A Weighted Meta-Analysis of Himalayan and Karakoram Glacier Mass Balance and Area Change: Heterogeneity Decomposition, the Karakoram Anomaly and the Method Effect

Abstract

Published estimates of glacier change in the Himalaya and Karakoram are usually compared as unweighted means or by narrative contrast, so the size of the regional differences and the share of disagreement that reflects real spatial variability rather than measurement noise remain unquantified. Here we apply a formal random-effects meta-analysis to published estimates of specific mass balance (k = 56 estimates from 21 source publications, observation periods spanning 1966 to 2019) and glacier area change (k = 16 estimates from 10 publications). Models were fitted by restricted maximum likelihood with source publication as a clustering level, and heterogeneity was decomposed with Cochran's Q, I2 and meta-regression on sub-region, period midpoint, method, spatial scale and period length. The pooled Himalaya-Karakoram mass balance was -0.28 m w.e. a−1 (95% CI -0.37 to -0.19), with high heterogeneity (I2 = 83%). The Karakoram pooled to -0.06 m w.e. a−1 (-0.10 to -0.02) with no detectable heterogeneity, and differed from the Himalayan mean by +0.34 m w.e. a−1 over matched post-2000 periods (p < 0.001), a difference that survived every sensitivity analysis. Sub-region alone explained about three quarters of the between-estimate variance. Contrary to expectation, the between-study comparison found no significant shift after 2000 in the Himalaya (-0.07 m w.e. a−1, p = 0.31), although within-study pre/post comparisons pointed the same way (-0.14 m w.e. a−1, p = 0.06), indicating that between-study heterogeneity is large enough to mask a real but modest acceleration. Glaciological estimates were more negative than geodetic estimates in the crude comparison (-0.31 m w.e. a−1, p = 0.026), but the difference disappeared once spatial scale was controlled (+0.04 m w.e. a−1 at glacier scale, p = 0.87), which suggests that the well-known negative bias of in situ series is a selection effect of small, low-lying monitored glaciers rather than a method artefact. Area change pooled to -0.32 % a−1 but with extreme heterogeneity (I2 = 98%) and strong small-study asymmetry, so weighted synthesis of this outcome is not yet robust. The results provide the first weighted regional baselines for the Himalaya and Karakoram and show that reporting of uncertainty, not the number of studies, is now the limiting factor for synthesis.