Vegetation Parameters 3

Vegetated coastal ecosystems such as a mangroves and saltmarshes have important biogeochemical roles, as well as offering system services (e.g. coastal armoring and storm surge mitigation). These features are perturbed by impacts associated with global climate change, including sea-level rise; increased frequency and magnitude of storms; and ocean acidification. Monitoring of terrestrial ecosystems utilizes the ever-improving spatial and temporal resolutions of satellite data, but these remote sensing resources are challenged by coastal ecosystems. High magnitude, high frequency sources of variation such as the tidal and wind states can mask underlying change. Saltmarshes in particular can have very strong phenological variations, and their ecological and hydrological condition can be drastically changed by episodic events in a matter of hours. In order to facilitate the use of long-term, high-resolution satellite data to monitor coastal ecosystem condition, these sources of variation require characterization, quantification and ultimately accurate monitoring, modeling, and prediction. Herein we present the methods and results from a pilot study in a saltmarsh linking fine-scale land cover, including vegetation by species and structure; tidal state; and phenological stage to the multi-spectral satellite signal. Throughout the 2015 growing season lidar and optical data were collected with terrestrial instruments covering seven co-located Landst 7 & 8 pixels in the Plum Island Long Term Ecological Research site (Massachusetts, USA). These data were collected within 30 minutes of Landsat overpass in order to reflect very similar conditions, with those areas containing tidally-active hydrological features prioritized for collection closest in time to satellite overpass. Achieving substantial spatial coverage in such a narrow window of temporal stability has been facilitated by recent developments and optimization of terrestrial instruments. In particular, the Compact Biomass Lidar (CBL), a 905nm, time-of-flight scanner provided the speed and portability (33 seconds per scan, 3.4kg weight) necessary to acquire lidar scans of the entire 6300m² area at the time of each satellite overpass. Optical and lidar data were combined to produce 10cm scale classification maps of land cover types and structure for each pixel. Coupled with hyperspectral data acquired for each saltmarsh cover type with an ASD spectrometer, these proportional components are analyzed for their contribution to the Landsat signal at the channel level. The compositional, tidal and phenological components of the Landsat 7 and 8 signals are quantified across the growing season. Additionally, the detailed 3D maps of the saltmarsh hydrological features were used to develop an interpolative model of tidal state, which can be used predictively, replacing the need for repeated overpass-coincident data collection. Goddard’s Lidar Hyperspectral and Thermal instrument package (G-LiHT) overflew the study sites in 2014 and in 2015, when coincident terrestrial data were collected. In addition to the spectral and lidar data at 1m resolution, analysis of the potential contribution of the thermal information from G-LiHT was explored.

These results are presented as a pathfinder for the application of similar methods to interpret coastal ecosystem conditions from Sentinel-2. Sentinel-2 offers additional channels and a superior spatial and temporal resolution to Landsat, therefore making future long-term monitoring even more viable and informative for  multi-source studies  of the high-frequency, high magnitude sources of variation inherent to coastal ecosystems.

