The spatiotemporal variability of soil moisture (θ) has rarely been studied at the field scale across different seasons and sites. Here, we utilized 9 months of θ data in two semiarid ecosystems of North America to investigate the key relationship between the spatial mean (〈θ〉) and standard deviation (σ _θ ) at the field-scale (~100 m). Analyses revealed a strong seasonal control on the σ _θ versus 〈θ〉 relation and the existence of hysteretic cycles where wetting and dry-down phases have notably different behavior. Empirical orthogonal functions (EOFs) showed that θ variability depends on two dominant spatial patterns, with time-stable and seasonally varying contributions in time, respectively. Correlations between EOFs and land surface properties also indicated that θ patterns are linked to vegetation (terrain and soil) factors at the site with higher (lower) vegetation cover. These physical controls explained the observed hysteresis cycles, thus confirming interpretations from previous modeling studies for the first time.