Taxonomic-based metrics, such as sensitive/tolerant species, taxonomic diversity indices, dominant species, etc., have been extensively used to indicate human disturbance in ecological assessments. However, these metrics would display variable associations with disturbance in both space and time due to inconsistent taxonomic identification or seasonal changes in community compositions. By comparisons, functional traits can be measured independently of species identification and could display strong and consistent linkages with human disturbance. In the present study, we surveyed benthic diatoms in 80 sites from 9 river networks located in Shenzhen of China, covering an extensive urban land use gradient, in the wet and dry seasons of 2021. Two algal experts identified and counted diatoms with one person responsible for one season’s samples. Then, diatom taxonomic metrics and functional traits were calculated to test the hypothesis that functional traits will display stronger and more consistent relationships with urban disturbance (indicated by % impervious surface area, %ISA) than of taxonomic metrics. We found that diatom community compositions and four species diversity indices all displayed substantial seasonal variation. Among measured diversity indices, only species richness responded significantly to %ISA change in both seasons. Fifteen species had average relative abundance >1% during our survey, in which five species were sensitive to %ISA changes in either season with two species displaying consistent responses along the %ISA gradient in both seasons. Considering functional traits, there was no evident seasonal variation in trait compositions. Percentages of attached growth form, cell size 2 and 3, and motile and plankton guilds responded significantly to %ISA changes in both seasons. Most of these traits were more responsive to %ISA than taxonomic-based metrics. Our findings imply that diatom functional traits were not sensitive to inconsistent identification and seasonal changes. Therefore, diatom functional traits may be better choices than taxonomic metrics for indicating human disturbance.