Accurate　and　timely　estimation　of　gross　domestic　product　(GDP)　is　essential　for　evaluating　economic　development.　Nighttime　light　(NTL)　data　effectively　estimate　subindustry　GDP,　yet　previous　studies　relied　on　single　panchromatic　bands.　Whether　multispectral　nighttime　remote　sensing　data,　detecting　spectral　differences　from　economic　activities,　improves　subindustry　GDP　estimates　remains　unverified.　This　article　leverages　multispectral　NTL　and　thermal　infrared　data　from　the　SDGSAT-1　satellite,　combined　with　land　cover　data,　to　estimate　subindustry　GDP　using　machine　learning　models.　We　compare　support　vector　machines,　neural　networks,　and　random　forest　(RF),　identifying　RF　as　the　optimal　model　due　to　its　lowest　RMSE　values　(9.16,　171.06,　and　180.51　for　primary,　secondary,　and　tertiary　industries,　respectively).　Empirical　results　demonstrate　that　multispectral　SDGSAT-1　data　significantly　outperforms　its　single　panchromatic　band　counterpart,　improving　R-2　values　for　secondary　and　tertiary　industries　from　0.58　to　0.88　and　0.68　to　0.90,　respectively.　Compared　to　VIIRS　NTL　data,　SDGSAT-1　further　reduces　spatial　misdistribution　over　farmland　and　industrial　zones,　achieving　a　7.7%　R-2　improvement　at　smaller　scale　(industrial　parks　level).　Key　factors　driving　GDP　estimation　vary　across　industries:　cropland　area　dominates　for　the　primary　industry;　thermal　infrared　and　red　light　intensity　for　the　secondary　industry;　and　blue　light　intensity　for　the　tertiary　industry.　These　findings　validate　the　superiority　of　multispectral　NTL　data　in　subindustry　GDP　estimation　and　offer　actionable　insights　for　enhancing　urban　economic　monitoring　and　policy　formulation.