The growing penetration of variable renewable energy sources (VRES) requires additional flexibility reserve to cope with the uncertainty in power system operation. Current industrial practice typically assumes a certain fraction of the VRES production forecast power as flexibility reserve, even though the VRES variability and uncertainty is a function of weather conditions. Therefore, this paper focuses on weather-driven flexibility reserve sizing and allocation for large-scale wind power installations. First, we propose a method, which generates statistically credible wind power forecast errors based on forecasts of various weather features, thus stressing a given wind power forecast. Then, these errors are mapped into a risk-based reserve requirement, which is then compared with the current extent-based and probability-based requirements. Additionally, the risk-, extent-, and probability-based reserve requirements are allocated to compare their cost and deliverability performance. Throughout the paper, we use real-world data to compute weather-driven flexibility reserve requirements and evaluate their performance using numerical experiments on a 1819-bus NYISO system model with both on- and off-shore wind power installations.