Development of spectral models for assessing soil fertility status in various agroecological sub-regions of Punjab
Loading...
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Punjab Agricultural University, Ludhiana
Abstract
The present study titled “Development of spectral models for assesing soil fertiity status in various
agroecological sub-regions of Punjab” was carried out in five agroecological sub-regions, namely, Submountainous
Siwalik Hills (SSH), Northeastern Undulating (NEU), Piedmont and Alluvial Plain
(PAP), Central Alluvial Plain (CAP) and Southwestern Alluvial Plain (SWAP). The objectives were to
develop visible-near infrared (vis-NIR) models for predicting soil properties and to examine portability
of spectral models across regions. GPS based soi fertility maps were prepared by using inverse distance
weighted (IDW) interpolation and classification. Highest soil organic carbon (SOC) and sand content
(mean 0.77 % and 74.2 %, respectively) in 0-15cm soil were observed in SSH, whereas highest
nitrogen, phosphorous and, manganese contents (mean value 104.5 kg ha-1, 40.8 kg ha-1, 7.8 ppm,
respectively) were found in NEU. The largest potassium, copper and silt levels (mean 173.1 kg ha-1,
5.21 ppm, 22.4 %, respectively) were reported in PAP. The CAP represented highest calcium
carbonate content (mean 1.36 %), while highest pH, electrical conductivity, iron and zinc (mean value
7.93, 0.95 dS/m, 29.5 ppm and 3.2 ppm, respectively) levels were observed in SWAP region. Spectral
signatures in vis-NIR range were collected using ASD Spectroradiometer. Partial least square
regression (PLSR) was emlpoyed to develop spectral models for assessing soil fertility attributes.
Various model evaluation indices e.g. root mean square error (RMSE), ratio of performance to
deviation (RPD) and coefficient of determination (R2) were used to evaluate predictive performance of
the PLSR models. The RMSE, RPD and R2 values varied between 0.05 - 0.39, 0.32 – 1.73 and 0.05-
0.69, respectively in SSH for various soil properties. Similarly, in NEU RMSE, RPD and R2 values
ranged between 0.05 – 0.39, 0.93 – 2.15 and 0.12 - 0.82, respectively. In PAP region, the RMSE, RPD
and R2 values varied between 0.07-2.90, 0.32-1.6 and 0.05-0.72, respectively. CAP region showed
RMSE, RPD and R2 values ranging between 0.03-0.43, 0.62-1.5 and 0.03-0.64, respectively. In case of
SWAP region, RMSE, RPD and R2 values ranged between 0.03-0.73, 0.72-1.5 and 0.05-0.68,
respectively. Moreover, in common model, RMSE, RPD and R2 values varied between 0.19-3.81, 0.16-
0.98 and 0.01-0.40. A comparison among models suggested about their non-portability across regions.
Based upon various performance indices, the CaCO3 and sand predictions were reliable, whereas
prediction of properties like SOC and Olsen-P was moderately reliable. The pH and EC, however,
could not be predicted much accurately. The study suggested about exploring better statistical tools to
enhance prediction performance of spectral models.