We examine the spatial patterns of near-surface air temperature (T_a) over a melting glacier using a multi-annual dataset from McCall Glacier, Alaska. The dataset consists of a 10-year (2005-2014) meteorological record along the glacier centreline up to an upper glacier cirque, spanning an elevation difference of 900 m. We test the validity of on-glacier linear lapse rates, and a model that calculates T_a based on the influence of katabatic winds and other heat sources along the glacier flow line. During the coldest hours of each summer (10% of time), average lapse rates across the entire glacier range from -4.7 to -6.7°C km^-1, with a strong relationship between T_a and elevation (R² > 0.7). During warm conditions, T_a shows more complex, non-linear patterns that are better explained by the flow line-dependent model, reducing errors by up to 0.5°C compared with linear lapse rates, although more uncertainty might be associated with these observations due to occasionally poor sensor ventilation. We conclude that T_a spatial distribution can vary significantly from year to year, and from one glacier section to another. Importantly, extrapolations using linear lapse rates from the ablation zone might lead to large underestimations of T_a on the upper glacier areas.