Adjusted CBA-Wheat model for predicting aboveground biomass in winter wheat from hyperspectral data

Abstract

Context or problem

Crop aboveground biomass (AGB) is a key indicator of photosynthesis and carbon cycle dynamics in agricultural ecosystems. The availability of accurate, real-time AGB data enables efficient resource management and precision farming. The crop biomass algorithm for wheat (CBA-Wheat) estimates winter wheat AGB using vegetation index (VI) and Zadoks stage (ZS), but acquiring ZS data through field surveys is challenging for large-scale applications.

Objective or research question

This study aimed to optimize the CBA-Wheat model by incorporating the concept of the relative day of the year (RDOY) as a replacement for ZS and combining it with VI to enhance the performance of the wheat growth model.

Methods

We proposed the concept of RDOY to replace the traditional ZS, thereby optimizing the CBA-Wheat model. The study used data from Xiaotangshan, Beijing, from 2013 to 2020 for model development. The validation dataset included 2021 Xiaotangshan data, 2010 suburban Beijing data, and 2012 Yucheng, Shandong data for testing the model’s temporal and spatial transferability. Additionally, we compared the performance of the CBA-Wheat_RDOY model with machine learning models, including Partial Least Squares Regression (PLSR) and Random Forest (RF).

Results

We found that the modified CBA-Wheat_RDOY model, utilizing the modified simple ratio vegetation index (MSR) as an input parameter, achieved the highest AGB estimation accuracy, with a coefficient of determination (R²) of 0.82 and a root mean square error (RMSE) of 1.71 t/ha. This result surpassed the performance of partial least squares regression (R² = 0.78, RMSE = 1.48 t/ha) and random forest (R² = 0.73, RMSE = 2.03 t/ha) models when RDOY was introduced.

Conclusions

Our findings highlight the effectiveness of introducing RDOY in improving the accuracy of winter wheat biomass estimation within the CBA-Wheat model. Moreover, RDOY is a superior alternative to traditional phenological observations and can potentially enhance the performance of conventional machine learning models.

Implications or significance

Compared with existing algorithms, the CBA-Wheat_RDOY model, grounded in RDOY, not only responds sensitively to various phenological stages but also exhibits improved inversion accuracy. This approach holds promising potential for enhancing the timeliness and spatial extrapolation of winter wheat AGB predictions, advancing precision agriculture and ecosystem management.