Abstract:
Due to improper agricultural and soil management, there has
been a drop in crop yield over the last few years and food security has become a major issue. For a country like India, with a
huge population to cater, the problem becomes more serious.
Since the inception of remote sensing in scientific agriculture
management, optical remote sensing along with field data has
been used for soil health monitoring and mapping. SAR or microwave remote sensing has an all-weather and high temporal data
availability which has found applications for various domains. For
soil health studies of multiple soil classes and sub-classes of same
type, both aerial and spaceborne SAR remote sensing is currently
in use for a multitude of monitoring and parameter modelling
approaches. This study utilizes Sentinel 1 A, C-band SAR remote
sensing data with VV and VH polarization channels for surface soil
moisture estimation for alluvial soil and its sub-types in Rupnagar
of Punjab state in India. While Index based OLS Regression
method for soil moisture estimation was done using backscatter
from Sentinel-1A SAR data, it was validated using Normalized
Differential Moisture Index (NDMI) generated from Sentinel 2
optical datasets. This approach though did not consider the actual
soil moisture data from field yet gave a low Root Mean Squared
Error (RMSE) of 0.5 and R2-statistics of 0.72 (72%) in training and
testing phases. The Index based OLS (Ordinary Least Squares)
Regression method for soil moisture estimation aims to establish
a technique for cases when field data is either not available or
the study area is not easily accessible. In the statistical approach
with field data, the same OLS model, when replaced by on-field
surface soil moisture data gave a RMSE of 1.9 and R2-statistics of
0.968 and 0.948 in training and testing phases respectively at
97.5% confidence level. The study is significant for using freely
available optical and SAR remote sensing data parameters synergistically, in a simplified manner for surface soil moisture estimation. The results have comparable accuracies given by studies
using commercial data and complex modelling approaches
