The Surface Energy Balance Algorithm for Land (SEBAL) framework in Graphics Processing Units (GPU) using Cuda and OpenCL

Abstract

Evapotranspiration (ET) is the second largest component of the terrestrial hydrological cycle on a global scale. The estimation of ET at regional scale using satellite imagery and algorithms allow the conversion of instantaneous measures in daily totals which represents a major contribution to the study of energy exchange between the biosphere and atmosphere. One of the most important remote sensing algorithms used to estimate the spatial partitioning of energy is the SEBAL - Surface Energy Balance Algorithm for Land. The time processing of this algorithm can be very high depending on the size of the image. Thus, the objective of this study was to propose a framework of the use of graphic processing units (GPU) to run the SEBAL to increase the performance and reduce the time required to run it. For implementation of SEBAL’s framework was used local data collected in a micrometeorological tower installed in an area located in the Northern Pantanal (Brazil) and Landsat 5 – TM image. We used the Model Maker function of the ERDAS 9.2 Software to run SEBAL algorithm, and its implementation in Java, CUDA and OpenCL languages. A daily ET estimated by SEBAL in JAVA, CUDA and OpenCL languages were similar to the value performed by ERDAS software. The daily ET computed by the ERDAS software was 10% lower and the SEBAL computed in JAVA language, CUDA and OpenCL was 9% lower than the ET obtained by the Bowen ratio method. The results showed that CUDA was better due to the Speed Up performance up to 35,000 times. Keywords: Wetland, evapotranspiration estimation, hyper seasonal savanna .