Abstract: The overall regional poverty in China was eliminated in 2020, but the relative poverty in the country will still exist for a long time. Therefore, it is necessary to conduct a long-term measurement and development analysis of poverty in poverty-stricken areas. However, conventional measurement methods based on socio-economic data have severe limitations. With four provinces (municipalities) in southwestern China as a case study, this study built a back propagation (BP) neural network model based on the particle swarm optimization algorithm and a nighttime light (NTL) dataset of long time series from 2000 to 2019 first. Then, this study constructed the multi-dimensional poverty indices based on socio-economic and geographical data to reflect the poverty in counties. Finally, this study established a poverty measure model by combining the long-time-series NTL data with the multidimensional poverty indices and produced the nighttime light multidimensional poverty index (NLMPI). Based on the NLMPI, the measure and spatial-temporal evolution analysis of poverty in counties were carried out. The study results are as follows. The NLMPI indicates that the four provinces (municipalities) in southwestern China had significantly differentiated multidimensional poverty in 2000. However, the proportion of counties at extremely low and low levels decreased while that of moderate-level counties increased owing to the national poverty alleviation efforts. From 2000 to 2019, the NLMPI of counties in southwestern China showed a positive spatial autocorrelation and the Moran’s I index showed a downward and then an upward trend. These results indicate that poverty aggregation weakened from 2000 to 2010 and poverty alleviation dispersed thereafter in the four provinces (municipalities) in southwestern China. The local spatial autocorrelation results indicate that the multi-dimensional poverty pattern in southwestern China was alleviated but unbalanced. This pattern was reflected in the high NLMPI values surrounded by high NLMPI values (the H-H aggregation type) in Chengdu-Chongqing, Kunming, and Guiyang and in the low NLMPI values surrounded by low NLMPI values (the L-L aggregation type) in northwestern Sichuan and western Yunnan. This study highlights the application of NTL data in research on regional poverty.