A lightweight SIF-based crop yield estimation model: A case study of Australian wheat

Abstract

As Australia's primary staple and export crop, wheat necessitates reliable yield mapping to ensure timely alerts about food insecurity. Conventional crop yields are estimated using either process-based or statistical models, but both face challenges in large-scale application due to the extensive data required. Recent studies have shown that the gross primary production (GPP) of plants can be mechanistically estimated from the fraction of open PSII reaction centers (q_L), solar-induced chlorophyll fluorescence (SIF), and readily accessible meteorological datasets including air temperature (T_air), dew-point temperature, and soil water content. q_L can be modeled as a function of SIF and T_air. Along with these theoretical advances, the resolution of satellite SIF has greatly improved, boosting the potential for accurate large-scale crop yield estimation. In this study, we develop a SIF-based lightweight crop model which uses q_L and SIF to track crop GPP. This approach allows for a direct mechanistic estimation of GPP without the need to explicitly account for numerous complex agro-climatic processes. We apply this model to estimate Australian wheat yields from 2019 to 2022. The model exhibits strong predictive power, explaining 86 % of wheat production variance at the regional level (RMSE: 91 kilotons, rRMSE: 7.24 %) and 91 % at the state level (RMSE: 1509 kilotons, rRMSE: 14.13 %). Australian wheat yields exhibit a positive correlation with soil water content and vapor pressure deficit (VPD) when VPD remains below 0.80 kPa. However, the correlation turns negative once VPD exceeds this threshold. We also identify the main sources of error in estimating wheat production as: (1) inaccuracies in estimating the harvested area of wheat, and (2) the relatively low spatial resolution of current satellite SIF data. Our model, with its lightweight design and its ability to mechanistically estimate crop photosynthetic CO₂ assimilation, offers a promising, novel framework for practical, large-scale crop yield mapping.