Relationship between tree species composition and phenology extracted from satellite data in Swedish forests

anglais. This study investigated the “relationship between tree species composition and phenology extracted from satellite data in Swedish forests”. The proposed method investigated in this study aims at mapping the fractional composition of deciduous/coniferous tree species and also the fractional composition of Norway spruce (Picea abies) and several pine (Pinus sp.). The fractions can then be used to classify a forest into forest types such as uniform deciduous/coniferous or mixed forests. The method uses field measurements for training a regression model against satellite derived seasonality parameters. The satellite derived phenological parameters consists of a time series of normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) values where parameters such as maximum, length, start and end of growing season were extracted with the software TIMESAT. The satellite system used was the Moderate-Resolution Imaging Spectroradiometer (MODIS). The result indicates that the EVI derived seasonality parameters correlated stronger against the fractional composition of deciduous/coniferous tree species as compared with NDVI derived seasonality. The correlation coefficient for the EVI derived seasonality was estimated to 0.88 for the best performing dataset and parameter. However, when validated against an independent dataset the accuracy proved to be low when the tested regression models were used to predict the fraction of deciduous/coniferous tree species composition. A source of error derives from differences in mapping scale between the satellite system (250x250 m) and that of the field data (either a plot size of a 7 m circle or a plot size 30x30 m). The differences in mapping scale are assumed be a major source of error. However, the correlation between satellite derived seasonality and the fractional tree species composition is strong enough to consider it worthwhile to investigate for future studies when better data will become available.

anglais. The changing of seasons is a slow process that is marked by natural annual events. Examples of these natural events include the budburst in spring or the colouring of trees in autumn. The seasonal changes of plants are a result of the adaption to the seasonal changes of their environment. These notations can become data of great scientific value when studied systemically. The systematic study of seasons is called phenology. Phenological studies can result in long, cohesive time series dataset of when phenological events occur through the year. Scientists are interested in such data because changes in phenological events can reveal changes in the environment indicating, for example, climate change. Hence, long cohesive time series datasets of phenological observations and measurements can be carriers of environmental information. The sciences of observing and gathering information about physical, biological, geometrical or chemical processes or properties of the planet Earth are part of a branch within science called Earth Observation. When observing the processes of the Earth, much can be gained by taking a step back and observing things from afar away to get a wider perspective. An observation made from far away without physical contact and with different instrument-based technologies is referred to as remote sensing. Remote sensing systems can, for example, consist of satellites generating images. These images contain information about for example vegetation health and development. The great advantage of remote sensing systems is that they offers the possibility to measure or observe large areas rather than individual specimens. Satellite measurements therefore facilitate the mapping vegetation or vegetation properties like crop yield, timber volumes, carbon stocks, land use, habitats or vegetation classes Traditional phenological observations have been made on specific specimens. However, with modern technology, satellite measurements of vegetation can be used to reconstruct phenological events like the budburst in spring. This is achieved by arranging the satellite images in time series, which allows the visualization of changes through time. In other words, phenological events can therefore effectively be mapped with chronologically arranged satellite measurements. But we can also turn things around. Mapping vegetation properties is usually done by using single satellite images. But the limitation of a single satellite image is that it contains only the information from one moment; that is to say, it is a snapshot. By arranging satellite images in a time series, with the information that describes the phenological changes of vegetation, might also facilitate the mapping of vegetation properties. This research uses satellite images arranged in a time series, which can be used to construct phenological estimations of events like season start, end, length and seasonal differences. The aim of the research is to develop a method by which the satellite derived phenological estimations can be used to map the fractional distribution of deciduous and coniferous forest. It is also tested by determining the fractional distribution of Norway spruce (Picea abies) and various pine species (Pinus sp.). This is done by statistically relating field observations of forest stands against the corresponding satellite measures. The results suggest that there exists a very strong statistically significant relationship between some of the satellitederived phenological estimations and the fraction of deciduous and coniferous forest. However, when the produced models were assessed against validation data, the accuracy of the predictions proved to be low. Many sources of errors do exist that contributed to low accuracy. The most important source of error is that the plot area of the field measurements only corresponds to a very small fraction of the area that the satellite measures in a single pixel. Normally, field data should consist of an area much larger than the area that the satellite is measuring, in order to get the methods like the tested one to work properly. This study concluded that the method works in principle but needs to be modified in several ways in order to get useful results: In a very near future, new and improved satellite systems will be launched which will make methods like the one tested much likely to generate more accurate results. Also, other statistical methods are worthwhile considering when repeating a study like this one. New spectral algorisms to treat satellite data are now available which is more suitable for the environment in which this was tested, i.e. the forested areas of Sweden.