Abstract Title: Innovative Application of CubeSat Imagery to Predict Wheat Yield Without Ground-Based Data

Abstract Submitted to: BIOGEOSCIENCES

Abstract Text:

Providing reliable, consistent and scalable projected crop yield data is one of the major challenges in monitoring and managing food security. Accurate yield forecasts, as early as possible prior to harvest, are critical for market stability, as well as for farmers, grain companies and governments. For years, the trade-off between high spatial and temporal resolutions has limited remotely-sensed applications such as estimating crop yield at the field and sub-field scales. The growing availability of CubeSats has opened the door to a new era of crop monitoring from space. This study ultimately aims to improve in-season yield predictions by coupling crop modelling and satellite images, with a focus on wheat in Australia. By contrast to most of the previous studies, this study attempted to develop a new approach to predict crop yield without ground-calibration data, which makes it applicable across different environments. In this process, a CubeSat-based sowing date detection method was developed to identify cultivated fields and the date when they were sown, PlanetScope images (with a spatial resolution of ~3 m) and Sentinel-2 images (with a spatial resolution of 10 m) were fused to create daily Leaf Area Index (LAI) datasets at 3 m resolution. Finally, the detected sowing dates and the LAI datasets were coupled with the APSIM-Wheat crop model to predict wheat yield at the field and subfield scales. As part of the process, ~2,000 simulations of APSIM were generated for each field, thus spanning a realistic range of possible environmental and on farm variables. Each APSIM simulation outputs a yield estimate as well as daily crop characteristics including LAI. Simulated LAI patterns were compared to the remotely-sensed LAI to choose the simulation reflecting best the studied crop. This method was found to be effective not only to produce field-scale yield estimations, but also to generate yield maps at 3 m resolution nearly three months before crop harvest.