Estimation of sugarcane biomass from Sentinel-2 leaf area index using an improved SAFY model (SAFY-Sugar)

Abstract

Assimilating crop biophysical traits (e.g., Leaf area index, LAI) derived from remote sensing data into a crop growth model provides an effective way for monitoring spatiotemporal variability of crop biomass and yield. However, traditional complex crop growth models generally require extensive input parameters and computation resources, limiting their applicability for large-area estimation using earth observation data. The Simple Algorithm for Yield Estimation model (SAFY), a semi-physical crop growth model grounded in light use efficiency theory has been widely adopted for satellite-based biomass estimation in major field crops. Despite its utility, SAFY cannot directly simulate sugarcane stalk biomass, a critical metric for sugarcane yield assessment. To address this, we developed SAFY-Sugar, a revised SAFY incorporating a temperature-driven stalk biomass module that partitions daily above-ground biomass into stalk biomass. Multi-temporal LAI (S2-LAI) was first inverted from vegetation indices of Sentinel-2 satellite using a semi-empirical model calibrated with the 250 m GLASS LAI product as reference. The estimated S2-LAI achieved an overall accuracy of 0.50 m²/m² in RMSE across selected vegetation indices. Two data assimilation strategies to assimilate the S2-LAI into the SAFY or SAFY-Sugar model for above-ground and stalk biomass estimation in fields were tested (1) independently optimizing SAFY and SAFY-Sugar parameters with S2-LAI alone, and (2) pre-optimizing the stalk module using independent farm measurements before assimilation. SAFY employed a fixed biomass allocation coefficient for stalk biomass estimation. Under the first strategy, SAFY-Sugar demonstrated large improvements in stalk biomass estimation (R² = 0.94, nRMSE = 26.09 %) compared to SAFY (R² = 0.92, nRMSE = 32.73 %). The second strategy further enhanced SAFY-Sugar’s accuracy (R² = 0.98, nRMSE = 14.34 %). For regional application in Chongzuo City (2020 – 2021) using the second strategy, SAFY-Sugar captured spatial yield variability, consistent with the government statistics (nRMSE = 6.61 %). By integrating satellite data assimilation, SAFY-Sugar provides a robust framework for monitoring sugarcane productivity across scales, advancing precision agriculture in sugarcane systems.

Plant phenotyping relevance

糖分作物のLAIから茎・地上部バイオマスを推定するモデルと衛星データ同化手法を開発し、精度検証と地域適用を行っており、植物形質取得・推定が研究の中心です。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.