Combine satellitebased geoinformation ground measurements and statistical methods
Combine satellite-based geoinformation, ground measurements and statistical methods to assess green-up trends in earth observational studies Doktor, D. Helmholtz for Environmental Research – UFZ Two. Centre observational sets: en-up dates d. GU derived from 1 [km] NDVI vs ground observatio Contact: daniel. doktor@ufz. de of budburst d. BB How can the two sets be compared ? 1
Why are we interested in trends in greenup dates ? 2
Plant Phenology, what is it ? • Phenology is the study of the timing of recurring life cycle events • For plants such events are budburst, leaf unfolding, blossoming, fruit ripening, leaf colouring etc. History: Time series since ~815 AD (cherry blossom in Japan) Phenological observation network started in 1882 Modern phenological networks include the International Phenological Gardens and National Networks 3
Why is phenology important ? • Track biotic response to climate change • Strongly influences atmospheric dynamics in the lower bou layer • Strong biophysical control on climate model simulations • Estimating biological productivity • Understanding land-atmosphere interactions • Modelling vegetative inputs into biogeochemical cycles 4
Sets of geoinformation • Satellite data: Daily observations of the Normalized Difference Vegetation Index. Influenced by clouds, atmospheric haze, viewing angle, sun position etc. • German Weather Service (DWD) phenological database, ground observations: Budburst dates of wildgrowing species • CORINE (Coordinated Information on the Environment) land-use cover map 5
Satellite products • Range of satellite products => Landsat, AVHRR, MODIS, Spot with all different spectral, temporal and spatial resolutions • Terra-MODIS data downloadable at no costs => Surface reflectance, NDVI, EVI, surface temperature etc. • Longer time series (since 1981) are provided by NOAA-AVHRR, but lower resolution and data quality 6
Satellite derived vegetation information NDVI composites over the Hainich from 1991 • Daily observations or composites of 1*1 k resolution • Based on reflection differences in the nea red wave lengths • Values generally between 0 -1 Range of methods to compute green-up dates => half-way between min and max, steepest slope, inflexion point of sigmoidal curve 7
Extract green-up date 8
Satellite and ground observational data NDVI, 1989 -2004, NOAA/AVHRR, 1 km Atmospheric filtering and corrections (Koslowsky et al. 2001, Koslowsky et al. 2003) Dynamic filtering BISE (Viovy et al. 1992) Ground observations, Observational network of German Weather Service (DWD) For Beech and Oak (N = 35195) (Schaber & Badeck 2002, 2005) 9
CORINE land-use data set 10
Scaling ground observations to the scale observed by satellites • In contrast to remote sensing ground observations do generally not provide integrated data => individual plant vs canopy • Low resolution, cloud contamination, atmospheric interferences, slight varia the geolocation • Green-up of understorey, influence of surrounding pixels (sensor‘s spatial reponsitivity) • Missing habitat maps for potential decomposition of the satellite signal 11
Geostatistical interpolation • External Drift Kriging (EDK) using a Digital Elevation Model of Germany (1*1 km) as external variable • Physical relationship between external variable and variable to estimated => Phenological pace dependent on tem-perature 12
Comparison of the two sets – a statistical approach • Geographically co-incident data of ground and satellite observations are not directly compared. Statistical approach comparing the parameters of the probability density function of the respective set (Doktor et al. 2005; Badeck et al. 2006) • Generation of subsets with different proportions of non-forest landuse according to CORINE (resolution of 0. 25 x 0. 25 km). Those subsets of 1 x 1 km forest pixels are classified with up to 0 through 9 non-forest 0. 25 x 0. 25 subpixels • Additionally, a subset is produced of which the mid-point of the NDVI pixels is inside a CORINE deciduous forest subpixel. This subpixel has to be surrounded by at least three rows of deciduous forest on any side 13
Expectation: slightly later green-up date than observed budburst Result: generally substantially earlier green-up than budburst Hypotheses: strong impact of non-tree vegetation elements on the signal, even in mainly forested landscapes 14
15 At 1 km resolution, satellite derived green-up dates for pixels classified as deciduous forest are very sensitive to admixture of other vegetation elements
Heterogeneity is a major cause for discrepancies Differences between the pdf's parameters of d. GU and d. BB are calculated. The relationship of these differences and the number of non-forest sub-pixels is modelled using a linear regression µ Q 1 Q 3 SD 1989 0. 98 0. 99 0. 85 1990 0. 99 0. 96 0. 75 1991 0. 97 0. 84 0. 97 0. 28 1992 0. 59 0. 86 0. 00 0. 74 1993 0. 99 0. 98 0. 95 0. 64 1994 0. 11 0. 46 0. 00 0. 81 1995 0. 98 0. 99 0. 72 1996 0. 71 0. 87 0. 45 0. 89 1997 0. 73 0. 75 0. 73 1998 0. 96 0. 94 0. 99 0. 62 1999 0. 97 0. 90 0. 99 0. 33 2000 0. 99 0. 98 0. 82 0. 85 2001 0. 98 0. 91 0. 99 0. 67 2002 0. 92 0. 89 0. 96 0. 60 2003 0. 76 0. 71 0. 55 2004 0. 96 0. 97 0. 93 0. 05 16
Green-up dates on a continental scale 17
Observed green-up trends • Studies based on satellite data with different spatial and spectral resolution • Use of different methods to determine green-up dates Myneni et. al (1997): plant growthscale in highwith • Mostly done increased on continental latitudes homegeneousnorthern land-use Schwartz et. al (2006) : earlier SOS over the Northern Hemisphere Stöckli et. al (2004): Trend for earlier green-up Delbart et. al (2004, 2005): Earlier green-up over Siberia 18
Trend of green-up dates 19
Conclusions I The Statistical approach most promising because remote sensing provides a statistical distribution of a random variable, not an exact representation of the state of the land surface or atmosphere at a particular pixel. The same holds true for ground observations as they sample from biological variability and landscapes with heterogeneous microclimates At 1 km resolution and in highly fragmented landscapes, a heterogeneous vegetation cover is a major cause for differences between satellite derived green-up dates and date of budburst of the main vegetation element observed on the ground. Therefore, at this spatial scale, green-up dates derived from satellite data and ground observations record two intrinsically different but related traits. Both can be linked potentially only through the use of high resolution land cover data and thus ground observation can be used for validation purposes with such an analysis of mixed signals 20
Conclusions II § Ground observational networks still provide unique information about plant phenology => go out and observe § Detected green-up trends based on satellite are of course valid but results have to be evaluated carefully 21
Carbon Sequestration dependent on changing temperature patterns and land-use change § Earlier start of the season increases sequestration, 16 % more absorbed PAR (negative feedback) § C-sequestration varies with land-use, i. e. forest sequesters more than crops § IPCC Guidelines for National Greenhouse Gas tories (GNGG) 22
Thank you for your attention ! 23
- Slides: 23