This study aims to improve the localization and assimilation of satellite observations in the visible and infrared spectral ranges to enhance predictions of clouds and convective processes. Understanding correlation structures between satellite observations and atmospheric state variables is crucial for successful data assimilation. We focus on examining vertical ensemble-based correlations from Himawari-8 channels (VIS0.64 or IR7.35) and tackle the challenge of vertical observation-space localization. Traditional distance-based localization methods are often suboptimal due to the multi-layered origin of observed radiation. We present empirical optimal localization (EOL) functions derived from a 1000-member ensemble convective-scale simulation to address this issue. Our research highlights the need for channel-specific and variable-specific localization strategies, emphasized by our analysis of two summer case studies that exhibit substantial situational variability in correlation structures, especially in the visible spectral range. Further, we explore various predictors for formulating dynamic, situation-specific vertical localization strategies, offering insights into their effectiveness and potential for advancing convective-scale satellite data assimilation.