B51H-0523
Spatial Resolution Effects of Remote Sensing Informed Soil Nutrient Models Based on Landsat 8, RapidEye, WorldView-2 and GeoEye-1 Images
Friday, 18 December 2015
Poster Hall (Moscone South)
Yiming Xu1, Sabine Grunwald2, Christopher M. Clingensmith2, Amr Abd-Elrahman3, Scot E. Smith4 and Suhas Wani5, (1)University of Florida, Soil and Water Science, Ft Walton Beach, FL, United States, (2)University of Florida, Soil and Water Science, Gainesville, FL, United States, (3)University of Florida, Gulf Coast Research and Education Center– Plant City Campus, Plant City, FL, United States, (4)University of Florida, School of Forest Resources & Conservation, GAINESVILLE, FL, United States, (5)International Crops Research Institute, Patancheru, India
Abstract:
Soil nutrient storage is essential and important to maintain food security and soil security in smallholder farm settings. The objective of this research was to analyze the spatial resolution effects of different remote sensing images on soil prediction models in Kothapally, India. We utilized Bayesian kriging (BK) to characterize the spatial pattern of total nitrogen (TN) and exchangeable potassium (Kex) in the topsoil (0-15 cm) at different spatial resolutions by incorporating spectral indices from Landsat 8 (30m), RapidEye (5m) and WorldView-2/GeoEye-1 images (2m). The band ratio of red to green, red to blue and green to blue, Crust Index and Atmospherically Resistant Vegetation Index from multiple images generally had high linear correlations with TN and Kex. The BK model of TN based on WorldView-2 and GeoEye-1 attained the highest model fit (R2=0.41) and lowest prediction error (RMSE=171.35 mg kg-1) compared with the BK models of TN based on Landsat 8 (R2=0.30; RMSE=182.26 mg kg-1) and RapidEye (R2=0.28; RMSE=183.52 mg kg-1). The BK model of Kex based on Landsat 8 had the highest model fit (R2=0.55) and the second lowest prediction error (RMSE=79.57 mg kg-1) compared with the BK models of Kex based on WorldView-2 and GeoEye-1 (R2=0.52; RMSE=79.42 mg kg-1) and RapidEye (R2=0.47; RMSE=83.91 mg kg-1). The lowest prediction fit and highest prediction error of soil TN and Kex models based on RapidEye suggest that the effect of fine spatial remote sensing spectral data inputs do not always lead to an increase of model fit. Soil maps based on WorldView-2 and GeoEye-1 have significant advantages in characterizing the variation of soil TN and Kex spatial pattern in smallholder farm settings compared with coarser maps. This research suggests that Digital Soil Mapping utilizing remote sensing spectral data from WorldView-2 and GeoEye-1 has high potential to be widely applied in smallholder farm settings and help smallholder farmers manage their soils and attain soil security.
