Boreal Forest

The seasonality of photosynthesis of boreal forests is an essential driver of the terrestrial carbon, water and energy cycles. However, current carbon cycle model results only poorly represent interannual variability and predict very different magnitudes and timings of carbon fluxes between the atmosphere and the land surface (e.g. Jung et al. 2011, Richardson et al. 2012). Reflectance-based satellite measurements, which give an indication of the amount of green biomass on the Earth's surface, have so far been used as input to global carbon cycle simulations, but they have limitations as they are not directly linked to instantaneous photosynthesis. As an alternative, space-borne retrievals of sun-induced chlorophyll fluorescence (SIF) from GOSAT, GOME-2, SCIAMACHY or OCO-2 boast the potential to provide a direct indication of the seasonality of boreal forest photosynthetic activity and thus to improve carbon model performances. SIF is a small electromagnetic signal that is re-emitted from the photosystems in the chloroplasts, which results in a direct relationship to photosynthetic efficiency.

In this contribution we examine the seasonality of the boreal forests with three different vegetation parameters, namely greenness, SIF and model simulations of gross primary production (gross carbon flux into the plants by photosynthesis, GPP). We use the enhanced vegetation index (EVI) to represent green biomass. EVI is calculated from NBAR MODIS reflectance measurements (0.05deg, 16 days temporal resolution) for the time from January 2007-May 2013. SIF data originate from GOME-2 measurements on board the MetOp-A satellite in a spatial resolution of 0.5deg for the time from 2007-2011 (Joiner et al. (2013), Köhler et al. (2014)). As a third data source, data-driven GPP model results are used for the time from 2006-2012 with 0.5deg spatial resolution. The method to quantify phenology developed by Gonsamo et al. (2013) is applied to infer the main phenological phases (greenup/onset of activity, maturity, senescence and end of season) from all 3 data streams. Maps of the transition dates (most of all the start of season) of EVI, SIF and GPP are derived and compared. Further, local comparisons of the annual cycle over several large scale regions and forest types are done.

Among other results, we find that in the boreal evergreen needleleaf forests both model GPP and SIF indicate much earlier onset of activity than EVI. This confirms – on a larger scale - findings from tower observations. Moreover, the end of activity occurs later in the case of SIF and GPP, which results in an overall longer growing season. Summer peak values of chlorophyll fluorescence, model GPP and greenness are reached approximately at the time of the annual temperature maximum one month after the illumination peak. In deciduous forests the length of the growing season indicated by the three proxies is very similar, however, SIF and GPP show large intraseasonal variability that cannot be identified using EVI. Also a slight decline in all three proxies can be observed from the end of June until August indicating that greenness and photosynthesis are already reduced to a small extent before autumn senescence starts and before the annual temperature maximum is reached. This might be due to higher sensitivity to illumination than to temperature at that time of year. These and other results show that satellite measurements of chlorophyll fluorescence reliably indicate plant activity and that they might be useful for benchmarking dynamic global vegetation and carbon cycle models.

Clear-cutting is the most drastic and wide-spread change that affects the hydrological and carbon-balance properties of forested land in the Boreal forest zone.

A time-series of 27 Sentinel-1 images was used to study the potential for mapping clear-cut areas in Boreal forest zone.  The time series covered a full year (2014-10-04 ... 2015-09-29) in 200-km-by-200-km study site in Finland.  The Sentinel-1 images were acquired in Interferometric Wide-swath (IW), dual-polarised mode (VV+VH).  All scenes were acquired in the same orbit configuration.  The Sentinel-1 scenes were down-loaded from the ESA scihub system as amplitude images (GRDH product).  The Sentinel-1 scenes were ortho-rectified with in-house software using a digital elevation model (DEM) produced by the Land Survey of Finland.  The Sentinel-1 amplitude data were normalized by the projected pixel area, which means that the corrected amplitude values are proportional to backscattering coefficient gamma-nought.

The temporal behaviour of C-band backscatter was studied for areas representing 1) areas clear-cut during the acquisition of the Sentinel-1 time-series, 2) areas remaining forest during the acquisition of the Sentinel-1 time-series, and 3) areas that had been clear-cut before the acquisition of the Sentinel-1 time-series.

