Soil water prediction based on its scale-specific control using multivariate empirical mode decomposition

Soil water (SW) is controlled by different factors operating in different intensities and scales. The objective of this study was to apply multivariate empirical mode decomposition (MEMD) in revealing scale-specific control of SW. Two data sets from different climates were used. One data set was soil water storage (SWS) of 0–140 cm measured at two different periods (recharge and discharge periods) from a transect at St. Denis National Wildlife Area in a Canadian prairie area (SDNWA). The other data set was soil water content (SWC) of 0–6 cm from two transects (bunge needlegrass and korshinsk peashrub) in the Laoyemanqu watershed on the Chinese Loess Plateau (LYMQ). In both areas, five environmental factors including elevation, sand, silt, clay, and organic carbon (OC) contents were measured at each sampling location. SW and environmental factors were separated into different intrinsic mode functions (IMFs) and residue representing different scales. The dominant components in terms of the percentages of total variations in SW were identified. At each scale, SW was controlled by one or multiple factors. Each IMF of SW or residue can be predicted with the corresponding IMF or residue of some environmental factors. The summation of all predicted IMFs and residue predicted well the SW at the measurement scale, which outperformed SW prediction based on simple linear regression between SW and environmental factors and regression between IMFs of SW and factors at the measurement scale. Organic carbon was the major predictor for SWS in SDNWA for both periods and soil particle composition was the major predictor for SWC in LYMQ. MEMD has a great potential in revealing the scale-specific control of other soil properties.

► MEMD separates SW and factors to different components representing different scales.
► MEMD revealed strong influences of factors on SW at different scales.
► SW at each scale can be predicted well with factors from the similar scale.
► Overall SW prediction based on MEMD outperformed those based on traditional methods.