Parameter and uncertainty estimation for maize, peanut and cotton using the SALUS crop model. (May 2015)

• Record Type: Journal Article
• Authors: Dzotsi, K.A.; Basso, B.; Jones, J.W.
• Is Part Of: Agricultural systems. Volume 135(2015)
• Page Start: 31
• Page End: 47
• Publication Date: 2015-05
• Subjects: Bayesian parameter estimation -- Crop modeling -- DSSAT -- Markov Chain Monte Carlo -- Metropolis-Hastings -- SALUS; Agricultural systems -- Periodicals; Agriculture -- Environmental aspects -- Periodicals; 338.16
• DOI: 10.1016/j.agsy.2014.12.003
• Languages: English
• ISSNs: 0308-521X

Abstract

Highlights: We estimate SALUS-Simple crop model parameters for maize, peanut and cotton. We evaluate the effect of data detail on the precision of estimated parameters. Data detail affects the variances of the posterior parameter distributions. Data detail affects the accuracy of model predictions obtained using the parameters. A stepwise approach reduces compensation errors during the parameter estimation. Abstract: The generic and simple version of SALUS (System Approach to Land Use Sustainability) crop model was recently integrated in the DSSAT (Decision Support System for Agrotechnology Transfer) cropping system model to provide an alternative approach to more complex crop models without the need for a detailed parameterization. A previous uncertainty and sensitivity analysis of the model (SALUS-Simple) established that accurate estimation of 15 of the 20 crop parameters required for predicting crop performance under water limitation was necessary to achieve reliable simulations. The present study used a Markov Chain Monte Carlo-based Bayesian stepwise approach for estimating crop parameters in SALUS-Simple using limited, end-of-season data (limited data case) and detailed in-season data (detailed data case). Independent testing were performed using data distributed with DSSAT version 4.5. Results of the detailed data case indicated that the estimated parameters resulted in smaller deviations between simulated and measured variables and in posterior parameter distributions with smaller variances. Independent testing showed that maize growth simulations (based on both data cases) were in good agreement with observations while peanut and cotton growth was simulated with mixed success. SALUS-Simple predictions using parameters estimated in the limited data case were concordant with observations for end-of-season biomass and yield, but simulations of in-season growth were degraded relative to the use of parameters estimated in the detailed data case. We conclude that the use of a sequential approach reduced compensation errors and, the use of a range of data types combined with a higher ratio between the number of data points and the number of estimated parameters significantly reduced uncertainties associated with the estimated parameters. Furthermore, model predictions based on mean parameter values can be regarded as reliable estimators of the expected values of the distributions of model predictions when an average prediction rather than a distribution is needed. Results from this study highlighted the principle that parameters estimated based on end-of-season data do not guarantee accurate prediction of in-season growth even if a Bayesian approach is used. The ability of the SALUS-Simple model to be parameterized or adapted for simulating canopy-level potential production of annual plants based on limited data is promising. Further testing of the model will help establish its response to different soils, climates and crops.