Search

We developed a large-area preprocessing framework for multisensor Landsat data, capable of processing large data volumes. Cloud and cloud shadow detection is performed by a modified Fmask code. Surface reflectance is inferred from Tanre's formulation of the radiative transfer, including adjacency effect correction. A precompiled MODIS water vapor database provides daily or climatological fallback estimates. Aerosol optical depth (AOD) is estimated over dark objects (DOs) that are identified in a combined database and image-based approach, where information on their temporal persistency is utilized. AOD is inferred with consideration of the actual target reflectance and background contamination effect. In case of absent DOs in bright scenes, a fallback approach with a modeled AOD climatology is used instead. Topographic normalization is performed by a modified C-correction. The data are projected into a single coordinate system and are organized in a gridded data structure for simplified pixel-based access. We based the assessment of the produced data set on an exhaustive analysis of overlapping pixels: 98.8% of the redundant overlaps are in the range of the expected ±2.5% overall radiometric algorithm accuracy. AOD is in very good agreement with Aerosol Robotic Network sunphotometer data ( $R^{2}$ : 0.72 to 0.79, low intercepts, and slopes near unity). The uncertainty in using the water vapor fallback climatology is approximately ±2.8% for the TM SWIR1 band in the wet season. The topographic correction was considered successful by an investigation of the nonrelationship between the illumination angle and the corrected radiance.

Satellite-derived land surface phenology (LSP) serves as a valuable input source for many environmental applications such as land cover classifications and global change studies. Commonly, LSP is derived from coarse-resolution (CR) sensors due to their well-suited temporal resolution. However, LSP is increasingly demanded at medium resolution (MR), but inferring LSP directly from MR imagery remains a challenging task (e.g., due to acquisition frequency). As such, we present a methodology that directly predicts MR LSP on the basis of the respective CR LSP and MR reflectance imagery. The approach considers information from the local pixel neighborhood at both resolutions by utilizing several prediction proxies, including spectral distance and multiscale heterogeneity metrics. The prediction performs well with simulated data $(R^{2} = 0.84)$ , and the approach substantially reduces noise. The size of the smallest reliably predicted object coincides with the effective CR pixel size (i.e., field-of-view). Nevertheless, even subpixel objects can be reliably predicted provided that pure CR pixels are located within the search radius. The application to real MODIS LSP and Landsat reflectance well preserves the phenological landscape composition, and the spatial refinement is especially striking in heterogeneous agricultural areas, where, for example, the circular shape of center pivot irrigation schemes is successfully restored at MR.

A novel multiannual land-cover-classification scheme for classifying hypertemporal image data is suggested, which is based on a supervised decision fusion (DF) approach. This DF approach comprises two steps: First, separate support vector machines (SVMs) are trained for normalized difference vegetation index (NDVI) time-series and mean annual temperature values of three consecutive years. In the second step, the information of the preliminary continuous SVM outputs, which represent posterior probabilities of the class assignments, is fused using a second-level SVM classifier. We tested the approach using the 10-day maximum-value NDVI composites from the ldquoMediterranean Extended Daily One-km Advanced Very High Resolution Radiometer Data Setrdquo (MEDOKADS). The approach increases the classification accuracy and robustness compared with another DF method (simple majority voting) and with a single SVM expert that is trained for the same multiannual periods. The results clearly demonstrate that DF is a reliable technique for large-area mapping using hypertemporal data sets.