Information retrieval from spatiotemporal data cubes is key to earth system sciences. Respective analyses need to consider two fundamental issues: First, natural phenomena fluctuate on different time scales. Second, these characteristic temporal patterns induce multiple geographical gradients. Here we propose an integrated approach of subsignal extraction and dimensionality reduction to extract geographical gradients on multiple time scales. The approach is exemplified using global remote sensing estimates of photosynthetic activity. A wide range of partly well interpretable gradients is retrieved. For instance, well known climate-induced anomalies in FAPAR over Africa and South America during the last severe ENSO event are identified. Also, the precise geographical patterns of the annual– seasonal cycle and its phasing are isolated. Other features lead to new questions on the underlying environmental dynamics. Our method can provide benchmarks for comparisons of data cubes, model runs, and thus be used as a basis for sophisticated model performance evaluations.