From sample to pixel: multi‐scale remote sensing data for upscaling aboveground carbon data in heterogeneous landscapes

• DOI: 10.1002/ecs2.2298
• Publication Date: 2018-08-22T19:30:06Z
• Authors: Pedro J. Leitão; Marcel Schwieder; Florian Pötzschner; José R. R. Pinto; Ana M. C. Teixeira; Fernando Pedroni; Maryland Sanchez; Christian Rogass; Sebastian van der...; Mercedes M. C. Bu...; Patrick Hostert
• Keywords: Moderate-resolution imaging spectroradiometer

Abstract In times of rapid global change, ecosystem monitoring is utmost importance. Combined field and remote sensing data enable large‐scale assessments, while maintaining local relevance accuracy. heterogeneous landscapes, however, the integration field‐collected with image pixels not a trivial matter. Indeed, much uncertainty in models that use to map larger areas lies on integration. this study, we propose fine spatial resolution (5 × 5 m 2 ) as auxiliary for upscaling field‐sampled aboveground carbon target (meso‐scale, i.e., 30 pixels. process, assess effects disaggregation extrapolation, without data. We test three study sites landscapes Brazilian savanna. thus compare two methods data—surface method, which uses weighting layer, regression applies model—with one method data—cartographic method. To evaluate our results, compared observed vs. estimated values (for known samples) at pixel level. Additionally, fitted random forest model assigned estimates satellite imagery assessed influence fraction extrapolated sampled performance. that, improves field‐based coarser also show surface more suitable disaggregation, approach preferable extrapolating non‐sampled fractions. datasets, included higher proportion values, generally delivered better than smaller datasets are assumed reliably reflect reality. Our enables link data, turn detailed mapping over large through optimized multi‐scale