Hyperspectral spectroscopic study of soil properties in acid soils of North East India.
Loading...
Date
2020-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
College of Post Graduate Studies in Agricultural Sciences, CAU-Imphal, Umiam
Abstract
Periodic monitoring of soil nutrient status for crop production is vital for growing population. The conventional soil analysis has several disadvantages which may be overcome by using real time optical remote sensing data. However, interpretation of nutrient content of soil is difficult using few wide spectral bands (5-10 bands). On the contrary, a large number of narrow spectral bands (350 to 2500 nm) are used in hyperspectral remote sensing. With these advantages, it is attempted to study spectral signature of soil and their relationship with soil attributes and to develop spectral indices using hyperspectral spectroscopy in acid soils of Ri Bhoi district of Meghalaya, India.
Spatial stratification of 21 different classes was generated from Land Use Land Cover (LULC) map prepared by National Remote Sensing Centre and soil order map prepared by National Bureau of Soil Survey and Land Use Planning. Five random soil sampling sites were selected from each stratum and recorded the spectral signature of soil in-situ. A composite soil sample at 0-10 cm depth was made from 10 random soil samples from each stratum. Soil attributes viz. pH, soil organic carbon (SOC), available nitrogen (N), available phosphorus (P2O5), available potassium (K2O), available zinc (Zn), sand, silt, clay and moisture were estimated using standard procedures. Field spectra were recorded using field portable spectroradiometer and laboratory soil spectra were recorded using contact probe. Suitable bands for each soil attribute were identified using Variable Importance in Prediction (VIP). Four commonly used indices such as Simple Ratio Index (SRI), Normalized Difference Index (NDI), Renormalized Difference Index (RDI) and Modified Simple Ratio Index (MSRI) were used for development of spectral indices for soil attributes. Soil attributes were also predicted using Partial Least Square Regression (PLSR) models.
Highest and lowest field spectral reflectance was found in Kharif Crop (0.03 to 0.27) & Double Crop (0.02 to 0.14) under Alfisols, Wastelands (0.04 to 0.34) and Deciduous Forest (0.02 to 0.16) under Inceptisols and Double Crop (0.04 to 0.31) & Abandoned Jhum (0.02 to 0.22) under Ultisols,respectively. Again, highest and lowest reflectance of field spectra was observed in Inceptisols (0.03 to 0.30) & Ultisols (0.02 to 0.22) under Abandoned Jhum, Ultisols (0.03 to 0.26) & Inceptisols (0.02 to 0.25) under Current Jhum, Ultisols (0.04 to 0.28) & Inceptisols (0.02 to 0.16) under Deciduous Forest, Ultisols (0.04 to 0.31) & Alfisols (0.02 to 0.14) under Double Crop, Alfisols (0.02 to 0.29) & Ultisols (0.01 to 0.29) under Evergreen Forest, Inceptisols (0.03 to 0.30) & Ultisols (0.02 to 0.25) under Kharif Crop and Inceptisols (0.04 to 0.34) & Alfisols (0.02 to 0.18) under Wastelands.
On the other hand, highest and lowest laboratory spectral reflectance values were recorded in Kharif Crop (0.10 to 0.62) & Deciduous Forest (0.09 to 0.49) in Alfisols, Evergreen Forest (0.10 to 0.62) & Current Jhum (0.09 to 0.51) in Inceptisols and Double Crop (0.09 to 0.62) & Deciduous Forest (0.09 to 0.54) in Ultisols. Again, highest and lowest laboratory spectral reflectance were recorded in Ultisols (0.09 to 0.58) & Alfisols (0.08 to 0.55) under abandoned jhum, Alfisols (0.09 to 0.57) & Inceptisols (0.09 to 0.51) under current jhum, Inceptisols (0.09 to 0.55) & Alfisols (0.09 to 0.49) under deciduous forest, Ultisols (0.09 to 0.62) & Alfisols (0.10 to 0.57) under Double Crop, Inceptisols (0.01 to 0.62) & Ultisols (0.08 to 0.55) under Evergreen Forest, Alfisols (0.10 to 0.62) & Inceptisols (0.11 to 0.58) under Kharif Crop and Alfisols (0.10 to 0.59) & Inceptisols (0.09 to 0.54) under wastelands.
Descriptive statistics of the soil attributes showed medium level of variability ( madian) and normal distribution of the data (skewness < 3.0).
VIP score for pH (2.65), SOC (2.87), N (2.75), P2O5 (2.41), K2O (2.19) Zn (2.93), sand (3.59), silt (2.44), clay (3.29), and moisture (3.15) has been found in 1383 nm, 1368 nm, 1391 nm, 1417 nm, 1417 nm, 1368 nm, 1368 nm, 1368 nm, 1368 nm and 1417 nm field spectra. Again, Highest VIP score for pH (2.03), SOC (1.80), N (1.73), P2O5 (2.12), K2O (2.39) Zn (2.38), sand (2.21), silt (2.69) and clay (1.89) has been found in 2203 nm, 2386 nm, 2205 nm, 2205 nm, 2396 nm, 2206 nm, 2205 nm, 2205 nm and 2267 nm laboratory spectra.
Ratio of Prediction to Deviation (RPD) value derived from field spectra was found to be highest in RDI (1697 nm, 1751 nm) for pH (1.06), SRI (615 nm, 628 nm) for SOC (1.11), RDI (413 nm, 2178 nm) for N (1.07), RDI (852 nm, 903 nm) for P2O5 (1.16), RDI (2246 nm, 2353 nm) for K2O (1.07), SRI (2445 nm, 2458 nm) for Zn (1.06), SRI (2369 nm, 2135 nm) for sand (1.10), RDI (776 nm, 779 nm) for silt (1.07), RDI (2089 nm, 2333 nm) for clay (1.15) and MSRI (2358 nm, 2005 nm) for moisture (1.07). Again, RPD value derived from laboratory spectra was found to be highest in SRI (1416 nm, 1394 nm) for pH (1.09), RDI (423 nm, 424 nm) for SOC (1.09), SRI (1994 nm, 1995 nm) for N (1.08), SRI (414 nm, 403 nm) for P2O5 (1.10), SRI (414 nm, 403 nm) for K2O (1.12), SRI (2445 nm, 2458 nm) for Zn (1.06), SRI (2410 nm, 2369 nm) for sand (1.08), RDI (1514 nm, 1515 nm) for silt (1.07) and RDI (2457 nm, 2459 nm) for clay (1.07).
Soil attributes predicted by PLSR model using field spectra showed highest RPD value in clay (1.60) which was followed by K2O (1.57), pH (1.56), soil moisture (1.42), P2O5 (1.41), sand (1.40), SOC (1.34), N (1.27), Zn (1.33) and silt (1.32). Again, Soil attributes predicted by PLSR model using laboratory spectra showed highest RPD value in soil pH (2.01) which was followed by Zn (1.83), sand (1.77), SOC (1.76), clay (1.62), silt (1.60), N (1.59), P2O5 (1.56) and K2O (1.54).
Spectral signature of soil was different under selected LULC. VIP has been found to be useful in studying the relationship of spectral reflectance with soil attributes. Spectral indices were found to poorly predict soil attributes in field as well as laboratory spectra. Soil pH, P2O5, K2O, sand, clay and moisture could be fairly predicted by PLSR models using field spectra. SOC, N and silt could be fairly predicted by PLSR model using laboratory spectra. Good quantitative prediction could be achieved for Zn through PLSR model using laboratory spectra. Again, very good quantitative prediction could be achieved for soil pH through PLSR model using laboratory spectra.
Description
Keywords
Soil properties, Acid soils