The effect of up-scaling soil properties and model parameters on predictive accuracy of DSSAT crop simulation model under variable weather conditions

• Published in: Geoderma Vol. 287; pp. 105 - 115
• Main Authors: Fry, Jessica, Guber, Andrey K., Ladoni, Moslem, Munoz, Juan D., Kravchenko, Alexandra N.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.02.2017
• Subjects: DSSAT-CSM model; Field scale; Soybean yield; Spatial variability; Upscaling; Spatial variability; DSSAT-CSM model; LTER KBS; TOC; RMSE; CN; Upscaling; ET; P; Field scale; DSSAT-CSM; Soybean yield; RAE; SSURGO
• DOI: 10.1016/j.geoderma.2016.08.012
• ISSN: 0016-7061

The development of environmentally and economically sound long term agricultural practices under changing climate conditions at the field scale requires implementation of predictive models that assess short and long term responses of agricultural systems to changing environmental conditions. Soil properties for such models are commonly taken from local or National Soil Surveys (SSURGO, STATSGO). This causes a mismatch between the modeling and data source scales. Different techniques can be implemented to upscale soil properties measured at plot scale for field-scale crop modeling, however little is known about the effect of scaling on the accuracy of crop models in predicting crop yields. The objective of this work was to examine: (i) the spatial variability of soybean yields across an agricultural field with different soils and inputs; (ii) how spatial variability in soil properties translates into the variability of measured and predicted yields; (iii) how scaling soil properties affects model accuracy in predicting soybean yields for different weather conditions. The study was conducted at LTER KBS in Michigan, USA. Soybean yield was measured at 2.2ha and 4.9ha fields at 2×5m resolution in 2010. Soil properties were measured at the plot scale in 19 locations, selected to represent two soil types and two management practices on 3 topographical elements (i.e. summit, slope and depression). The DSSAT-CSM was calibrated and validated on soybean yield data measured at the plot scale in 2010 and 2013. The validated model was used to generate yields for 22years with varying weather conditions in all selected locations. Then the model parameters were scaled up using different techniques, such as averaging plot-scale measured soil properties, averaging the model parameters estimated for each plot using measured soil properties, and using typical soil profile descriptions and the SSURGO soil database. The results of this study showed high variability in soybean yield for two soils and two management practices, which was associated with variability in soil texture and organic carbon content in the top 20-cm soil layer, but not with surface topography. Despite considerable differences in parameters, all upscaling techniques performed reasonably well for different weather conditions However, model performance appeared to be site specific in this study. •Spatial variability in soil properties and soybean yield was studied in field.•Calibrated and validated DSSAT-CSM generated yield for 22 weather scenarios.•Four techniques for upscaling DSSAT-CSM parameters were tested to predict yield.•All upscaling techniques provided reasonable estimates of yield at field scale.•The model was sensitive to runoff curve numbers, but was not to soil properties.