Mapping and modeling of soil chemical properties distribution using hyperspectral satelitte imagery
Student name: Mr Muvunyi Germain
Guide: Dr Chander Kumar Singh
Year of completion: 2014
Host Organisation: TERI University
Abstract: Hyperspectral remote sensing serves as a potential tool for soil chemical properties
mapping, it has particularity to acquire images in hundreds narrow and contiguous
spectral bands so that it can allow assessment of each individual pixel to extract
detailed soil information and characteristics with high accuracy. In addition,
Hyperspectral remote sensing plays a key role in precision agriculture which
requires rapid, updated and accurate information about soil chemical properties.
Soil chemical properties are mostly taken as soil nutrients for plants or as indices of
nutrient supply for plants. In the present work, around nine soil chemical elements
;SOM, T.N , K , P , S ,Ca Mg, Al and CEC were taken into consideration due to
their important contribution in soil chemical quality evaluation and in agriculture
productivity as well. The main objective of this study was to provide the predictive
models which explain the spatial distribution of each selected soil chemical
properties in high altitude land of Rwanda, and to provide precise and quantitative
information in form of digital soil maps using Hyperspectral remote sensing
technology. To perform this study, we used 76 soil samples which have been
analyzed in laboratory to extract soil concentration at each location; soil samples
were analyzed using mid-infrared spectrometer. On the other hand, using ENVI 4.7,
a Hyperion Image (L1 product) was atmospherically corrected and geo-referenced
with RMSE of 0.4268 .A shapefile of soil samples was created and overlaid on
Hyperion Image to facilitate extraction of soil spectrum at each sampling location.
PLSR model was used to establish relationship between soil chemical concentration
and soil spectral response in XLSTAT2014. PLSR model with Hyperion soil spectra
and soil chemical properties measured from the field provided estimations of SOM
with an R2 of 0.88 , N with an R2 of 0.6759 , S with an R2 of 0.725 , K with an R2
of 0.639 , P with an R2 of 0.847 , Ca with an R2 of 0.5145 , C.E.C with an R2 of
0.6607 for validations . On the hand, PLSR model resulted in mediocre prediction
accuracy with an R2 of 0.347 for Aluminum and R2
of 0.284 for Magnesium.
Moreover, the PLSR output models help to generate spatial distribution maps
showing the quantity of each chemical property at every location. The digital soil
maps are abundantly dominated by the range of chemical concentration of 1.01 to
2.19 % for SOM, of 9 to 13.2 % for TN, of 0.02 to 0.66 ppm for K, of 2.01 to 5 ppm
of P, of 0 to .3.25 Cmol/ kg of Ca, of 1 to 16.9 and 17 to 26 ppm of S, and of 0.2 to
12.5 Cmol/ kg of C.E.C. These maps should serve as reference while taking decision
for a future usage of soil or for any other kind of spatial planning related to
agriculture management and productivity.
Keywords: Hyperspectral, Soil chemical properties, Partial least square regression,
Hyperion image, Maps
