TY - JOUR
T1 - Rapid and simultaneous estimation of certain soil physico-chemical properties by regression modelling using the hyperspectral signature of agricultural soils
AU - Arunageetha, S.
AU - Rajendran, S.
AU - Kumar, P. S.Senthil
AU - Kumaraperumal, R.
AU - Raja, S.
AU - Kannan, P.
N1 - Funding Information:
Finally, we found AdaBoost to overfit the NDCG score for low number of iterations during the validation process. This fact indicates the presence of label noise in the learning-to-rank datasets, according to experiments conducted by [13] using artificial data. We note here that noise might come either real noise in the labeling, or from the deficiency of the overly simplistic feature representation which is unable to capture nontrivial semantics between a query and document. As a future work, we plan to investigate the robustness of our method to label noise using synthetic data since this is an important issue in a learning-to-rank application: while noise due to labeling might be reduced simply by improving the consistency of the data, it is less trivial to obtain significantly Acknowledgments. This work was supported by the ANR-2010-COSI-002 grant of the French National Research Agency.
PY - 2010/8
Y1 - 2010/8
N2 - Application of remote sensing in crop production has gained its popularity in recent times due to an increased concern on issues like land degradation, soil pollution, etc. caused by imbalanced fertilization of agricultural lands. Past research in this area has focused primarily on the use of spectral signature of crops as an indirect indicator and as a response factor to the nutrient management systems imposed in soil. Spectral features of various soil properties have not been fully evaluated, especially, concerning the studies on soil fertility evaluation. In this study, spectral evaluation on selected soils was done to determine wavelengths and /or combinations of wavelengths that are indicative of certain soil properties (pH, EC, OC, available phosphorus, available sulphur and available potassium). Further, based on the reference spectral points and reflectance inflection difference (RID) values, various prediction models were developed using simple linear regression (SLR) and multiple linear regression (MLR) approaches and were evaluated as a foundation towards the establishment of functional spectral library' for the coastal soils of Tamil Nadu. The results revealed that certain spectral reflectance at specific wavelengths and RID value based spectral regions have been proved to exhibit significant relationships with specific soil properties. The MLR based models were found to provide better estimates of the soil attributes than SLR ones and the models involving RID values were highly effective for quantifying the soil attributes than the individual spectral reflectance values of the soil. Laboratory spectral reflectance (SR) yielded high correlations with traditional laboratory analyses using MLR equations with R2 values > 0.70 for all the soil attributes studied. Therefore, for various soil fertility evaluation studies, soil properties can be predicted directly, better with VNIR reflectance spectroscopy than the conventional approach involving the indirect laboratory methods.
AB - Application of remote sensing in crop production has gained its popularity in recent times due to an increased concern on issues like land degradation, soil pollution, etc. caused by imbalanced fertilization of agricultural lands. Past research in this area has focused primarily on the use of spectral signature of crops as an indirect indicator and as a response factor to the nutrient management systems imposed in soil. Spectral features of various soil properties have not been fully evaluated, especially, concerning the studies on soil fertility evaluation. In this study, spectral evaluation on selected soils was done to determine wavelengths and /or combinations of wavelengths that are indicative of certain soil properties (pH, EC, OC, available phosphorus, available sulphur and available potassium). Further, based on the reference spectral points and reflectance inflection difference (RID) values, various prediction models were developed using simple linear regression (SLR) and multiple linear regression (MLR) approaches and were evaluated as a foundation towards the establishment of functional spectral library' for the coastal soils of Tamil Nadu. The results revealed that certain spectral reflectance at specific wavelengths and RID value based spectral regions have been proved to exhibit significant relationships with specific soil properties. The MLR based models were found to provide better estimates of the soil attributes than SLR ones and the models involving RID values were highly effective for quantifying the soil attributes than the individual spectral reflectance values of the soil. Laboratory spectral reflectance (SR) yielded high correlations with traditional laboratory analyses using MLR equations with R2 values > 0.70 for all the soil attributes studied. Therefore, for various soil fertility evaluation studies, soil properties can be predicted directly, better with VNIR reflectance spectroscopy than the conventional approach involving the indirect laboratory methods.
KW - Hyperspectral
KW - Reflectance spectroscopy
KW - Remote sensing
KW - Soil nutrients
KW - Soil testing
UR - http://www.scopus.com/inward/record.url?scp=79957790624&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=79957790624&partnerID=8YFLogxK
M3 - Article
AN - SCOPUS:79957790624
SN - 0972-3226
VL - 11
SP - 339
EP - 344
JO - Research on Crops
JF - Research on Crops
IS - 2
ER -
