Global weather forecasts are of great economic value for society, but geographical differences in forecast accuracy can create new - and potentially exacerbate existing - economic inequalities. Regional differences in forecast accuracy are particularly relevant if weather forecasts are considered as an important tool to help reduce the negative effects of future climate change such as mortality from extreme temperature events. In this paper, we provide a comprehensive global analysis of the accuracy of short-term numerical weather predictions of temperature and relate our findings to both existing economic inequalities and inequities in global weather monitoring infrastructure. We report three main results: First, temperature forecasts are currently substantially more accurate in high income countries than in low income countries. A seven-day-ahead forecast in a high-income country is on average more accurate than a one-day-ahead forecast in a low income country. Second, while forecast accuracy has improved steadily between 1985 and the present - with the largest increases in the 1990s - there is a persistent gap between high income and low income countries. Third, the infrastructure for weather observations is highly unequally distributed across countries, with fewer land-based weather stations and radiosondes in poorer countries. These inequalities grow even larger when lower reporting rates are taken into account. Remedying these differences in infrastructure would help close the forecast accuracy gap.