Heiko Balzter

Cellular automata models for vegetation dynamics

A short review of cellular automata models in ecology is given. Introduction of a spatial dimension into a discrete-time Markov chain leads to a class of cellular automata called spatio-temporal Markov chains (STMC). The behaviour of the model is determined by its temporal and spatial orders. It has been proven that STMC models have ergodic distributions in certain cases for spatial order 0. Simulations suggest that STMC models of higher spatial order also have ergodic distributions. The model of Dytham (1995) is demonstrated to be an STMC. Modelling population dynamics of three plant species on a lawn, two STMC models of different order are compared. The model validation shows a good agreement for Glechoma hederacea, but large deviations for Lolium perenne and Trifolium repens. The species-dependent performance of the models can be explained by selective grazing. Modifications of the transition matrices are used to examine possible causes of the deviations.

Markov chain models for vegetation dynamics

A theoretical implementation of Markov chain models of vegetation dynamics is presented. An overview of 22 applications of Markov chain models is presented, using data from four sources examining different grassland communities with varying sampling techniques, data types and vegetation parameters. For microdata, individual transitions have been observed, and several statistical tests of model assumptions are performed. The goodness of fit of the model predictions is assessed both for micro- and macrodata using the mean square error, Spearman’s rank correlation coefficient and Wilcoxon’s signed-rank test. It is concluded that the performance of the model varies between data sets, microdata generate a lower mean square error than aggregated macrodata, and time steps of one year are preferable to three months. The rank order of dominant species is found to be the most reliable prediction achievable with the models proposed.

Large-Scale Mapping of Boreal Forest in SIBERIA using ERS Tandem Coherence and JERS Backscatter Data

Siberias boreal forests represent an economically and ecologically precious resource, a significant part of which is not monitored on a regular basis. Synthetic aperture radars (SARs), with their sensitivity to forest biomass, offer mapping capabilities that could provide valuable up-to-date information, for example about fire damage or logging activity. The European Commission SIBERIA project had the aim of mapping an area of approximately 1 million km2 in Siberia using SAR data from two satellite sources: the tandem mission of the European Remote Sensing Satellites ERS-1/2 and the Japanese Earth Resource Satellite JERS-1. Mosaics of ERS tandem interferometric coherence and JERS backscattering coefficient show the wealth of information contained in these data but they also show large differences in radar response between neighbouring images. To create one homogeneous forest map, adaptive methods which are able to account for brightness changes due to environmental effects were required. In this paper an adaptive empirical model to determine growing stock volume classes using the ERS tandem coherence and the JERS backscatter data is described. For growing stock volume classes up to 80 m3/ha, accuracies of over 80% are achieved for over a hundred ERS frames at a spatial resolution of 50 m.

Forest mapping and monitoring with interferometric synthetic aperture radar (InSAR)

A synthetic aperture radar (SAR) is an active sensor transmitting pulses of polarized electromagnetic waves and receiving the backscattered radiation. SAR sensors at different wavelengths and with different polarimetric capabilities are being used in remote sensing of the earth. The value of an analysis of backscattered energy alone is limited due to ambiguities in the possible ecological factor configurations causing the signal. From two SAR images taken from similar viewing positions with a short time-lag, interference between the two waves can be observed. By subtracting the two phases of the signals, it is feasible to eliminate the random contribution of the scatterers to the phase. The interferometric correlation and the interferometric phase contain additional information on the three-dimensional structure of the scattering elements in the imaged area. A brief review of SAR sensors is given, followed by an outline of the physical foundations of SAR interferometry and the practical data-processing steps involved. An overview of applications of InSAR to forest mapping and monitoring is given, covering tree-bole volume and biomass, forest types and land cover, fire scars, forest thermal state and forest canopy height.

Analysing forest recovery after wildfire disturbance in boreal Siberia using remotely sensed vegetation indices

