Please use this identifier to cite or link to this item:
Authors: Priya Bhattacharya
Advisor: Pragati Pramanik Maity
Title: Prediction of soil health indicators under conservation and conventional agricultural practices using different machine learning approaches
Language: en_US
Type: Thesis
Agrotags: null
Abstract: production systems can increase soil erosion, decrease soil health and water quality, and the ability to attain sustainable agricultural production systems. In the current study, conservation agriculture (CA) and conventional tillage (CT) fields existing in different villages of Nilokheri block of Karnal district, Haryana were surveyed and soil samples were collected for analysis of soil physical, chemical and biological health indicators. Prediction of soil health indicators were done by multilinear regression (MLR) and Artificial Intelligence (AI) based machine learning approaches like Artificial Neural Network (ANN) and Support Vector Machine (SVM).The results showed that bulk density (BD) of 0-15 cm soil layer was 7% higher in CA 3 yrs. plot whereas in CA 6yrs. And CA 9 yrs., BD values were 2% and 3% higher than CT. In 15-30 cm soil layer average BD values were 3.4 to 6.5% higher than the surface 0-15 cm soil layer. Total organic carbon (TOC) in 0-15 cm soil layer was found to be greater by 40% in CA 3 yrs. plot whereas it was higher by 55.1 and 67% in CA6 yrs. and CA 9 yrs. as compared to CT. The changes in soil water content at field capacity (SWCFC) were not significant for 0-15cm soil layer for all the treatments and The saturated hydraulic conductivity (SHC) increased in CA 9 yrs. treatment by 7.34 cm/hr as compared to CT which was significant. In CA for 0-15 cm and 15-30 cm soil depth, labile pools were 36 and 22% more than CT and in both the soil layer, in CA plots recalcitrant pool was 12 and 9% more than CT. Increase in microbial biomass carbon (MBC) values of 0-15cm soil layer over CT were 18.57, 47.08, and 71.5 % for CA 3 yrs., CA6 yrs. and CA 9 yrs., respectively. In prediction of SHC, SWCFC and mean weight diameter (MWD), the means and medians were comparable for each parameter, and the skewness and kurtosis coefficients were close to zero for SHC, MWD sand, silt, clay, OC and BD. These results indicated that the frequency distributions of each parameter were symmetrical and apparently stationary (in relation to mean). ANN with two neurons in hidden layer had better performance in predicting SHC and MWD than multi- linear regression. And ANN with four hidden layer was best in predicting SWCFC because of its non-linear function. SVM was best model with lowest root mean square error (RMSE) (0.103 and 0.12 for 96 training and testing datasets) and mean absolute percent error (MAPE) values (10.96 and 16.91% for training and testing data) in prediction of MWD. Results of Friedman test showed that performance of SVM in prediction of MWD was significantly better as compared to ANN and SVM but performance of SVM (in case of SHC) and ANN (in case of SWCFC) was better but not significant. Plots under CA showed sustainability in different village but CT system were sustainable with high input use. Technique for order of preference by similarity to ideal solution (TOPSIS) scoring showed that among 11 indicators used, organic carbon (OC) had maximum weightage (0.96) towards sustainability. It can be concluded that the use of SVM should be extended to different problems for constructing some rules for future applications and sustainable system with high input use should be studied in details as these systems are more vulnerable to any changes like climate and agricultural management practices. 97
Description: T-10090
Subject: Agricultural Physics and Meteorology
Theme: Prediction of soil health indicators under conservation and conventional agricultural practices using different machine learning approaches
These Type: M.Sc
Issue Date: 2019
Appears in Collections:Theses

Files in This Item:
File Description SizeFormat 
  Until 2020-09-18
1.93 MBAdobe PDFView/Open Request a copy

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.