Soil solarization is a non-chemical method for disinfecting soil solely by solar radiation. The basic principle is to cover the soil with a mulching film during the hottest period of the year so that the soil temperatures increase to levels that are lethal to many soil-borne plant pathogens and weeds. The best results, however, are achieved in closed greenhouses, where temperature regimes in the soil are strongly influenced by the shape and dimensions of the greenhouse as well as by the characteristics of the greenhouse covering material. The modeling of thermal regimes in the soil during so-larization treatment is an important issue; it can be useful to estimate the required duration of the treatment in relation to the climatic conditions and the efficacy of the technique in reducing infections due to soil-borne pathogens. To this aim, several studies have modeled the physical processes of the soil-mulch-greenhouse system. The application and reliability of these models require accurate knowledge of the thermo-physical properties of each component of the system, which are sometimes difficult to measure. Neural network (NN) models represent an alternative and widely accepted method for studying physical problems and offer a way to tackle complex systems that may be difficult to define. However, until now, no work in the literature has described the use of NNs for this specific application. In this article, an innovative approach is proposed based on NN models that use as input only climatic data. The development and validation of the NN model were performed using data collected during soil solarization treatments carried out in two full-scale commercial greenhouses that differed in building features. The study showed that a multi-layer perceptron (MLP) network with one hidden layer of 60 neurons and continuous (sigmoid) transfer functions was very efficient. More specifically, the network used outside air temperature, outside solar radiation flux, and time of day as input variables and provided air temperature and solar radiation flux inside the greenhouse as well as soil temperatures at different depths as output variables. The results of the validation show that the modeled NNs estimate the output variables with high accuracy when trained at least once with data measured in the modeled greenhouse. The results for the modeled soil temperatures are particularly remarkable considering that the obtained precision is of the same order of magnitude as the accuracy of the sensors used in the field trials. The results obtained with the designed networks for cases different from those considered in the training can only be used as an indication because they are the outcome of an extrapolation. Nevertheless, the proposed NN model can be used as a reference for an NN of wider effectiveness obtained by training it on a large set of data from different case studies.

Neural Networks for Predicting Greenhouse Thermal Regimes During Soil Solarization

D'EMILIO, ALESSANDRO;PORTO, SIMONA MARIA;CASCONE, Giovanni
2012

Abstract

Soil solarization is a non-chemical method for disinfecting soil solely by solar radiation. The basic principle is to cover the soil with a mulching film during the hottest period of the year so that the soil temperatures increase to levels that are lethal to many soil-borne plant pathogens and weeds. The best results, however, are achieved in closed greenhouses, where temperature regimes in the soil are strongly influenced by the shape and dimensions of the greenhouse as well as by the characteristics of the greenhouse covering material. The modeling of thermal regimes in the soil during so-larization treatment is an important issue; it can be useful to estimate the required duration of the treatment in relation to the climatic conditions and the efficacy of the technique in reducing infections due to soil-borne pathogens. To this aim, several studies have modeled the physical processes of the soil-mulch-greenhouse system. The application and reliability of these models require accurate knowledge of the thermo-physical properties of each component of the system, which are sometimes difficult to measure. Neural network (NN) models represent an alternative and widely accepted method for studying physical problems and offer a way to tackle complex systems that may be difficult to define. However, until now, no work in the literature has described the use of NNs for this specific application. In this article, an innovative approach is proposed based on NN models that use as input only climatic data. The development and validation of the NN model were performed using data collected during soil solarization treatments carried out in two full-scale commercial greenhouses that differed in building features. The study showed that a multi-layer perceptron (MLP) network with one hidden layer of 60 neurons and continuous (sigmoid) transfer functions was very efficient. More specifically, the network used outside air temperature, outside solar radiation flux, and time of day as input variables and provided air temperature and solar radiation flux inside the greenhouse as well as soil temperatures at different depths as output variables. The results of the validation show that the modeled NNs estimate the output variables with high accuracy when trained at least once with data measured in the modeled greenhouse. The results for the modeled soil temperatures are particularly remarkable considering that the obtained precision is of the same order of magnitude as the accuracy of the sensors used in the field trials. The results obtained with the designed networks for cases different from those considered in the training can only be used as an indication because they are the outcome of an extrapolation. Nevertheless, the proposed NN model can be used as a reference for an NN of wider effectiveness obtained by training it on a large set of data from different case studies.
Greenhouse climate ; Multi-layer perceptron; Soil solarization
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/20.500.11769/10826
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 7
  • ???jsp.display-item.citation.isi??? 5
social impact