Table 2 (a) Spatially cross validated (CV) results for energy access at microregions for 2013 and comparison with existing works. (b) Temporally validated nowcasts for 2020 with DHS-2019 data aggregated at regions and comparison with existing works. Corr. refers to Pearson’s r correlation, Rank corr. refers to Spearman’s rank correlation and RMSE refers to Root Mean Square Error, which is disaggregated by urban and rural areas. The target values are normalized between 0 and 1. The p-values for all targets is less than 0.001, unless annotated with a ‘*’. RMSE is in same units as the target, with lower values indicating a better fit. The columns highlighted Linear (EO) and Linear (NL) are comparative methods, where the former denotes a linear model with our covariates and the later denotes a linear model with only nighttime lights data. Results for past approaches on all covariates are reported in columns GPR (Gaussian Process Regression [12]) and GBR (Gradient Boosted Regression [25]). Spatial CV procedure was performed 100 times with different train-test split of the data, the standard deviation across the multiple runs is reported within simple brackets. In temporal validation, the testing data belonged to 2020 and the training data to 2013, 2015, and 2017. Since this procedure was run a single time, there are no standard deviations to be reported

From: Accurate intercensal estimates of energy access to track Sustainable Development Goal 7