Publications

Rudke, AP; Fujita, T; de Almeida, DS; Eiras, MM; Xavier, ACF; Abou Rafee, SA; Santos, EB; de Morais, MVB; Martinsa, LD; de Souza, RVA; Souza, RAF; Hallak, R; de Freitas, ED; Uvo, CB; Martins, JA (2019). Land cover data of Upper Parana River Basin, South America, at high spatial resolution. INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION, 83, UNSP 101926.

Abstract
This study presents a new land cover map for the Upper Parana River Basin (UPRB-2015), with high spatial resolution (30 m), and a high number of calibration and validation sites. To the new map, 50 Landsat-8 scenes were classified with the Support Vector Machine (SVM) algorithm and their level of agreement was assessed using overall accuracy and Kappa coefficient. The generated map was compared by area and by pixel with six global products (MODIS, GlobCover, Globeland30, FROM-GLC, CCI-LC and, GLCNMO). The results of the new classification showed an overall accuracy ranging from 67% to 100%, depending on the sub-basin (80.0% for the entire UPRB). Kappa coefficient was observed ranging from 0.50 to 1.00 (average of 0.73 in the whole basin). Anthropic areas cover more than 70% of the entire UPRB in the new product, with Croplands covering 46.0%. The new mapped areas of croplands are consistent with local socio-economic statistics but don't agree with global products, especially FROM-GLC (14,9%), MODIS (33.8%), GlobCover (71.2%), and CCI (67.8%). In addition, all global products show generalized spatial disagreement, with some sub-basins showing areas of cropland varying by an order of magnitude, compared to UPRB-2015. In the case of Grassland, covering 25.6% of the UPRB, it was observed a strong underestimation by all global products. Even for the Globeland30 and MODIS, which show some significant fraction of pasture areas, there is a high level of disagreement in the spatial distribution. In terms of general agreement, the seven compared mappings (including the new map) agree in only 6.6% of the study area, predominantly areas of forest and agriculture. Finally, the new classification proposed in this study provides better inputs for regional studies, especially for those involving hydrological modeling as well as offers a more refined LU/LC data set for atmospheric numerical models.

DOI:
10.1016/j.jag.2019.101926

ISSN:
0303-2434