Page 113 - AJWEP-22-6

SWAT-based LULC impacts on groundwater recharge

Figure 3. Methodologies in this study
Abbreviations: CUP: Calibration and uncertainty program; DEM: Digital elevation model; GIS: Geographical
information system; HRU: Hydrological response units; LULC: Land use and land cover; MLC: Maximum
likelihood classification; SWAT: Soil and water assessment tool.

nearest station with a long-term, continuous record. The 2.2.3. Accuracy assessment of LULC map
dataset spans the period from 1979 to 2022 and includes The accuracy of remote sensing classifications is crucial
variables such as evapotranspiration, temperature, to ensure that the classifications are consistent with field-
relative humidity, sunshine hours, and wind speed. grade reference data. 3,34,35 The classification of LULC types
Due to the presence of missing or inconsistent data, was further validated by the researchers’ experiences,
an averaging method was applied for bias correction. literature reviews, and Google Earth Pro. This study
Stations with significant data gaps were excluded from utilized both qualitative and quantitative methodologies,
the analysis. adhering to established scientific recommendations. One
35
of the most popular tools for achieving this is the confusion
2.2.2. LULC analysis matrix, from which important measures such as overall
The LULC detection and change analysis for the target accuracy (OA), producer’s accuracy (PA), user accuracy
watershed were performed with ERDAS IMAGINE (UA), and the kappa coefficient (κ ) can be derived. OA is
c
2015 (Hexagon, Sweden) and ArcGIS 10.8 software a measure of global classification accuracy, while UA is a
(Esri, United States). In this study, satellite images measure of the ability of a pixel classified in the presumed
with cloud cover <10% of the imagery were selected to category to be similar to the real-world category. PA
acquire accurate LULC data (Table 2). Landsat 7 was represents the proportion of a map in which a reference
used for the 2000 LULC map, Landsat 5 for the 2010 pixel is correctly classified, whereas κ represents the
c
LULC map, and Landsat 8 for the 2022 LULC map. proportion of agreement between the classification and
Image processing and classifications were conducted reference data after correcting for chance. The κ values
c
using algorithms such as contrast enhancement, edge have the following interpretations: >0.80 indicates high
detection, and haze correction to enhance the satellite agreement, 0.40–0.80 indicates moderate agreement, and
images based on the ERDAS IMAGINE 2015 software. <0.40 indicates poor agreement 35,36 The κ was calculated
c
The maximum likelihood classification method was using Equation I:
employed, based on region-specific training information, x
to delineate LULC within the watershed. This algorithm i N (I)
ii
2
has been used previously. 33 c N x

Volume 22 Issue 6 (2025) 107 doi: 10.36922/AJWEP025180139