مروری: داده گواری سنجش از دور در کالیبراسیون مدلهای رشد محصول: با تاکید بیشتر بر روشهای تحلیل حساسیت

Introduction: Crop growth models are convenient tools for understanding and predicting the interactions between crop growth, environmental conditions, and land management practices such as irrigation (Servin-Palestina et al., 2022; Li et al., 2023). The crop models comprise several computational stages and parameters along with climate, soil, crop, management, temperature, salinity, fertility, and water stress conditions (Akbari et al., 2024b). These factors can significantly challenge model calibration, even potentially leading to uncertainties in the results (Guo et al., 2019).

To successfully run a crop simulation model, the selected model needs to be calibrated with accurate crop model parameters based on local soil conditions, weather, management practices, and other conservative/non-conservative parameters that may be difficult to measure locally (Shan et al., 2021). To address this challenge, it demands a spatially explicit data assimilation strategy that incorporates the observed data to calibrate the model parameters. It minimizes the difference between observed data and the state variables simulated by the crop growth model and then estimates the model parameters (Hoefsloot et al., 2012). The assimilation of the satellite-derived products and their spatial variability as pixels in each farm into such models can, to some extent, resolve the uncertainties introduced by the assumption of homogeneity in croplands (Hoefsloot et al., 2012; Jin et al., 2017).

Calibrating a crop growth model for the specific location and agricultural conditions of a region can thus be a powerful tool for developing effective water management strategies that enhance production while minimizing water consumption (Hsiao et al., 2009). Furthermore, when calibrating crop growth models on a spatial scale beyond an individual farm, it is necessary to reduce the uncertainties related to input data to account for the lack of information about land management and the structure of the model. To cope with these limitations (e.g., data scarcity, uncertainty), the pragmatic practice seeks to simplify crop growth models with fewer parameters and data requirements. It is therefore necessary to determine the minimum number of effective parameters in each of the crop growth models to achieve a more accurate and optimal model calibration, or, in other words, apply a sensitivity analysis (SA) (Silvestro et al., 2017).

Material and methods: This study examines the critical role of sensitivity analysis (SA) in model parameter calibration as a part of assimilating satellite data into crop growth models with the purpose of improving the accuracy of crop growth simulations and yield estimation. Calibration involves adjusting model parameters with the purpose of minimizing discrepancies between observed variables and model simulations. Undoubtedly, the choice of the calibration method for optimizing crop model parameters depends on the specific model and requires knowledge of its most influential and sensitive parameters, as can be defined by SA. In this light, aiming for optimizing assimilation of data streams of satellite products in crop growth modeling, it is indispensable to identify the most sensitive model parameters and those that can be fixed. Restricting input parameters reduces redundancy and uncertainty, particularly when calibrating models for specific study sites. This review explores the diversity of SA techniques applicable to crop growth models, aiding readers in choosing the best SA method for their needs.

Results and discussion: The study reveals that global SA methods are predominantly employed in crop growth model calibration practices, especially when data streams of satellite products have been assimilated into the model. As a general trend in the reviewed studies, the EFAST model tends to outperform Sobol in use and accuracy. In cases where complexity is high, it is suggested to use the Morris method for screening parameters in combination with applying the EFAST model to reduce computational complexity and crystallize the most effective parameters at relatively high accuracies. Furthermore, this review study shows the ensemble, i.e., combination, of global SA methods Morris and EFAST outperforms other SA methods in calculation efficiency and in precision of identifying driving parameters. Such ensemble strategies excel in finding the driving parameters, which can lead to fine-tuned calibration and, in turn, to more precise crop growth model simulations.

Conclusion: We conclude that an ensemble of global SA methods is an appropriate choice for overcoming the challenges and limitations of each technique and reducing the computational complexities when introducing satellite data assimilation into common crop growth models.