The following observations were made:

The separation between clear-cut areas and forest was generally low;

Under certain acquisition conditions, clear-cut areas were well separable from forest;

The good scenes were acquired: 1) in winter during snow cover and temperature above zero degrees Celsius, and 2) in late summer towards the end of a warm and dry period;

The separation between clear-cut and forest was higher in the winter/snow/above-zero scenes than in the dry summer scenes.

An algorithm is proposed for reliable mapping of clear-cut areas with a time-series of Sentinel-1 data:

collect all Sentinel-1 scenes acquired in the chosen orbit configuration in dual-polarized interferometric Wide-swath mode

for each winter scene, compute the contrast between agricultural and forested areas,

if the contrast in a winter scene is higher than a predefined threshold add the new scene to the feature set of further analysis,

select the best summer scene based on meteorological data: the scene with the longest period without rain before the image acquisition,

make a clustering of the selected images and label the clusters based on reference data from the ground or from interpretation of cloud-free optical images acquired in summer time.

In point 2 above, the contrast between agricultural areas and forest is used, not between clear-cut and forest.  This is because agricultural areas and clear-cut areas have similar backscatter level in good winter images and because agricultural areas are easier to identify than clear-cut areas when moving to new monitoring area.

This study was part of EU/FP7 project NorthState (grant 606962).

Spaceborne radar has found limited use in wall-to-wall mapping of forest variables so far, despite the availability of global observations from several missions. While many studies have documented the sensitivity of, in particular, long wavelength radar backscatter observations to forest variables such as growing stock volume (GSV) or aboveground biomass (AGB), the large-scale application faces a number of specific challenges, such as the pronounced sensitivity of the measurements to changing environmental imaging conditions (e.g., freeze/thaw, moisture variations) and the limited availability of in situ data for a locally adaptive calibration of models, relating the signal to the forest biophysical variable of interest.

A first global radar based map of forest GSV at ~1 km resolution is presented (see LP15 paper by Santoro et al.), which exploits the global archive of hypertemporal medium- to low-resolution ENVISAT ASAR observations and the BIOMASAR retrieval approach (Santoro et al., 2015). Key features of the algorithm deployed are the adaptive calibration of models, relating backscatter to GSV, without the need for in situ observations, and the minimization of noise (environmental, measurement) in the estimates by means of a weighted combination of retrieval results from a large number of multi-temporal observations. Validation efforts for forests across the northern hemisphere demonstrated the ability of hyper-temporal C-band to adequately depict the spatial distribution of forest GSV. Yet, limitations of the ASAR product were found to be associated with the low resolution of the ScanSAR data and the limited sensitivity of C-band to GSV in the densest forests (e.g., humid tropical).

In the frame of the ESA DUE GlobBiomass project, we investigate options for complementing and, possibly, improving the existing map using globally available high-resolution (~30m) mosaics of ALOS PALSAR dual-polarization backscatter (Shimada et al., 2014) and Landsat reflectances (Hansen et al., 2013). The goals of this study are i) to develop algorithms that allow for exploiting the improved sensitivity of longer wavelength L-band radar to GSV as well as the higher resolution of PALSAR and Landsat, and ii) document uncertainties associated with the estimation of GSV at global scale with datasets that are currently available globally. The global mapping efforts are supported by a number of regional GlobBiomass projects, which provide the means for validating against regionally tuned/optimized maps and optimization of the modeling for different forest ecosystems.

For the global mapping of GSV with PALSAR, we adopt the BIOMASAR algorithm. The performance of this algorithm when used for L-band was so far verified in regional studies for temperate forests (Cartus et al., 2012) and boreal forests (Santoro et al., 2014). Limited experience exists concerning the feasibility of the approach, which bases on a semi-empirical Water-Cloud-type of model, in the tropics and savanna forest ecosystems. The investigations therefore aim at confirming the validity of the modeling assumptions across the major forest biomes as well as the feasibility of the approach for calibrating the model, which is achieved with the aid of global Landsat-based canopy density maps and ICESAT GLAS derived forest heights. To exploit the spatial detail provided by global mosaics of Landsat reflectances, empirical regression tree algorithms such as randomForest or Cubist are tested for downscaling the 1km GSV estimates from ENVISAT ASAR to 30m. Initial tests of this approach, which has previously been used to downscale MODIS canopy density maps to the 30m pixel size of Landsat (Sexton et al., 2013), indicate the possibility to improve the spatial characterization of the GSV distribution in the global GSV map, in particular in heterogeneous forest areas where the low resolution provided by ASAR is insufficient. The current status and results of research activities within this GlobBiomass study will be presented at the symposium.

