University of Salford (GB)
Ground Reference Observations Underlying Novel Decametric Vegetation Data Products from Earth Observation (GROUNDED EO) aims to exploit cutting-edge machine learning approaches, ground data collection capabilities, and data fusion methods to develop improved decametric vegetation biophysical products from Sentinel-2 and -3, ultimately facilitating better environmental decision making.
The objectives of the project are to:
Vegetation covers approximately 70% of the terrestrial surface, influencing biogeochemical processes through controls on the exchange of water, carbon, and energy with the atmosphere. Accurate estimates of biophysical variables describing vegetation condition and dynamics are required as inputs to models of crop yield, carbon exchange, and the weather and climate systems, which are fundamental to developing successful environmental policy, and play a crucial role in informing effective climate change mitigation strategy. Accounting for approximately 30% of the land surface and approximately 50% of its gross primary productivity (GPP), the world’s forests are crucial in terms of carbon storage, but also provide a range of important ecosystem services, acting as a source of fibre, fuel and timber. Meanwhile, agricultural crops, which account for approximately 13% of the land surface, represent the main source of the world’s food, whose security needs to be ensured in the context of an increasing global population. High quality EO-derived estimates of vegetation biophysical variables are essential for monitoring and managing these resources.
Whilst current retrieval algorithms including the Sentinel-2 Level-2 Prototype Processor (SL2P) can now provide routine decametric (10 m to 100 m) estimates of biophysical variables including leaf area index (LAI) and the fraction of absorbed photosynthetically active radiation (FAPAR), they are subject to known biases due to assumptions embedded within the radiative transfer models used in their training. Meanwhile, their temporal frequency (≥ 5 days) cannot resolve rapid changes in vegetation status (e.g. due to stress, pests, and disease), particularly in cloud-prone environments. GROUNDED EO will provide a fundamental advancement upon these algorithms. By taking advantage of cutting-edge machine learning approaches and ground data collection capabilities, retrieval algorithms will be trained on real EO data and contemporaneous ground reference observations, enabling biases due to radiative transfer model assumptions to be avoided. Spatiotemporal data fusion will then enable improved temporal coverage to be attained.
In achieving these objectives, GROUNDED EO will also deliver a step change in the provision of ground reference observations, which have been limited in quantity, typically being obtained through one-off field campaigns restricted to the peak of the growing season, and which have also suffered from inconsistencies and the presence of unquantified measurement uncertainties. By collecting and harmonising observations from novel automated field instrumentation and recent environmental monitoring networks using a standardised processing chain, and by adopting traceable uncertainty quantification methods developed under the Fiducial Reference Measurements (FRM) programme, these limitations will be addressed.
Overall, by increasing the accuracy and spatiotemporal coverage of decametric LAI and FAPAR products, GROUNDED EO has potential to reduce uncertainties in downstream applications, providing improved estimates of variables such as crop yield and GPP, and ultimately resulting in better environmental decision making.