Wildfires have major effects on forest dynamics, succession and the carbon cycle in the boreal biome. They are a significant source of carbon emissions, and current observed changes in wildfire regimes due to changes in climate could affect the balance of the boreal carbon pool. A better understanding of postwildfire vegetation dynamics in boreal forests will help predict the future role of boreal forests as a carbon sink or source. Time series of Normalized Difference Vegetation Index (NDVI) and Normalized Difference Shortwave Infrared Index (NDSWIR) derived from Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Terra satellite were used to investigate whether characteristic temporal patterns exist for stands of different ages in the Siberian boreal forests and whether their postwildfire dynamics are influenced by variables such as prewildfire vegetation cover. Two types of forests, evergreen needle-leaf (ENF) and deciduous needle-leaf (DNF), were studied by analysing a sample of 78 burned forest areas. In order to study a longer time frame, a chronosequence of burned areas of different ages was built by coupling information on location and age provided by a forest burned area database (from 1992 to 2003) to MODIS NDVI and NDSWIR time series acquired from 2001 to 2005. For each of the burned areas, an adjacent unburned control plot representing the same forest type was selected, with the aim of separating the interannual variations caused by climate from changes in NDVI and NDSWIR behaviour due to a wildfire. The results suggest that it takes more than 13 years for the temporal NDVI and NDSWIR signal to recover fully after wildfire. NDSWIR, which is associated to canopy moisture, needs a longer recovery period than NDVI, which is associated to vegetation greenness. The results also suggest that variability observed in postwildfire NDVI and NDSWIR can be explained partially by the dominant forest type: while 13 years after a fire NDVI and NDSWIR are similar for ENF and DNF, the initial impact appears to be greater on the NDVI and NDSWIR of ENF, suggesting a faster recovery by ENF.

Coupling of Vegetation Growing Season Anomalies and Fire Activity with Hemispheric and Regional-Scale Climate Patterns in Central and East Siberia

An 18-yr time series of the fraction of absorbed photosynthetically active radiation (fAPAR) taken in by the green parts of vegetation data from the NOAA Advanced Very High Resolution Radiometer (AVHRR) instrument series was analyzed for interannual variations in the start, peak, end, and length of the season of vegetation photosynthetic activity in central and east Siberia. Variations in these indicators of seasonality can give important information on interactions between the biosphere and atmosphere. A second-order local moving window regression model called the “camelback method” was developed to determine the dates of phenological events at subcontinental scale. The algorithm was validated by comparing the estimated dates to phenological field observations. Using spatial correlations with temperature and precipitation data and climatic oscillation indices, two geographically distinct mechanisms in the system of climatic controls of the biosphere in Siberia are postulated: central Siberia is controlled by an “Arctic Oscillation–temperature mechanism,” while east Siberia is controlled by an “El Nino–precipitation mechanism.” While the analysis of data from 1982 to 1991 indicates a slight increase in the length of the growing season for some land-cover types due to an earlier beginning of the growing season, the overall trend from 1982 to 1999 is toward a slightly shorter season for some land-cover types caused by an earlier end of season. The Arctic Oscillation tended toward a more positive phase in the 1980s leading to enhanced high pressure system prevalence but toward a less positive phase in the 1990s. The results suggest that the two mechanisms also control the fire regimes in central and east Siberia. Several extreme fire years in central Siberia were associated with a highly positive Arctic Oscillation phase, while several years with high fire damage in east Siberia occurred in El Nino years. An analysis of remote sensing data of forest fire partially supports this hypothesis.

Land Degradation Assessment Using Residual Trend Analysis of GIMMS NDVI3g, Soil Moisture and Rainfall in Sub-Saharan West Africa from 1982 to 2012

Areas affected by land degradation in Sub-Saharan West Africa between 1982 and 2012 are identified using time-series analysis of vegetation index data derived from satellites. The residual trend (RESTREND) of a Normalized Difference Vegetation Index (NDVI) time-series is defined as the fraction of the difference between the observed NDVI and the NDVI predicted from climate data. It has been widely used to study desertification and other forms of land degradation in drylands. The method works on the assumption that a negative trend of vegetation photosynthetic capacity is an indication of land degradation if it is independent from climate variability. In the past, many scientists depended on rainfall data as the major climatic factor controlling vegetation productivity in drylands when applying the RESTREND method. However, the water that is directly available to vegetation is stored as soil moisture, which is a function of cumulative rainfall, surface runoff, infiltration and evapotranspiration. In this study, the new NDVI third generation (NDVI3g), which was generated by the National Aeronautics and Space Administration-Goddard Space Flight Center Global Inventory Modeling and Mapping Studies (NASA-GSFC GIMMS) group, was used as a satellite-derived proxy of vegetation productivity, together with the soil moisture index product from the Climate Prediction Center (CPC) and rainfall data from the Climate Research Unit (CRU). The results show that the soil moisture/NDVI pixel-wise residual trend indicates land degraded areas more clearly than rainfall/NDVI. The spatial and temporal trends of the RESTREND in the region follow the patterns of drought episodes, reaffirming the difficulties in separating the impacts of drought and land degradation on vegetation photosynthetic capacity. Therefore, future studies of land degradation and desertification in drylands should go beyond using rainfall as a sole predictor of vegetation condition, and include soil moisture index datasets in the analysis.