References:

Cartus, O., Santoro, M., & Kellndorfer, J. (2012). Mapping forest aboveground biomass in the Northeastern United States with ALOS PALSAR dual-polarization L-band. Remote Sensing of Environment, 124, 466–478.

Hansen, M. C., Potapov, P., Moore, R., et al. (2013). High-resolution global maps of 21st-century forest cover change. Science, 342(6160), 850–3.

Santoro, M., Wegmuller, U., Fransson, J. E. S., Schmullius, C., "Regional mapping of forest growing stock volume with multitemporal ALOS PALSAR backscatter," Proceedings of IGARSS'14, Quebec City, 13-18 July, pp. 2313-2316, 2014.

Santoro, M., Beaudoin, A., Beer, C., Cartus, O., et al. (2015). Forest growing stock volume of the northern hemisphere: Spatially explicit estimates for 2010 derived from Envisat ASAR. Remote Sensing of Environment, 168, 316–334.

Sexton, J., Song, X. P., Feng, M., et al. (2013). Global, 30-m resolution continuous fields of tree cover: Landsat-based rescaling of MODIS Vegetation Continuous Fields with lidar-based estimates of error. International Journal of Digital Earth, 6, 427-448.

Shimada, M., Itoh, T., Motooka, T., Watanabe, M., Shiraishi, T., Thapa, R., & Lucas, R. (2014). New global forest/non-forest maps from ALOS PALSAR data (2007–2010). Remote Sensing of Environment, 155, 13-31.

The Boreal ecosystem, corresponding to a forest dominated zone in northern Eurasia and North America, covers ~17% of the terrestrial land surface (FAO, 1999), corresponds to ~38% of global forest area and contains ~40% of global soil carbon (Committee I., 2010).  The ecosystem plays a major role within the Earth’s climate and carbon cycle (Kasischke and Stocks, 2000) and also support land uses such as forestry and, increasingly, oil and gas.   Environmental monitoring of boreal forests is essential to ensure responsible resource development.   Land cover and leaf area index (LAI) are two fundamental quantities used both as environmental indicators and as input to ecosystem and atmosphere models.  In the past, medium resolution satellite imagery has had limited ability  to quantify LAI in Boreal ecosystem due to sensitivity to land cover, saturation at high LAI and  uncertainty with atmospheric correction (Fernandes et al. 2004). 

In this study we make use of both Landsat OLI and Sentinel2 MSI imagery acquired within a two week period, in combination with extensive field measurements of LAI, over a highly disturbed region of the Canadian boreal ecosystem to address two research questions:  1.  What is the consistency between LAI products derived from both sensors separately?  2.  Can the sensors be used synergistically by combining them at a radiometric level prior to land cover and Lai estimation? 

Climate change has altered vegetation productivity and structure, carbon sequestration and many other processes in the higher northern latitudes. The taiga-tundra ecotone (TTE) which is the circumpolar transition zone from 50° to 75° north latitude is expected to change in form and distribution from global warming. Some evidence of this has already been documented in small-scale studies but large amounts of uncertainty remains in current global estimates of circumpolar forest extent. Establishing the TTE extent is challenging because of variations in vegetation structure, site-level interactions between microclimate, topography, winter snow depth, wind, edaphic conditions and other factors. These interactions have produced TTE forest patterns that include sporadic forest cover patches to growth-stunted trees resembling shrubs. These patterns are evident at local scales, but are often not apparent in moderate resolution imagery from most earth observing satellites. Such scale problems have contributed to large TTE geographic uncertainties. Recent advances using the global archive of Landsat data have provided vegetation continuous fields (VCF) at 30-m resolution, which may improve TTE delineation. Furthermore, U.S. federal access to sub-meter DigitalGlobe data provides the means to improve estimates of the extent of the TTE with observations of individual trees and structure from stereoscopic observations. We are currently processing these data with super-computing resources to extract detailed forest structure information over vast and remote areas of the TTE. These derived products will be used to optimize the Landsat VCF delineation of the ecotone. The resulting products could become a baseline to identify sites of potential TTE structure change and be used in ecosystem models. An overview and early results of our current work will be presented.