Application of regression kriging and machine learning methods to estimate soil moisture constants in a semi-arid terrestrial area

Tuncay T., Alaboz P., Dengiz O., Baskan O.

COMPUTERS AND ELECTRONICS IN AGRICULTURE, cilt.212, 2023 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 212
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1016/j.compag.2023.108118
  • Dergi Adı: COMPUTERS AND ELECTRONICS IN AGRICULTURE
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Aerospace Database, Applied Science & Technology Source, BIOSIS, CAB Abstracts, Communication Abstracts, Compendex, Computer & Applied Sciences, Environment Index, Food Science & Technology Abstracts, INSPEC, Metadex, Veterinary Science Database, Civil Engineering Abstracts
  • Ondokuz Mayıs Üniversitesi Adresli: Evet

Özet

In the current study, the use of regression-kriging (RK), artificial neural networks (ANN), support vector machines (SVM), and random forest (RF) methods from machine learning algorithms, were used to estimate field capacity (FC), permanent wilting point (PWP), available water content (AWC) and their performance was compared. A data set obtained from 354 surface soil samples taken randomly, mostly from agricultural areas is used. The soil data set includes pH, EC, calcium carbonate equivalent (CaCO3 equivalent), particle size distribution, and bulk density (BD) values. The results showed that while FC showed a negative strong correlation (p < 0.001) with sand (r:-0.69), BD (r:-0.85), and silt (r:-0.47), it showed a positive strong correlation (p < 0.001) with C (r: 0.90). Similarly, PWP showed a negative strong correlation with (p < 0.001) sand (r:-0.73), BD (r:0.88), and silt (r:-0.42) but a positive strong correlation (p < 0.001) with C (r: 0.90). While AWC showed a negative strong correlation (p < 0.001) with sand (r:-0.61), BD (r:-0.76), it found a positive strong correlation (p < 0.001) with FC (r: 0.97), clay (r: 0.83), and PWP (r: 0.74). In the stepwise regression results showed that particle size were prominent as the most important factor in the regression equation created for FC, PWP and AWC. Moreover, FC is the most important factor to predict AWC. For the soil FC, ANN was best with excellent accuracy (RPD = 2.71), followed by SVM (2.42), RF (2.21) while RK was poor accuracy (1.10 and 1.04). Similarly, among the machine learning algorithms (RF and SVM), ANN obtained superiority by producing lower RRMSE (7.84%), RMSE (2.83%), MAE (2.37%), MAPE (7.45%), with the largest Lin's concordance correlation coefficient (LCCC) (0.961) compared to other methods. For PWP and AWC, ANN was the best algorithm with excellent and good accuracy RPD 3.17 and 1.95 respecively. In addition, other machine learning algorithms have been the same value range in terms of LCCC. Therefore, we recommend the ANN machine-learning algorithm is more favorable to predict FC, PWP and AWC than both RK and other machine learning methods.