Weekly soil moisture forecasting with multivariate sequential, ensemble empirical mode decomposition and Boruta-random forest hybridizer algorithm approach

Prasad, Ramendra and Deo, Ravinesh C. and Li, Yan and Maraseni, Tek (2019) Weekly soil moisture forecasting with multivariate sequential, ensemble empirical mode decomposition and Boruta-random forest hybridizer algorithm approach. Catena, 177. pp. 149-166. ISSN 0341-8162

Abstract

Soil moisture forecasts are vital for environmental monitoring, the health of ecological systems, hydrology, agriculture and understanding the soil characteristics. In this study, we design a new multivariate sequential predictive model that utilizes the ensemble empirical mode decomposition (EEMD) algorithm hybridized with extreme learning machines (ELM) to forecast soil moisture (SM) over weekly horizons. The EEMD data pre-processing utility is a self-adaptive tool, which does not require a predefined basis function and avoids frequency-mode mixing issues. The proposed multivariate sequential EEMD model is designed to sequentially demarcate model predictor variables and the target (SM) into analogous intrinsic mode functions (IMFs) and a residue component using the EEMD process, to address the complexities and associated non-linearities in hydrologic-based inputs. To validate the new approach, four diversely characterized sites in Australia's Murray-Darling Basin are purposely selected where 13 weekly hind-casted predictors are collated from the Australian Water Availability Project WaterDyn physical model. After the sequential EEMD transformation process, a two-stage feature selection employing cross-correlation and random forest based Boruta wrapper algorithm is adopted to extract pertinent features from the hydro-meteorological predictor series to construct a hybridized multivariate sequential EEMD-Boruta-ELM model. Comprehensive model evaluation using statistical metrics and diagnostic plots against alternative methods: hybrid multivariate adaptive regression splines (MARS) (EEMD-Boruta-MARS) and classical MARS and ELM, establish the superiority of hybrid EEMD-Boruta-ELM model, yielding relatively low errors and high performance. The study ascertains that the EEMD-Boruta-ELM hybrid model can be explored as a pertinent data-driven tool for relatively short-term soil moisture forecasts, thus advocating its practical use in near real-time hydrological and pedological applications.


Statistics for USQ ePrint 36181
Statistics for this ePrint Item
Item Type: Article (Commonwealth Reporting Category C)
Refereed: Yes
Item Status: Live Archive
Additional Information: Permanent restricted access to Published version, in accordance with the copyright policy of the publisher.
Faculty / Department / School: Current - Faculty of Health, Engineering and Sciences - School of Agricultural, Computational and Environmental Sciences
Date Deposited: 12 Mar 2019 23:49
Last Modified: 13 Mar 2019 00:13
Uncontrolled Keywords: multivariate sequential ensemble empirical mode decomposition; Boruta feature selection; soil moisture; Murray-Darling basin; extreme learning machine
Fields of Research : 08 Information and Computing Sciences > 0801 Artificial Intelligence and Image Processing > 080108 Neural, Evolutionary and Fuzzy Computation
07 Agricultural and Veterinary Sciences > 0701 Agriculture, Land and Farm Management > 070101 Agricultural Land Management
Socio-Economic Objective: E Expanding Knowledge > 97 Expanding Knowledge > 970107 Expanding Knowledge in the Agricultural and Veterinary Sciences
Identification Number or DOI: 10.1016/j.catena.2019.02.012
URI: http://eprints.usq.edu.au/id/eprint/36181

Actions (login required)

View Item Archive Repository Staff Only