Skip to main content
ARS Home » Midwest Area » Morris, Minnesota » Soil Management Research » Research » Publications at this Location » Publication #285738
Title: A comparative study between nonlinear regression and artificial neural network approaches for modelling wild oat (Avena fatua) field emergence

Author
item CHANTRE, GUILLERMO - Universidad Nacional Del Sur (UNS)
item BLANCO, ANIBAL - Universidad Nacional Del Sur (UNS)
item Forcella, Frank
item VAN ACKER, RENE - University Of Guelph
item SABBATINI, MARIO - Universidad Nacional Del Sur (UNS)
item GONZALEZ-ANDUJAR, JOSE - Consejo Superior De Investigaciones Cientificas (CSIC)

Submitted to: Journal of Agricultural Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/20/2012
Publication Date: 1/23/2013
Citation: Chantre, G.R., Blanco, A.M., Forcella, F., Van Acker, R.C., Sabbatini, M.R., Gonzalez-Andujar, J.L. 2013. A comparative study between nonlinear regression and artificial neural network approaches for modelling wild oat (Avena fatua) field emergence. Journal of Agricultural Science. 152(2):254-262. doi: 10.1017/S0021859612001098.

Interpretive Summary: State-of-the-art computer models that assist crop consultants and others to predict when weeds emerge after crop sowing use a statistical technique called "non-linear regression" (NLR) to match the level of seedling emergence with an index that combines soil temperature and soil moisture into a single variable referred to as "soil hydrothermal time." An alternative mathematical approach is to use "artificial neural networks" (ANN) for the same purpose. Unlike NLR, ANN allow soil temperature and soil moisture to remain as separate and independent variables. Although the mathematics of ANN are more complicated than those of NLR, the independence of the two critical variables, soil temperature and soil moisture, permit greater flexibility and accuracy in predictions of the timing and extent of seedling emergence. An international team from Argentina, Canada, Spain, and the USA tested the accuracy of NLR and ANN using several data sets of wild oat emergence and associated soil microclimate from Manitoba (Canada); Minnesota, Montana, and North Dakota (USA); and South Australia (Australia) to develop and calibrate an emergence model for wild oat. Four independent data sets (Manitoba, Montana, North Dakota, and South Australia) were used to validate the model. Results indicated a much higher level of accuracy of ANN compared to NLR. These outcomes are of greatest utility for other researchers who develop models of emergence for weeds from around the world with the ultimate beneficiaries being crop consultants and advisors who use these models to make recommendations to growers regarding weed control.

Technical Abstract: Non-linear regression techniques are used widely to fit weed field emergence patterns to soil microclimatic indices using S-type functions. Artificial neural networks present interesting and alternative features for such modeling purposes. In this work, a univariate hydrothermal-time based Weibull model and a bivariate (hydro-time and thermal-time) artificial neural network were developed to study wild oat emergence under non-moisture restriction conditions using data from different locations worldwide. Results indicated a higher accuracy of the neural network in comparison to the non-linear regression approach due to the improved descriptive capacity of thermal-time and the hydro-time as independent explanatory variables. These outcomes suggest the potential applicability of the proposed modeling approach in the design of weed management decision support systems. Furthermore, due to specific ecological adaptations, the development of a single accurate predictive emergence model for A. fatua, applicable to a wide range of climatic environments, is difficult at the present time.