Mitigating the effects of omission errors on area and area change estimates

Abstract

Information on Earth's land surface and change over time has never been easier to obtain, but making informed decisions to manage land well necessitates that this information is accurate and precise. In recent years, due largely to the inevitability of classification errors in remote sensing-based maps and the marked effects of these errors on subsequent area estimates, sample-based area estimates of land cover and land change have increased in importance and use. Area estimation of land cover and change by sampling is often made more efficient by a priori knowledge of the study area to be analyzed (e.g., stratification). Satellite data, obtained free of cost for virtually all of Earth's land surface, provide an excellent source for constructing landscape stratifications in the form of maps. Errors of omission, defined as sample units observed as land change but mapped as a stable class, may introduce considerable uncertainty in parameter estimates obtained from the sample data (e.g., area estimates of land change). The effects of omission errors are exacerbated in situations where the area of intact forest is large relative to the area of forest change, a common situation in countries that seek results-based payments for reductions in deforestation and associated carbon emissions. The presence of omission errors in such situations can preclude the acquisition of statistically valid evidence of a reduction in deforestation, and thus prevent payments. International donors and countries concerned with mitigating the effects of climate change are looking for guidance on how to reduce the effects of omission errors on area estimates of land change. This article presents the underlying reasons for the effects of omission errors on area estimates, case studies highlighting real-world examples of these effects, and proposes potential solutions. Practicable approaches to efficiently splitting large stable strata are presented that may reduce the effects of omission errors and immediately improve the quality of estimates. However, more research is needed before further recommendations can be provided on how to contain, mitigate and potentially eliminate the effects of omissions errors.