Mid-Latitude Versus Tropical Scales of Predictability and Their Implications for Forecasting

Weather predictability varies between tropical and middle latitudes: rotational effects enable forecasts on moderate spatial scales up to 10 days in middle latitudes, while longer term predictions are less reliable; in contrast, tropical weather is challenging to predict at short lead times, but seasonal forecasts are more accurate due to the influence of larger-scale oscillations, such as slowly varying oceanic surface conditions. This behaviour has been demonstrated in previous studies, but has yet to be focused on in detail, despite its importance to the development of forecasting systems in Tropical regions. This study systematically evaluates precipitation in weather prediction models across both regions using the fractions skill score, evaluating performance at progressively longer lead times and averaging scales, and compares the results with an evaluation based on upper air error kinetic energy. The results confirm that the prediction systems perform better on smaller scales and shorter lead times at middle latitudes and on larger scales and longer lead times at tropical latitudes. A “crossover” in performance is seen at forecast lead times of 5–7 days, a result that appears to be consistent across a range of model resolutions, and occurs both when specifically comparing European and African domains and when comparing whole latitude bands. This differential pattern of model skill even occurs for machine learning-based forecast models, suggesting that it is a fundamental property of the atmosphere rather than an effect of the construction of currently used operational forecasting systems. These findings highlight the need for different forecasting methodologies in tropical regions to address the lack of short-term predictability and leverage long-term statistical predictability.