Estimating Field Capacity from Volumetric Soil Water Content Time Series Using Automated Processing Algorithms

Abstract

We evaluated competing approaches for automated soil water cycles analysis that use widely available R packages based on pattern recognition and machine learning (findpeaks [R‐FP], symbolic aggregate approximation [R‐SAX], and density histogram [R‐DH]), and a MATLAB code based on soil water dynamic principles (SWDP). These approaches were applied to three SMS datasets. Our empirical results showed superiority of R‐SAX for identifying valid soil water cycles, probably due to benefiting from training sets to calibrate to correct cycles. Two other approaches (SWDP and R‐FP) provided similar results without need of training sets or preprocessing data. Three approaches for estimating field capacity were applied to valid cycles, R‐FP, regression of exponential decay (SWDP‐R), and estimated “knee” of curve (SWDP‐K). Each performed similarly to the expert defined values, with R‐FP and SWDP‐R generally performing best across analyses. Results of this study also show temporal dynamics of θFC within datasets used here. There is potential for optimizing θFC and a need for automated, objective analysis to leverage dynamics in irrigation management and modeling.