The　spatiotemporal　non-stationary　relationships　between　2D/3D　urban　features　and　land　surface　temperature　(LST)　introduce　uncertainty　to　the　quantitative　exploration　between　them.　This　study　focused　on　the　urban　building　spaces　of　＂furnace　city　＂　Fuzhou　and　explored　the　quantitative　relationships　between　urban　features　and　ECOSTRESS　diurnal　LSTs　from　a　block　perspective.　Our　results　revealed　that:　(1)　Compared　to　the　ordinary　least　squares　regression　model,　the　multi-scale　geographically　weighted　regression　model　can　better　capture　the　spatiotemporal　non-stationary　relationships.　(2)　Largest　patch　index　of　building　patches　(LPI_B)　and　building　height　(BH)　have　the　greatest　impact　on　the　variations　in　daytime　and　nighttime　LSTs,　respectively.　The　interaction　between　largest　patch　index　of　vegetation　patches　(LPI_V)　and　LPI_B　has　the　largest　enhancing　effect　on　daytime　LST,　while　that　between　BH　and　LPI_B　enhances　nighttime　LST　the　most.　(3)　The　diversification　of　architectural　morphology　highlights　the　equal　importance　of　both　2D　and　3D　building　features　in　influencing　LST　variations.　Meanwhile,　the　standardization　of　urban　greening　emphasizes　the　greater　significance　of　2D　vegetation　features　compared　to　3D.　(4)　Based　on　varying　spatial　characteristics,　differentiated　urban　renewal　schemes　should　be　adopted.　These　findings　can　deepen　our　understanding　of　spatiotemporal　non-stationarity,　which　cannot　be　ignored　in　urban　thermal　environment　research.