- Remote sensing (1) (remove)
- Evaluation of Modis Products Over Four European Ecological Study Sites (2009)
- Global vegetation is a key component of the climate system due to its key role in geosphere-biosphere-atmosphere interactions. Understanding these processes is of important for predicting future climate and the future state of terrestrial ecosystems. Land surface properties such as the land cover type and leaf area index (LAI) are used as essential inputs in many hydrological, ecological, and climate models. They are key parameters that describe the functioning of vegetation and are required for modeling vegetation productivity, land surface climatology, global carbon budgets and agricultural outputs as influenced by resource management. Successful modeling of these processes to quantitatively and accurately characterize global dynamics requires definition of these parameters periodically and globally with high accuracy. For this purpose the MODIS-based land surface products were designed and are now regularly available worldwide. Nevertheless, analyses based on MODIS inputs of land cover and LAI must be tested with respect to their reliability, in order that we can trust and use the outputs from simulation models quantifying water and carbon balances at large scale. The purpose of the research reported here is to determine the reliability of the MODIS spectral reflectance, land cover and LAI products for European landscapes which are highly fragmented and not necessarily homogeneous at the 1 km scale characteristic of MODIS products. A stepwise analysis has been carried out for reflectance, land cover and LAI products, comparing results from ground truth data and from high resolution remote sensing images (Landsat) to the coarser scale MODIS information. In this way, the influence of landscape fragmentation on the MODIS products should be clear and advice can be given about how they should be used in land surface modelling efforts. Four European locations were chosen for study; landscapes dominated by deciduous forest at Hesse, France; by coniferous forest at Tharandt, Germany, and by forest and grassland in mountainous terrain in the Berchtesgaden National Park, Germany and in Stubai Valley, Austria. All of these landscapes, however, have a mixture of land use. In order to compare measurements at intensive study plots with MODIS (1 km resolution), it was necessary to build a bridge via remote sensing data derived with Landsat TM (30 m resolution). It was demonstrated that for all study sites, the registration accuracy of Landsat TM images did not deviate by more than half of one pixel, and that the root mean square of error (RMSE) was less than 0.3 pixel when utilizing at least 40 ground control points and nearest-neighbor resampling technique. Comparing Landsat images with aerial photography clearly demonstrated that specific study sites on the ground could be identified and that the measured characteristics could be associated with Landsat pixel properties. The evaluation results showed that the MODIS reflectance product is reasonably accurate (less than 10 % absolute error). Certainly it is appropriate to utilize reflectance data from the two types of satellite images and to use these information in comparative examinations of land cover mapping and leaf area index estimation. The land cover comparison demonstrates that both the scale applied in classifications and the number and type of land use categories that are permitted lead to important shifts in the characterization of land cover when moving from 30 m to 1 km resolution of MODIS. Fragmentation in European landscapes is a fundamental problem encountered in the use of MODIS products. A true representation of the land surface cannot be obtained from the current MODIS land cover classifications at 1 km scale. The use of these descriptors in models describing land surface properties may potentially lead to large errors. Thus, exchange between the land surface and the atmosphere of water and CO2 as estimated by models using MODIS inputs will have a high level of uncertainty, and the results must be considered with caution. The problems in classification that are encountered lead to further difficulties in land surface characterization, since the retrieval of LAI uses land cover as an input variable. At the peak of vegetation development, MODIS LAI appears to strongly underestimate values of the Landsat based maps. During winter, the comparison is even worse, but is not consistent from grassland to deciduous forest and coniferous forest. The results cast doubt on the usefulness of MODIS LAI products as input to continental scale simulation models for carbon and water balances, at least in Europe where land cover is highly modified and fragmented due to centuries of human use and management. Use of the MODIS products in Europe requires that new techniques be considered to search for compatibility in averaging and aggregating information on land cover and reflectance that is used to estimate LAI for large areas. Keywords: Remote sensing; vegetation; MODIS; Landsat; LAI; Land cover classification; reflectance; evaluation.