Mitigating geolocation errors in nighttime light satellite data and global CO2 emission gridded data

Abstract

Accurate geospatial modeling of greenhouse gas (GHG) emissions is an essential part of the future of global GHG monitoring systems. Our previous work found a systematic displacement in the high-resolution carbon dioxide (CO2) emission raster data of the Open-source Data Inventory for Anthropogenic CO2 (ODIAC) emission product. It turns out this displacement is due to geolocation bias in the Defense Meteorological Satellite Program (DMSP) nighttime lights (NTL) data products, which are used as a spatial emission proxy for estimating non-point source emissions distributions in ODIAC. Mitigating such geolocation error (~1.7 km), which is on the same order of the size of the carbon observing satellites field of view, is especially critical for the spatial analysis of emissions from cities. In this paper, there is proposed a method to mitigate the geolocation bias in DMSP NTL data that can be applied to DMSP NTL-based geospatial products, such as ODIAC. To identify and characterize the geolocation bias, we used the OpenStreetMap repository to define city boundaries for a large number of global cities. Assumption is that the total emissions within the city boundaries are at the maximum if there is no displacement (geolocation bias) in NTL data. Therefore, it is necessary to find an optimal vector (distance and angle) that maximizes the ODIAC total emissions within cities by shifting the emission fields. In the process of preparing annual composites of the nighttime stable lights data, some pixels of the DMSP data corresponding to water bodies were zeroed, which due to the geolocation bias unreasonably distorted the ODIAC emission fields. Hence, an original approach for restoring data in such pixels is considered using elimination of the factor that distorted the ODIAC emission fields. It is also proposed a bias correction method for shifted high-resolution emission fields in ODIAC. The bias correction was applied to multiple cities from the different continents. It is shown that the bias correction to the emission data (elimination of geolocation error in non-point emission source fields) increases the total CO2 emissions within city boundaries by 4.76% on average, due to reduced emissions from non-urban areas to which these emissions were likely to be erroneously attributed.