The largest and most dynamic agricultural frontier in Brazil is known as MATOPIBA, an area that covers part of the Cerrado biome. Within this region, Western Bahia stands out as a large producer of soy and cotton. There are no studies that quantify carbon stocks for different land uses and land cover types in Western Bahia, which hinders comprehension of the role of agricultural expansion in carbon dynamics and the development of sustainable agriculture policies. Here, we evaluate how the land use changes in this region have affected the carbon balance in the aboveground biomass (AGB), belowground biomass (BGB), and soil reservoirs. We collected soil samples for areas with different land uses and land cover types to estimate soil carbon stocks (SCS) and combined remote sensing results and modeling techniques to develop a historical reconstruction of spatial patterns of SCS, AGB, and BGB during the period 1990-2018. The replacement of areas from the forest formations class with pasture and rainfed agriculture reduced the 100 cm depth SCS (SCS100) by 37.3% (p = 0.031) and 30.3% (p = 0.053), respectively. By contrast, the conversion of pasture and rainfed agriculture to irrigated agriculture increased SCS100 by 34% (p = 0.034) and 26.5% (p = 0.022), respectively. Spatial changes in historical carbon stocks are strongly associated with land use changes that occurred between 1990 and 2018. We estimated a non-significant loss of 61.9 Tg-C (p = 0.726) from the total carbon stocks (calculated as the sum of AGB, BGB, and SCS) of which 80% of the losses came from soil stocks, 11% from BGB, and 8% from AGB. These findings reveal the need to monitor carbon stocks in sandy soils to reduce the uncertainties of estimates and support the development of effective sustainable agriculture policies. The best alternatives for reducing carbon losses in the Cerrado are to maintain natural forest cover and to recover soils through sustainable soil management, especially in pasturelands where soil carbon stocks are lowest.