Validation of Sentinel-2 fAPAR products using ground observations across three forest ecosystems

Putzenlechner, Birgitta; Castro, Saulo; Kiese, Ralf; Ludwig, Ralf; Marzahn, Philip; Sharp, Iain und Sanchez-Azofeifa, Arturo (2019): Validation of Sentinel-2 fAPAR products using ground observations across three forest ecosystems. In: Remote Sensing of Environment, Bd. 232, UNSP 111310

Abstract

Accurate estimates of Essential Climate Variables (ECV) such as the fraction of Absorbed Photosynthetically Active Radiation (fAPAR) are essential for assessing global carbon balances. ESA's Sentinel-2 (S2) mission with its decametric resolution enables to derive fAPAR information at single forest stand. Validation studies on previously existing satellite-derived fAPAR products have found considerable discrepancies, especially in forest ecosystems that exceed relative discrepancies of 10% (i.e. 0.05 for absolute values) set as target accuracy by the Global Climate Observing System (GCOS). This study presents the validation of S2 fAPAR products using direct radiation measurements of 2017 at three different forests, located in Central Europe (mixed-coniferous forest in temperate mid-latitude), North America (boreal-deciduous forest) and Central America (tropical dry forest, TDF). We measured incoming and transmitted PAR every 10 min synchronously using Wireless Sensor Networks (WSN) and calculated a two-flux fAPAR estimate. We validated the S2 fAPAR product with different instantaneous ground fAPAR estimates (e.g. instantaneous fAPAR at 10:00 and 14:00 local solar time) and daily average fAPAR. We considered uncertainties of ground data, i.e. bias related to the presence of colored autumn leaves and influences of solar zenith angle. Overall, we found high discrepancies between the S2 fAPAR product and ground measurements, indicating that the S2 fAPAR product systematically underestimated (negative values for bias in percent) the ground observations. The highest agreement was observed at the boreal-deciduous forest stand with a bias of -13% (R-2 = 0.67). The Central American and European sites reported deviations of -20% (R-2 = 0.68) and -25% (R-2 = 0.26), respectively. At all sites, we found evidence that particularly the influence of colored leaves during the senescence periods lead to bias of the ground data. Further, the choice of temporal fAPAR estimate, i.e. daily average fAPAR or a certain instantaneous fAPAR estimate, lead to partly different results in the correlation analysis with the S2 fAPAR product. However, considering sources of uncertainties of ground data, we emphasize that only the boreal-deciduous site in Canada fulfilled the accuracy requirements set by the GCOS. In contrast to absolute values, we found strong agreement on phenological changes at all three sites. Specifically, the influence of species composition on seasonal variability of fAPAR across the European mixed coniferous site was well-represented in the S2 fAPAR product. As for the representation of spatial variability, we found highest agreement at the boreal-deciduous forest stand (BIAS = -22%, R-2 = 0.93), whereas spatial variability was least represented at the TDF site (BIAS = 125%, R-2 = 0.97). We conclude that the S2 fAPAR product has strong capabilities for assessing temporal variability of fAPAR, but due to low accuracy of absolute values currently limited options to feed global production efficiency models and assess global carbon balances.

• Dokumententyp: Zeitschriftenartikel
• Fakultät: Geowissenschaften > Department für Geographie
• Themengebiete: 500 Naturwissenschaften und Mathematik > 550 Geowissenschaften, Geologie
• ISSN: 0034-4257
• Sprache: Englisch