The Application of Lidar in Woodland Bird Ecology

Habitat quality is fundamental in ecology, but is difficult to quantify. Vegetation structure is a key characteristic of avian habitat, and can play a significant role in influencing habitat quality. Airborne lidar provides a means of measuring vegetation structure, supplying accurate data at high post-spacing and on a landscape-scale, which is impossible to achieve with field-based methods. We investigated how climate affected habitat quality using great tits (Parus major) breeding in woodland in eastern England. Mean chick body mass was used as a measure of habitat quality. Mean canopy height, calculated from a lidar digital canopy height model, was used as a measure of habitat structure. The influence of canopy height on body mass was examined for seven years during which weather conditions varied. The slopes and correlation coefficients of the mass/height relationships were related linearly to the warmth sum, an index of spring warmth, such that chick mass declined with canopy height in cold, late springs, but increased with height in warm, early springs. The parameters of the mass/height relationships, and the warmth sum, were also related linearly to the winter North Atlantic Oscillation index, but with a time lag of one year. Within the same wood, the structure conferring “best” habitat quality differed between years depending on weather conditions.

Observations of forest stand top height and mean height from interferometric SAR and LiDAR over a conifer plantation at Thetford Forest, UK

Estimates of forest stand mean height using airborne LiDAR (light detection and ranging) instruments have been reported previously with accuracies comparable to traditional ground‐based measurements. However, the small area covered by a LiDAR sensor in a single aircraft overpass is a significant hindrance for large‐scale forest inventories. In comparison, airborne interferometric synthetic aperture radar (InSAR) systems are also able to make estimates of surface height, but the swath coverage is often far greater, typically five or ten times that of the LiDAR coverage. A set of interferometric data takes was acquired by the ESAR airborne sensor over a managed pine plantation at Thetford Forest, UK. Scattering phase centre height estimates were made from two single‐pass X‐band acquisitions and polarimetric repeat‐pass L‐band acquisitions and compared with height estimates made from a separate LiDAR acquisition. The relationship between the scattering phase centre heights and stand top heights estimated, and the accuracy of stand top height estimates estimated from InSAR and LiDAR is quantified by the root mean square error (rmse). General yield class models by the Forestry Commission (UK) were used to estimate stand top height from a GIS database used for forest management. The longer wavelength L‐band radiation penetrates deeper into the canopy than the X‐band, and the scattering phase centre height is affected by both forest structural parameters (canopy density, understorey, and gaps) and sensor parameters (look‐angle and reduced coherence through temporal and volume decorrelation). Consequently, a simple translation of scattering phase centre height into stand top height gives noisy results for L‐band, with observed rmse values between ±3.1 m in the near‐range and ±6.4 m in the far‐range. The X‐band based top height estimates are more accurate, with rmse between ±2.9 m in the near‐ and ±4.1 m in the far‐range, which can be further reduced by an empirical incidence angle correction. Stand top height estimates from LiDAR achieved an rmse of only ±2.0 m. The X‐band scattering phase centre heights have also been related to mean stand height and are comparable with heights observed from the LiDAR sensor and field measurements. An rmse of ±2.5 m for the mean stand height estimates based on the X‐band dataset was found. Finally, we briefly discuss error propagation from the use of a terrain model, here provided by the Ordnance Survey.

Accuracy assessment of a large-scale forest cover map of central Siberia from synthetic aperture radar

Russias boreal forests host 11% of the worlds live forest biomass. They play a critical role in Russias economy and in stabilizing the global climate. The boreal forests of central and western Siberia represent the largest unbroken tracts of forest in the world. The European Commission funded SIBERIA project aimed at producing a forest map covering an area of 1.2 million square kilometres. Three synthetic aperture radars (SAR) on board the European remote sensing satellites ERS-1 and ERS-2 and the Japanese Earth resources satellite JERS-1 were used to collect remote sensing data. Radar is the only sensor capable of penetrating cloud cover and imaging at night. An adaptive, model-based, contextual classification to derive ranked total growing stock volume classes suitable for large-scale mapping is described. The accuracy assessment of the Siberian forest cover map is presented. The weighted coefficient of agreement κw is calculated to quantify the agreement between the classified map and the reference data. First, the classified map is compared with Russian forest inventory data (κw = 0.72). The inherent uncertainty in the forest inventory data is simulated by allowing for fuzziness. The effect of uncertainty on the unweighted coefficient of agreement κ is stronger than that on the weighted coefficient of agreement κw. Second, the map is compared with a more reliable, independent posterior ground survey by Russian forestry experts (κw = 0.94). The follow-on project SIBERIA-II started in January 2002 and is striving to develop multisensor concepts for greenhouse gas accounting (www.siberia2.uni-jena.de).