Accounting for spatiotemporally correlated errors in wind speed for remote surveys of methane emissions

Abstract. Spatiotemporally correlated errors in wind speeds estimated via numerical weather prediction (NWP) models – where estimated wind speeds and associated uncertainties at similar times or at nearby sites are likely to be correlated – are an important, but to date neglected, determinant of overall uncertainties in remote surveys of emissions in the oil and gas and other sectors. In this work, we develop a methodology to model such errors using publicly available, anemometer-measured wind speeds at weather stations within a region of interest. The method is parsed into two: (1) creation of a region-averaged wind speed error model to provide the probability of the true wind speed given an NWP model-estimated wind speed while ignoring autocorrelation, and (2) development of a Gaussian copula model for the region of interest (ROI) to capture spatiotemporal autocorrelation. The combined model for total wind uncertainty, including spatiotemporally autocorrelated errors, is demonstrated using the oil and gas-producing region of northeastern British Columbia, Canada as a case study. We also provide additional combined models for the Canadian provinces of Alberta and Saskatchewan, the U.S. state of North Dakota, and Colombia. Finally, we share a simple python code to interface with these models and to simplify application by others. The combined models show varying correlations in wind speed errors, which are attributable to the variability of terrain in the ROIs and the relative accuracy of different NWP models. Results further reveal how temporal correlations and hence uncertainties in aggregated emissions can be minimized through remote survey design, where waiting at least two days before revisiting a site and phase shifting re-surveys by approximately six hours can avoid both near-field and diurnal patterns in temporal autocorrelation.