The Carbon Cycle Data Assimilation System (CCDAS, Rayner 2005, Knorr 2008)  is used to increase our knowledge about the global carbon cycle by reducing the uncertainties in regional sources/sinks in the terrestrial biosphere. This is achieved through the simultaneous assimilation  of complementing satellite data streams, including column integrated CO 2 (XCO2), into the land biosphere model BETHY (Knorr 2000) and a subsequent propagation of posterior uncertainties to terrestrial fluxes. BETHY simulates carbon assimilation and plant and soil respiration, embedded within a full energy and water balance. Atmospheric CO2 concentrations are derived from CO2 surface fluxes
 using the precomputed Jacobian of the Transport Model TM3. The data assimilation system uses the Bayesian approach by defining a cost function which quantifies the misfit between observations and their model equivalence, assuming Gaussian error distribution.
The cost function depends on a number of process parameters of BETHY and the mean global CO2 concentration. The best fit of the model to the observations is at the minimum of the cost function. It is found by a BFGS algorithm using the gradient information computed efficiently by the adjoint model. The adjoint code is generated by FastOpt's Automatic Differentiation tool Transformation of Algorithm in Fortran (TAF).
By this procedure dynamically consistent Net and Gross Surface CO2 fluxes are found. Data streams integrated in CCDAS are:
 - Fraction of Absorbed Photosynthetic Active Radiation (fapar) and Leaf Area Index (effective LAI) from Two-stream Inversion Package (TIP-GlobAlbedo)
- soil moisture from SMOS
- surface CO2 concentration at FluxNet sites
- XCO2 satellite product from SCIAMACHY 13
All model-data misfits are weighted by the inverse error estimates. The TIP GlobAlbedo fapar and LAI data misfits are weighted by their inverse error correlation matrix (preliminary results of QA4ECV). Prior estimates of the process parameters are used and deviations are weighted with assumed uncertainties. Posterior error estimates of the parameters are derived from the Hessian of the cost function at the minimum.
This error is propagated by the adjoint and tangent of the prediction operator to error estimates of predicted quantities (Scholze 2007). Again these derivative codes (Hessian, adjoint, tangent) are generated by TAF. By the assimilation of XCO2 data the posterior error estimates of net terrestrial carbon fluxes in 2010 are noticable reduced compared to assimilation without the XCO2 data.

Global climate change research suggest that extreme weather events are expected in higher frequency, intensity, and duration at global scales and these events will have drastic changes on both the agricultural and forest landscapes. Higher temporal frequency moderate resolution data are required for monitoring agriculture production and forest monitoring. To meet this need, multi-source (Landsat TM/ETM and Sentinel-2) land imaging products have been proposed and are being currently generated by NASA Earth Exchange (NEX, http://nex.nasa.gov).NEX is a state-of-the-art collaborative supercomputing platform that provides ready-to-use data, models and codes for performing scientific analysis on large Earth science data sets. The computational efficiency on the NEX platform has been showcased in various large scientific projects dealing with Landsat data and very high resolution satellite data processing architectures. NEX also provides knowledge capture and provenance tools that are really helpful in managing scientific worlflows in a production environment.

In collaboration with the research scientists at NASA's Goddard Space Flight Center and NASA's land cover/land use change program, we are integrating the 6S radiative transfer (RT) model, BRDF correction modules and spectral response function calibration modules to generate harmonized daily Surface Reflectance products from Landsat 8 OLI and Sentinel-2 MSI. The near-daily Sentinel-2 / Landsat 8 surface reflectance product will be provided once the ESA release S2 data for public access. We have also deployed ESA’s sen2corr tool (atmospheric correction tool for Sentinel-2) on NASA NEX, which can be used for comparison with the 6S output. Cross-validation of the surface reflectance products will be conducted and higher level products will be generated (e.g. LAI and FPAR) from domain experts who will access to the NEX system. The harmonized data products (L8 and S2) will be used to monitor agricultural and forest productivity across regional to continental scales. 

The High Arctic Zone is characterized by a short and intense growing season, where even small changes in the timing of the growing season highly influences the plant production and the population dynamics of most animals, birds, and insects. Changes in the growing season are also important for the feedback loop to the climate. The study area is the High Arctic archipelago of Svalbard, located at about 76°30ʹ–80°50ʹN, and where the warmest areas have about 7 °C in July mean temperature. The main aim of this study is to map the growing season at three different scales, where each scale is closely connected and complementary in a way that observations on a finer scale provides the base for interpreting the scale above.

On a field-scale we use near-surface hourly time-lapse RGB phenology cameras (‘phenocams’) for the years 2013-2015. These cameras are placed in a range of different vegetation types, and from these RGB images we calculate different indices. From the indices we then automatically calculate phenophases as onset, peak and end of the growing season, on both species and plant-community levels.

