time-series analysis without requiring supplementary acquisition information. By systematically and consistently providing CARD4L data, the burden on global satellite data users is expected to be significantly reduced, ultimately enhancing the usability of the data. The delivery of CARD4L data will adhere to the Guiding Principles of Findable, Accessible, Interoperable, and Reusable (FAIR) including various methods such as systematic data processing, hosting platforms, and toolkits made available to users (Wilkinson et al. 2016; Musen et al., 2022; Jacobsen et al., 2020; Weigel et al., 2020).

However, the sheer volume of data generated by the extensive Earth observation data and services provided by over 300 satellites poses a substantial challenge for analysts, scientists, and even non-experts (FrontierSI, 2020). This emphasizes the urgent need to establish a comprehensive FAIRified data model framework for computational workflows, aimed at fully unlocking the potential of Earth observation products in the environmental sector and beyond. Recent advancements in storage and computing power have not only made this cost-effective but also feasible to process and analyze data across various scales. Machine Learning (ML), a fundamental component of modern science and industry, significantly simplifies scientific analysis by efficiently identifying patterns, outliers, and discrepancies in datasets, streamlining data preprocessing, and promoting the development of more accurate, robust, and scalable applications (Peng et al. 2023).

Extensive research has been conducted in various domains focusing on the FAIRified data model framework integrated with Machine Learning (ML). One notable outcome of this research is the development of the Source-augmented Partial Convolution v2 model (SAPC2), which serves as an innovative solution for pixel reconstruction. SAPC2, built upon a partial convolution-enabled U-Net framework, leverages a complete source temporally adjacent to the reconstruction process. It employs an encoder-decoder structure to extract high-level features (Chen M, 2020). Traditionally, many applications have tended to disregard observations affected by cloud coverage or sensor saturation. A recent study has shed light on the issue of errors of commission arising from the Fmask technique. It emphasizes the importance of continued inclusion and enhancement of quality assurance bands in Analysis-Ready Data (ARD). The study advises users to utilize pixel quality flagging to mitigate potential biases under varying conditions (Ernst, 2018). This approach offers significant potential for standardizing the documentation of workflows and increasing trust in EO data products, thereby increasing productivity and utilization in environmental sector and other sectors. Furthermore, in-depth research efforts have been dedicated to dynamic environmental simulations, covering the intricate interplay between human activities and land-use changes (Searchinger et al., 2018; Newbold et al., 2015), climate dynamics (Findell et al., 2017), water resource management (Spera et al., 2016), and the socioeconomic system (Hostert et al., 2011).

Overall, this research advocates for a comprehensive approach that seamlessly incorporates the FAIR Guiding Principles into machine learning workflows. This integration aims to facilitate the generation of Analysis Ready Data (ARD) to significantly improve efficiency and reproducibility in the field of land observation and monitoring. The central tenet of this work underscores the critical significance of standardized data management procedures and automated workflows in empowering data-driven investigations in the environmental sector. By embracing this integrated framework, the global scientific community can fully harness the untapped potential of ARD products. This approach fosters collaborative efforts and paves the way for groundbreaking discoveries. Furthermore, it actively promotes interoperability between public and commercial data sources, enabling a wide array of applications. This approach is a cornerstone of a more sustainable and informed approach to land management, ensuring the optimal utilization of Earth Observation (EO) products across a multitude of sectors.