A bayesian approach to crop Model calibration under unknown error covariance

Hue, Carine; Tremblay, Marie; Wallach, Daniel

doi:10.1198/108571108x335855

Cited by 8 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…NIS approximates the a posteriori distribution with a vector of weighted parameters or variables. This method has been applied to crop model inversions (Hue et al, 2008) but not to the assimilation of satellite imaging. Actually, when the processed entities are numerous for a similar set of forcing inputs (i.e.…”

Section: Normalised Importance Sampling and Look-up Tablementioning

confidence: 99%

AgriCarbon-EO: v1.0.1: Large Scale and High Resolution Simulation of Carbon Fluxes by Assimilation of Sentinel-2 and Landsat-8 Reflectances using a Bayesian approach

Wijmer¹,

Bitar²,

Arnaud³

et al. 2023

Preprint

View full text Add to dashboard Cite

Abstract. Soil carbon storage is a well identified climate change mitigation solution. The extensive in-situ monitoring of the soil carbon storage in cropland for agricultural policy and offset carbon markets is prohibitive, especially at intra-field scale. For this reason, comprehensive Monitoring, Reporting and Verification (MRV) of soil carbon and its explanatory variables at large scale needs to rely on remote sensing and modelling tools that provide the spatio-temporal dynamics of the carbon budget and it’s components at high resolution with associated uncertainties. In this paper, we present AgriCarbon-EO v1.0: an end-to-end processing chain that enables the estimation of carbon budget components of major crops and cover crops at intra-field resolution (10 m) and large scale (over 110×110 km) by assimilating remote sensing data in physically-based radiative transfert and agronomic models. The data assimilation in AgriCarbon-EO is based on a novel Bayesian approach that combines Normalised Importance Sampling (NIS) and Look-Up Table (LUT) generation. This approach propagates the 10 m uncertainties across the processing chain from the reflectances to the output variables. The chain considers as input a land cover map, multi-spectral reflectance maps from the Sentinel-2 and Landsat-8 satellites, and daily weather forcing. The PROSAIL radiative transfer model is inversed in a first step to obtain Green Leaf Area Index (GLAI). The GLAI time series are then assimilated into the SAFYE-CO2 crop model taking into consideration their uncertainty. The chain is applied over winter wheat in the south-west of France during the cropping seasons 2017 and 2019. We compare the results against the net ecosystem exchange measured at the FR-AUR ICOS site (RMSE = 1.69 - 2.4 gC m−2 , R2 = 0.88 - 0.88), biomass (RMSE = 250 g m−2 , R2 = 0.9), and combine harvester yield maps. We quantify the difference between pixel and field and pixel scale simulations of biomass (bias = -47 g m−2 , -39 % variability), and the impact of the number of remote sensing acquisitions on the outputs (-66 % of mean uncertainty of biomass). Finally, we conduct a coherency analysis at regional scale to test the consistency of the observed patterns with soil texture, altitude and exposition variability. Results show higher biomass for higher clay soils and earlier emergence and senescence for south western exposition.

show abstract

Section: Normalised Importance Sampling and Look-up Tablementioning

confidence: 99%

AgriCarbon-EO: v1.0.1: Large Scale and High Resolution Simulation of Carbon Fluxes by Assimilation of Sentinel-2 and Landsat-8 Reflectances using a Bayesian approach

Wijmer¹,

Bitar²,

Arnaud³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Parameters were first pre-selected (Hue et al, 2008;Makowski et al, 2006) based on expert knowledge. Parameter values were elicited from an expert and are given in Table 2.…”

Section: Selection Of Model Parametersmentioning

confidence: 99%

“…One advantage of Bayesian inference is the use of prior information (Sexton et al, 2016). The posterior probability distribution obtained by conditioning on one dataset can then be used as a prior distribution for the next dataset in a sequential manner (Hue et al, 2008). This approach, called Bayesian sequential updating (BSU), would be more computationally efficient than having to re-calibrate the model to all previous datasets, every time new data are available.…”

mentioning

confidence: 99%

A Bayesian sequential updating approach to predict phenology of silage maize

Viswanathan

Weber

Gayler

et al. 2021

Preprint

View full text Add to dashboard Cite

Abstract. Crop models are tools used for predicting year to year crop development on field to regional scales. However, robust predictions are hampered by factors such as uncertainty in crop model parameters and in the data used for calibration. Bayesian calibration allows for the estimation of model parameters and quantification of uncertainties, with the consideration of prior information. In this study, we used a Bayesian sequential updating (BSU) approach to progressively incorporate additional data at a yearly time-step to calibrate a phenology model (SPASS) while analysing changes in parameter uncertainty and prediction quality. We used field measurements of silage maize grown between 2010 and 2016 in the regions of Kraichgau and Swabian Alb in southwestern Germany. Parameter uncertainty and model prediction errors were expected to progressively reduce to a final, irreducible value. Parameter uncertainty reduced as expected with the sequential updates. For two sequences using synthetic data, one in which the model was able to accurately simulate the observations, and the other in which a single cultivar was grown under the same environmental conditions, prediction error mostly reduced. However, in the true sequences that followed the actual chronological order of cultivation by the farmers in the two regions, prediction error increased when the calibration data was not representative of the validation data. This could be explained by differences in ripening group and temperature conditions during vegetative growth. With implications for manual and automatic data streams and model updating, our study highlights that the success of Bayesian methods for predictions depends on a comprehensive understanding of inherent structure in the observation data and model limitations.

show abstract

“…Bayesian calibration (BC) is a statistical method that is frequently applied in the estimation of effective values of model parameters (Hue et al, 2008;Iizumi et al, 2009Iizumi et al, , 2011Tao et al, 2009;Ceglar et al, 2011), as a way of dealing with parameter uncertainty and measurement errors. Probability distribution of model parameters ((Â p ) are conditioned on the model output observations available (O) and the model used (M).…”

Section: Model Calibration: Bayesian Approachmentioning

confidence: 99%