Journal of Remote Sensing & GIS

Abstract

Unbiasing a Stochastic Endmember Interpolator Using ENVI Object-BasedClassifiers, a Farquhar's Single Voxel Leaf Photosynthetic ResponseExplanatory Model and Boolean Time Series Statistics for ForecastingShade-Canopied Simulium damnosum s.l. Larval Habitats in Burkina Faso

Benjamin Jacob*, Robert J Novak, Laurent Toe, Moussa S Sanfo, Semiha Caliskhan, Alain Pare, Laurent Yameogo, Mounkaila Noma and Thomas Unnasch

Endmember spectra recovered from sub-meter resolution data [e.g., Quick Bird visible and near infra-red (NIR) 0.61m wavebands ratio] of an arthropod-related infectious disease aquatic larval habitat can act as a dependent variable within a least squares estimation algorithm. By so doing, seasonal endemic transmission -oriented risk variables can be accurately interpolated. Spectral mixing, however, is a problem inherent to multi-dimensional canopy-oriented arthropod-related infectious disease larval habitat feature attributes resulting in few image sub-pixel spectra representing "pure" targets. This can lead to a biased endmember target signature due to spectrally unquantitated mixed sub-pixel radiance originating from different canopy-oriented larval habitat object types. An erroneous endmember larval habitat signature will render inconsistent residual forecasts in a stochastic/deterministic interpolator. In this analyses we spectrally extracted and decomposed multiple canopied endmembers surface-oriented sub-meter resolution pixel reflectance values derived from a georeferenced QuickBird imaged canopied larval habitat of Similium damnosum s.l. (Figure 1), a black fly vector of onchocerciasis in an epidemiological riverine study site in Burkina Faso. We employed ENVI object-based classifiers, a 3-Dimensional radiative transfer equation and the Li-Strahler geometric-optical model to perform the pixel decomposition. Thereafter, the georeferenced larval habitat and the within canopy radiance values (e.g., Precambrian rock) were spectrally isolated and weighed using a robust Successive Progression Algorithm (SPA) within a Boolean domain. The decomposed endmember then rendered a robust spectral signature in ArcGIS which was subsequently kriged to identify unknown, unsampled productive S. damnosum s.l. larval habitats along a Burkina Faso river system using a blind study format. The validation model revealed a 100% correlation among the predicted georeferenced productive black fly habitat sites based on the seasonal-sampled larval density counts values.