|
|
Bimonthly Since 1986 |
ISSN 1004-9037
|
|
|
|
|
Publication Details |
Edited by: Editorial Board of Journal of Data Acquisition and Processing
P.O. Box 2704, Beijing 100190, P.R. China
Sponsored by: Institute of Computing Technology, CAS & China Computer Federation
Undertaken by: Institute of Computing Technology, CAS
Published by: SCIENCE PRESS, BEIJING, CHINA
Distributed by:
China: All Local Post Offices
|
|
|
|
|
|
|
|
|
|
Abstract
The application of optimal soil nutrients is considered one of the challenging problems faced by farmers mainly because of its dependence on the spatial variability of soil elements. Experimental estimation of soil elements across all agricultural farms is a hard problem to address. Although there are several studies proposed to estimate soil elements employing statistical, geostatistical, computational, or by using AI techniques, most of these methods either replete computational resources or lack accuracy. All these factors point toward the need for a novel method that is more accurate and uses less computational resources for the accurate prediction of soil nutrients. In this study, we ensembled Machine Learning Regression algorithms and Geostatistical estimations for the first time to develop models to predict Soil Organic Carbon (SOC), one of the most important soil elements. Soil nutrient data (2648 samples) pertaining to Alappuzha District, Kerala, India, collected during 2019-20 was used for the study. Geo-environmental predictors generated from remote sensing data such as topography, vegetation, land surface temperature, and precipitation were used for developing the prediction model. Predictions of Geostatistical and Machine Learning models were used as features for stacked ensemble modeling using the Stochastic Gradient Boosting algorithm. Models were generated and a repeated 10-fold cross-validation technique was used to evaluate the model performance. The stacked ensemble model resulted in very good prediction accuracy with an R2 of 81 % and the lowest RMSE of 0.13. The results showed that the ensembling of MLA and Geostatistical predictions considerably improved the prediction accuracy of the SOC when compared to traditional methods. Finally, the prediction model was applied to a data frame with a 200 X 200-meter spatial interval, and the prediction results were visualized in a geospatial framework.
Keyword
Spatial Prediction; Geostatistical Estimation; Machine Learning; Ensemble Modeling; Soil Organic Carbon
PDF Download (click here)
|
|
|
|
|