A mechanistic understanding of soil respiration is a major impediment to predicting terrestrial C fluxes spatially and temporally. Automated measurements of soil respiration offer the high-resolution information necessary to observe temporal variation in soil respiration, but spatially these measurements are under-represented in water-limited and non-forested ecosystems. We measured soil respiration with automated chambers over the growing season, at two sites with the same semi-arid climate, but with different dominant vegetation, perennial grasses and shrubs in the Owens Valley, CA, USA. An isotope mass balance technique was used to partition soil respiration into autotrophic and heterotrophic components at two time points, early and late growing season. Results showed large differences in the magnitude of growing season soil respiration between the two sites (910 versus 126 g C m−2 for grasses and shrubs respectively over 5 months). We attribute this to site differences in soil water availability and belowground allocation and productivity. Diel patterns of soil respiration between the two sites were similar. Temperature explained most of the diel variability in the early growing season, when soil moisture was greatest. As soil moisture declined over the growing season, diel patterns became increasingly decoupled temporally from temperature due to increased water-limitation on surface heterotrophic sources and hypothesized strong photosynthetic control over soil respiration rates. Partitioning of soil respiration into autotrophic and heterotrophic sources showed the dominance of autotrophic sources across seasons and ecosystems. However, heterotrophic respiration was more dynamic from early to late growing season, declining in the grass ecosystem; and a surprising increase in the shrub ecosystem, attributed to warming of the soil profile enhancing microbial decomposition at depth.