The unequal global distribution of weather forecast accuracy and the value of ground-based observations

Abstract

<p class="western">Recent appraisals of the value of accurate weather forecasts for labor supply decisions (Song 2022), heat mortality (Shrader et al. 2022), and in the construction sector (Downey et al. 2022) show that they exceed multiple times their production costs. Given the large economic benefits of accurate weather forecasts, differences in forecast accuracy represents an important source of potential economic inequality. This paper examines the global distribution of the accuracy of weather forecasts and critically assesses its implications for global inequalities. The paper first describes the current situation and past trends in the accuracy of forecasts, then examines distributional implications, and finally relates differences in accuracy to climatic, geographical, economic, and institutional factors. To do this, the paper combines historical weather analyses and forecasts from several meteorological agencies with climate reanalysis, station measurements, and high-resolution economic data. To attribute differences in forecast accuracy to differences in measurement infrastructure, the paper makes use of quasi-random historical appearance and disappearance of weather stations using modern econometric methods for causal inference. Prior estimates of the value of forecasts are then used to translate differences in forecast accuracy into economic values. Finally, projections of future climate change are used to inform the discussion about changes in the demand for and benefits of forecasts. The paper builds on previous work on the global distribution of forecast accuracy (e.g. Bauer et al. 2015) by focusing on impact-relevant meteorological variables (2m temperature, precipitation), by conducting the analysis at a more granular spatial resolution (subnational in this paper versus hemisphere in prior work), by combining the analysis of weather forecasts with socioeconomic data, by comprehensively examining both differences in the cross-section and over time, and by using methods of causal inference to attribute changes in forecast accuracy to its determinants. Preliminary results show large improvements in forecast accuracy since 1985 everywhere but especially in relatively poor low-latitude countries in the Global South. These trends can be attributed primarily to the use of satellite data which has led to a global convergence in forecast accuracy. In all years, forecast errors tend to be largest in (relatively rich) mid- and high-latitude countries due to large day-to-day fluctuations. At the same time, measures of forecast skill based on correlations tend to feature lower skill in low-latitude countries and hence worse forecasts in relatively poor world regions. This is particularly problematic because projections of climate models suggest increases in temperature variability especially in those regions of the world. By the time of the EGU conference, results on the other parts of the analysis will also be ready for presentation, including a statistical attribution of differences in weather forecast accuracy to differences in the ground-based monitoring infrastructure.</p>