Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks
Boulos, D. (2020). Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks. EKB Journal Management System, 41(OCTOBER- DECEMBER), 435-445. doi: 10.21608/asejaiqjsae.2020.123803
Diia Boulos. "Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks". EKB Journal Management System, 41, OCTOBER- DECEMBER, 2020, 435-445. doi: 10.21608/asejaiqjsae.2020.123803
Boulos, D. (2020). 'Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks', EKB Journal Management System, 41(OCTOBER- DECEMBER), pp. 435-445. doi: 10.21608/asejaiqjsae.2020.123803
Boulos, D. Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks. EKB Journal Management System, 2020; 41(OCTOBER- DECEMBER): 435-445. doi: 10.21608/asejaiqjsae.2020.123803

Predicting Soil Productivity Resulted from Organic Matter Addition by Using Neural Networks

Alexandria Science Exchange Journal
Article 2, Volume 41, OCTOBER- DECEMBER - Serial Number 4, December 2020, Page 435-445  XML PDF (837.27 K)
Document Type: Original Article
DOI: 10.21608/asejaiqjsae.2020.123803
View on SCiNiTO View on SCiNiTO
Author
Diia Boulos email
Soil Physics - Desert Research Center - Cairo - Egypt
Abstract
Artificial neural networks (ANN) model is used for predicting some soil physical properties [soil bulk density (Bd), available water (AW), infiltration rate (I)], soil spinach productivity (Pro) and water use efficiency (WUE) under three different types of organic matter [Town refuse (TR), Farmyard manure (FYM) and Compost (COM)] with three rates [10, 15 and 20 ton/fed] for each treatment. Multilayer feedforward ANN with 8 neurons in input layer, 10 and 20 neurons for first and second hidden layers respectively and 5 neurons in output layer was trained using a backpropagation learning algorithm. The ANN model was trained with data collected from previous literatures (668 observations for training and 223 observations for testing). The model inputs were [Sand (S), Silt (Si), Clay (C), Town refuse (TR), Farmyard manure (FYM), Compost (COM), Electrical conductivity of irrigation water (EC) and Irrigation applied water (IR)]. The model outputs were [soil bulk density (Bd), available water (AW), infiltration rate (I), soil spinach productivity (Pro) and water use efficiency (WUE)]. Verification of the ANN model in prediction was done using field experimental data which carried out in El Sadat City (Data that ANN model has never seen before). Root mean square error (RMSE) and correlation coefficient (R2) were used to evaluate the ANN model. Validation and testing for the ANN model were done after careful and extensive training. The RMSE between measured and predicted values for soil bulk density (Bd), available water (AW), infiltration rate (I), soil spinach productivity (Pro) and water use efficiency (WUE) were 0.00909 Mg/m3, 0.10528 %, 0.23878 mm/h, 14.28973 kg/fed and 0.26762 kg/m3. While the R2 were equal to 0.99955, 0.99947, 0.99902, 0.99998 and 0.96883 respectively. The high R2 for output parameters recall indicated for excellent prediction of the ANN model for the data has never seen before.
 
Keywords
neural networks; soil physical properties; soil spinach productivity; water use efficiency
Statistics
Article View: 203
PDF Download: 280