On a local to a regional scale, we use time-series of Landsat 8 images for the years 2014 and 2015. Due to the location close to the North Pole and the polar orbit of Landsat 8, data are obtained several times a week. Moreover, all the paths 22-29 and 211-217, within row 3-4 and 240-241, cover the central part of Svalbard. Our datasets for the 2014 and 2015 seasons (May-September) are composed of 85 and 74 images, respectively. For the former season, we selected the 17 best images, with a 7-day average sampling interval. For the latter, we selected 23 images with a 8-day average sampling interval.

For the entire archipelago Svalbard, for the 2000-2015 period, we use MODIS data, the 8-days composites products MOD09A1 and MOD09Q1.

Due to scattered vegetation cover, ice and snow, short season, and very frequent cloud cover, cloud detection on Svalbard is a challenging task. For both the Landsat 8 data and MODIS data the first step is to remove the cloud cover, where we use a combination of three cloud removing methods (quality information provided with the data, own algorithms, and manual removal). This combination of methods worked well, but is time-consuming as it requires manual interpretation of cloud cover. Then we interpolate the cloudy parts and smooth the time-series data. The onset of the growing season is mapped with a NDVI threshold method, which shows high correlation with photos from the time-lapse cameras. For mapping the end of the growing season a combination of different methods have to be used, depending on the land cover type. However, this work is still not completed.

The results demonstrate a multi-scale mapping of the seasonal-dynamics from species level, to vegetation types, and to ecosystems. Due to the high availability of Landsat 8 data from Svalbard, the study area creates a unique opportunity for a high spatial and temporal resolution mapping of the seasonal dynamics, and is a good preparation for planned Sentinel-2 based mapping. The work is in part funded by the ESA Prodex project ‘Sentinel-2 for high north vegetation phenology’.

River deltas represent rapidly developing geomorphological features among Earth landforms characterized by continuous dynamic changes. Shape of delta is controlled by sediment deposition, vegetation cover and erosion processes, mostly biophysical forcing driven. River sediment supply represents the major gain in the delta and vegetation contributes to sediment cohesion reducing its erodibility. On the contrary, removal of coastal sediments by wave action together with natural and human induced subsidence determine erosion processes. Thematic maps generated from different RS datasets describe Essential Variables (EV) that were used to characterize the evolution of the Po deltaic system (Italy).

Decadal time series analysis of Total Suspended Matter (TSM) generated from ENVISAT MERIS data was used to characterize variability of river plumes, with a special focus to flood river discharge conditions. Evolution of sandy spits, including sand and vegetation dynamic, and coastal erosion trends were quantitatively estimated from Landsat time series for the period 1991-2015. Cover abundance maps derived from Linear Spectral Mixing Analysis (LSMA) were further analyzed using Empirical Orthogonal Function (EOF) method, in order to find spatial and temporal patterns of cover changes. Rates of subsidence were estimated from vertical land velocity applying Small BAseline Subset (SBAS) method to ENVISAT ASAR SLC data for the period 2003-2010. Coastal wave field for the period 2008-2013 were downscaled using Radial Basis Functions (RBF) interpolation forced using both atmospheric modeled winds and SAR estimated wind from ENVISAT ASAR WSM data.

For the considered time period results prove that, despite some areas in proximity of the main river mouths show progradation due to accumulation of recently delivered river sediments, Po river delta is dominated by wave action. Erosion processes are more active in the southern coast: Significant Wave Height exceed by a factor of approximately 1.6 when compared to northern coast suggesting an asymmetric response to wave action and sediment fluxes. TSM concentrations from ten-years time series reveal that the removal of fine sediment from buoyant plumes occurs within 10 km from Po delta, while coarser material is settling in the proximity of river mouths and not deposited farther than 3 km. Subsidence rates have a spatial gradient moving seaward, reporting values ranging between -1 and -7 mm y^-1.

This research study provides innovative maps and models that describe macrosystems river deltas that could be effectively used in the context of strategies implementation for the integrated coastal zone management.