Analysis of the spectral signature of soils and prediction of nutrients through remote sensing and GIS based models | |
| Author | Waktola, Daniel Kassahun |
| Call Number | AIT DISS. no. SR-03-02 |
| Subject(s) | Soils--Analysis Soil fertility |
| Note | Submitted in partial fulfillment of the requirement for the degree of Doctor of Philosophy |
| Publisher | Asian Institute of Technology |
| Abstract | Employing remote sensing tools for monitoring and mapping of soil fertility parameters could offer enhanced possibilities. This is due to the objectivity of data, wider area coverage, continuous monitoring, and faster acquisition capabilities than the conventional methods. However, apart from the challenges induced from plant cover, other obstacles like indiscriminability of soil nutrients from satellite platform and the incompatibility of the sensors' resolutions considerably hinder the direct application of the technology. This research was undertaken with a prime objective of developing predictive models for Soil Organic Matter (SOM), phosphorus (P), potassium (K), and iron (Fe), from quantified spectral-chemical patterns. Data were computed from multi sources of detecting platforms, viz., laboratory-based spectrometer, field-based photometer, and optical satellite-based sensors. Samples were collected from plough layers of Lop Buri tropical soils, Thailand, during a satellite-synchronized ground survey. Both radiometric and chemical analysis were carried out following the standard field and laboratory procedures. The first part of the research addressed the modeling of nutrients from the field and laboratory generated reflectance spectra. The data were synthesized across a 10, 20, 50, and 100 nm bandwidth categories. Accordingly, a single, double, and 100 nm-based bandwidth categories were synthesized from the field spectral data. Stepwise Multiple Regression, Polynomial models, and Artificial Neural Network (ANN) analysis were employed for the identification of nutrient-sensitive bands and subsequent model development. The results revealed higher degree of predictions obtainable from the laboratory than the field conditions. ANN appeared to offer better prediction than the Multiple Regression and Polynomial models. Furthermore, the narrower bandwidth categories have outperformed the broader bandwidth category, where the trend, in this research, is termed as "bandwidth decay effect''. In the second pa1t, the radiometrically, geometrically, and atmospherically corrected IRS-ID satellite data were processed through several indices to yield nutrient-specific prediction models. These models were built through the synthetically cloned channels, called "Spech·al Band Cloning" (SBC). The SBC had interwoven the fine-scaled spectrometer signatures with the course-scaled satellite data. This enabled the development of an original modeling approach. The conclusion that emerges from this model is that, although SBC has performed lower than the spectrometer-based models, it has produced adequate results, which were impossible to realize from the satellite alone. An R2 of 0.72, 0.62, 0.70, and 0.68 were attained for SOM, P, K, and Fe, respectively. Measured and predicted maps of each soil nutrient type, along with the overall weighted fertility surface, were generated from the interpolated nutrient surfaces, performed on a GIS environment. Models were validated on unused (independent) dataset, vis-a-vis the conventionally measured soil nutrients, and had attained an R2 of0.69, 0.49, 0.64, and 0.60 for SOM, P, Kand Fe, respectively. |
| Year | 2003 |
| Type | Dissertation |
| School | School of Advanced Technologies (SAT) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Space Technology Application and Research (SR) |
| Chairperson(s) | Tripathi, Nitin K.;Honda, Kiyoshi; |
| Examination Committee(s) | Apisit Eiumnoh; |
| Scholarship Donor(s) | Alemaya University / World Bank; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2003 |