Abstract
Ecosystem Service (ES) mapping has become a key tool in scientific assessments of human-nature interactions and is being increasingly used in environmental planning and policy-making. However, the associated epistemic uncertainty underlying these maps often is not systematically considered. This paper proposes a basic procedure to present areas with lower statistical reliability in a map of an ES indicator, the vegetation carbon stock, when extrapolating field data to larger case study regions. To illustrate our approach, we use regression analyses to model the spatial distribution of vegetation carbon stock in the Brazilian Amazon forest in the State of Pará. In our analysis, we used field data measurements for the carbon stock in three study sites as the response variable and various land characteristics derived from remote sensing as explanatory variables for the ES indicator. We performed regression methods to map the carbon stocks and calculated three indicators of reliability: RMSE-Root-mean-square-error, R2-coefficient of determination - from an out-of-sample validation and prediction intervals. We obtained a map of carbon stocks and made explicit its associated uncertainty using a general indicator of reliability and a map presenting the areas where our prediction is the most uncertain. Finally, we highlighted the role of environmental factors on the range of uncertainty. The results have two implications. (1) Mapping prediction interval indicates areas where the map's reliability is the highest. This information increases the usefulness of ES maps in environmental planning and governance. (2) In the case of the studied indicator, the reliability of our prediction is very dependent on land cover type, on the site location and its biophysical, socioeconomic and political characteristics. A better understanding of the relationship between carbon stock and land-use classes would increase the reliability of the maps. Results of our analysis help to direct future research and fieldwork and to prevent decision-making based on unreliable maps.