Interpreting the uncertainty of model-based and design-based estimation in downscaling estimates from NFI data: a case-study in Extremadura (Spain)

Abstract

Remotely sensed data are increasingly used together with National Forest Inventory (NFI) data to improve the spatial precision of forest variable estimates. In this study, we combined data from the 4th Spanish National Forest Inventory (SNFI-4) and from the 2nd nationwide Airborne Laser Scanning (ALS) survey to develop predictive forest inventory variables (total over bark volume (V), basal area (G), and annual increase in total volume (IAVC)) and aboveground biomass (AGB) models for the eight major forest strata in the region of Extremadura that are included in the Spanish Forest Map (SFM). We generated maps at 25 m resolution by applying an area‐based approach (ABA) and 758 sample plots measured with good positional accuracy within the SNFI-4 in Extremadura (Spain). Inventory performance is mainly influenced by spatial scale and vegetation structure. Therefore, in this study, we conducted a comparative analysis of statistical inference methods that can characterize forest inventory variables and AGB uncertainty across multiple spatial scales and types of vegetation structure. Predictions at pixel level were used to produce county, provincial, and regional model-based estimates, which were then compared with design-based estimates at different scales for different types of forest. We developed and tested both methods for forested area (cover, 19,744.15 km2), one province (9126.78 km2), and two counties (1594.42 km2 and 2076.76 km2, respectively) in Extremadura. The resulting relative standard error (SE) for regional level forest type-specific model-based estimates of V, G, IAVC, and AGB ranged from 3.34%–14.46%, 3.22%–12.50%, 4.46%–16.67%, and 3.63%–12.58%, respectively. The performance of the model-based approach, as assessed by the relative SE, was similar to that of the design-based approach at regional and provincial levels. However, the precision of SNFI model-based estimates was higher than that of estimates based on only the plot observations in small areas (e.g. at county level). The standard errors (SE) for model-based inferences were stable across the different scales, while SNFI design-based errors were higher due to the small sample sizes available for small areas. The findings indicate that SNFI-model based maps could be used directly to estimate forest inventory variables and AGB in the major forest strata included in the Spanish Forest Map, leading to potentially large economic savings.

Keywords

• lidar
• mapping
• model-based inference
• design-based inference
• uncertainty
• spanish national forest inventory (snfi)