The impact of scaling-up on biophysical variable estimation
Jensen, Indiana State University, firstname.lastname@example.org
Mausel, Indiana State University, email@example.com
Moran, Indiana University, firstname.lastname@example.org
Brondizio, Indiana University, email@example.com
Menzies, Indiana State University, menziesjohn@ hotmail.com
This study employed regression and artificial neural network techniques to model leaf area index (LAI) dynamics in the Santarem, Brazil area using ASTER, ETM+, MODIS and IKONOS satellite data to determine the effect of scaling-up with different satellite sensor resolutions. LAI values were gathered at 76 field locations characteristic of many different cover types: mature forest, secondary succession, pasture, cropped land, barren land, and urban area. The field data were used with satellite data to assess multiple regression and artificial neural network LAI predictive abilities. Assessments of model accuracy were determined by calculating RMSE. Results demonstrated that neural networks perform better than multiple regression for LAI modeling. In addition, ASTER data provided the most accurate LAI models - probably because its spatial resolution was characteristic of the 20 x 20 m sampling scheme. The poorest modeler of LAI was found to be MODIS with its 250 and 500 m resolutions indicating that much biophysical detail is lost when using coarser remote sensing data. New and expanded research is being conducted in the Altamira area of Brazil that builds upon the positive results of the Santarem study. Much more intensive neural network classification and modeling has begun in the Altamira area using multi-temporal and multi-resolution remotely sensed data supported by detail field measurements. Improved neural network modeling of selected bio-parameters using lower resolution spectral data informed by samples of higher resolution spectral and field data is anticipated.
Science Theme: LC (Land Use and Land Cover Change)