Assimilation of soil moisture and canopy cover data improves maize simulation using an under-calibrated crop model

Abstract

Parameter calibration is normally required prior to crop model simulation, which can be a time-consuming and data-intensive task. Meanwhile, the growth stages of different hybrids/cultivars of the same crop often show some similarities, which implies that phenological parameters calibrated for one hybrid/cultivar may be useful for the simulation of another. In this study, a data assimilation framework is proposed to reduce the requirement for parameter calibration for maize simulation using AquaCrop. The phenological parameters were uniformly scaled from previous research performed in a different location for a different maize hybrid, and other parameters were taken from default settings in the model documentation. To constrain simulation uncertainties, soil moisture and canopy cover observations were assimilated both separately and jointly in order to update model states. The methodology was tested across a rain-fed field in Nebraska for 6 growing seasons. The results suggested that the under-calibrated model with uniformly scaled phenological parameters captured the temporal dynamics of crop growth, but may lead to large estimation bias. Data assimilation effectively improved model performance, and the joint assimilation outperformed single-variable assimilation. When soil moisture and canopy cover were jointly assimilated, the overall yield estimates (RMSE = 1.24 t/ha, nRMSE = 11.48%, R 2 = 0.695) were improved over the no-assimilation case (RMSE = 2.01 t/ha, nRMSE = 18.61%, R 2 = 0.338). Sensitivity analyses suggested that the improvement was still evident with temporally sparse soil moisture observations and a small ensemble size. Further testing using observations within 90 days after planting demonstrated that the method was able to predict yield around 3 months before harvest (RMSE = 1.7 t/ha, nRMSE = 15.74%). This study indicated that maize yield can be estimated and predicted accurately by monitoring the soil moisture and canopy status, which has potential for regional applications using remote sensing data.

Monitoramento de plantas em casas de vegetação para assimilação de dados

“…Sendo assim, as conclusões são limitadas e dificultam comparações. A despeito disso, estudos sobre valores adequados para os parâmetros de incerteza são necessários, uma vez que a decisão de como atribuir esses valores de incerteza é um problema em aberto e diversos autores a tomam de formas distintas: alguns autores trabalham com erros variáveis [Lu et al 2021], outros usam valores fixos também relacionados aos erros do modelo de conversão [Huang et al 2016], outros estimam com base na variância de observações coletadas em campo [Zhao et al 2013] enquanto outros usam estratégias que independem dessa incerteza [Chen et al 2018;Vazifedoust et al 2009].…”

A comprehensive assessment of in situ and remote sensing soil moisture data assimilation in the APSIM model for improving agricultural forecasting across the U.S. Midwest

“…Numerous past studies have used SDA to constrain crop model estimates, using observations on leaf area index (e.g., Nearing et al, 2012;Ines et al, 2013;Ma et al, 2013;Chen et al, 2018;Lu et al, 2021), soil moisture (Kivi et al, 2022), biomass (e.g., Linker and Ioslovich, 2017) and evapotranspiration (e.g., Huang et al, 2015). For example, a synthetic study by Zhu et al (2017) found that the assimilation of coarse resolution surface soil moisture data into a coupled soil water-groundwater numerical model constrained soil moisture estimates in the first 50 cm of the soil profile despite explicitly unaccounted spatial heterogeneity in soil properties.…”

Process-based crop models calibration and structure affect data assimilation for estimating sugarcane yield

“…This agrees with , which showed that the EnKF had better yield estimation for maize when calibration was applied prior to LAI data assimilation. In contrast, Lu et al (2021) improved the accuracy of maize yield estimation without prior genotype-specific calibration. In their study, the EnKF method was employed to assimilate canopy cover data for adjusting the phenological parameters of the AquaCrop model for each crop season.…”

“…Therefore, studies with this methodology in areas where the reduction factors are present, or there would be data flaws in describing the environment, such as soil characterization and climate data, may result in a more pronounced accuracy gain due to the use of DA methods (Gilardelli et al, 2019). Thirdly, other state variables simulated by DS, such as above ground biomass (Lu et al, 2021) plant height , soil moisture , canopy nitrogen accumulation (Li et al, 2015), and canopy cover (Lu et al, 2021) could be also used to enhance the model accuracy. Yet, the used PBM was developed to run without assimilations from start to end, for this study we just adjusted the state variables directly related to the LAI, plant weight, and leaf area.…”

“…Moreover, due to the difficulty of quantifying the variability of genotypes in large areas, this information is commonly resumed to just one calibration in DA simulations Tewes et al, 2020a). Lu et al (2021) proposed a framework to use the EnKF to simulate Maize, without using a specific genotype calibration and presented satisfactory improvement compared to open-loop (OP) simulations. However, recent studies emphasized that calibrating the model parameters that are specific for cultivar or region should be targeted before DA methods , and this should be more stressed in crop-specific PBM, due to a large number of parameters to describe the crops cultivar .…”