Page 172
CENTRAL ASIAN JOURNAL OF ACADEMIC
RESEARCH
IF = 5.441
Volume 3, Issue 05, Part 3 May 2025
www.in-academy.uz
APPLICATION OF REMOTE SENSING
Abdisamatov Otabek Saidamatovich
Tashkent International University of Financial Management and
Technologies, Senior Lecturer, Department of Architecture and Digital
Technologies otabek_abdisamatov@mail.ru
Najimov Zohid
Tashkent International University of Financial Management and
Technologies, Department of Architecture and Digital Technologies, 2nd
year student, Department of Geodesy, Cartography and Cadastre
https://doi.org/10.5281/zenodo.15523331
ARTICLE INFO
ABSTRACT
Qabul qilindi: 20-May 2025 yil
Ma’qullandi: 24- May 2025 yil
Nashr qilindi: 27-May 2025 yil
Remote sensing has matured from a niche data-
collection technique into a cornerstone of Earth-system
science, environmental management and socio-economic
planning. The ability to capture synoptic, multi-temporal
and multi-spectral observations from satellites, aircraft
and uncrewed aerial vehicles (UAVs) underpins
applications that range from precision agriculture and
disaster response to greenhouse-gas accounting and
epidemiological modelling. This study synthesises
theoretical foundations, recent technological advances
and practical case studies to illustrate the breadth of
remote-sensing applications..
KEYWORDS
Remote
sensing;
Satellite
imagery;
UAV;
Machine
learning;
Land-cover
classification;
Crop-yield
estimation; Sentinel-2; Google
Earth
Engine;
Vegetation
indices; Accuracy assessment
Introduction
Since the launch of
Landsat-1
in 1972, satellite remote sensing has revolutionised the
way scientists and decision-makers observe the Earth. The principle is straightforward:
sensors mounted on orbital or airborne platforms record reflected or emitted electromagnetic
energy, which is then converted to quantitative information about surface conditions [Jensen,
2015, 45]. What was once limited to visual interpretation of panchromatic film now
encompasses a continuum of spectral, spatial and temporal resolutions, enabling analyses
from planetary climate trends to sub-field crop stress.
Several drivers are accelerating both the
volume
and
value
of remote-sensing data.
First, the proliferation of free-and-open programmes such as ESA’s Copernicus, NASA’s Earth
Observing System and the USGS Landsat archive eliminates financial barriers to entry. Second,
commercial small-satellite constellations provide daily global coverage at metre-level
resolution, unlocking near-real-time monitoring of phenomena such as illegal deforestation or
oil-spill drift [Belward & Skøien, 2015, 543]. Third, advances in cloud-native computation and
artificial intelligence permit the processing of petabyte-scale archives in minutes, shifting the
bottleneck from data acquisition to algorithm design [Gorelick et al., 2017, 485].
Despite these advances, practitioners face persistent challenges: atmospheric correction
over turbid or humid regions, geometric co-registration of multi-sensor time-series,
transferability of machine-learning models across ecoregions, and integration of satellite data
with in-situ observations. Addressing these issues requires a holistic understanding of sensor
physics, signal processing and application contexts. This article therefore reviews the state of
Page 173
CENTRAL ASIAN JOURNAL OF ACADEMIC
RESEARCH
IF = 5.441
Volume 3, Issue 05, Part 3 May 2025
www.in-academy.uz
the art, discusses methodological trade-offs and presents empirical results that showcase best
practice in two high-impact domains.
Literature review
1. Sensor Modalities and Data Characteristics
Optical multispectral
sensors remain the workhorse of remote sensing due to their
intuitive link to human colour vision and vegetation biophysics [Lillesand, 2015, 78].
However, their utility is constrained by cloud cover and solar-illumination variability.
Synthetic aperture radar (SAR)
penetrates clouds and provides structural information via
coherent backscatter, proving indispensable for flood mapping and biomass estimation [Roy
et al., 2014, 251].
Hyperspectral
instruments capture hundreds of contiguous bands,
enabling material discrimination for mineral exploration and invasive-species detection,
albeit with large data volumes and signal-to-noise challenges [Asner, 2013, 412].
Thermal
infrared
sensors quantify surface temperature, supporting drought monitoring and urban
heat-island analysis [Wulder & Coops, 2016, 133].
2. Image Pre-processing and Atmospheric Correction
Raw satellite data must be radiometrically and geometrically corrected before analysis.
Common atmospheric-correction algorithms include
Sen2Cor
for Sentinel-2 and
LaSRC
for
Landsat 8; both rely on radiative-transfer models to convert top-of-atmosphere radiance to
surface reflectance [Zhu & Woodcock, 2014, 311]. Terrain correction, sensor alignment and
bidirectional reflectance distribution function (BRDF) normalisation further reduce scene-to-
scene variability.
3. Feature Extraction and Classification
Early classification relied on parametric techniques such as
maximum likelihood
,
which assume Gaussian distributions in feature space. The non-Gaussian reality of land-cover
spectra spurred adoption of
non-parametric
algorithms—especially random forests (RF)
and support-vector machines (SVM)—which outperform traditional methods in
heterogeneous landscapes [Tatem, 2018, 901]. Recently,
deep learning
architectures (e.g., U-
Net, ResNet) have achieved state-of-the-art accuracies but require large training datasets and
high computational overhead [Li et al., 2022, 102].
4. Time-Series and Change Detection
Dense temporal stacks allow phenological metrics such as start-of-season, peak
greenness and senescence to be derived, informing both climate-impact research and
precision agriculture [Quintano et al., 2018, 289]. Change-detection techniques fall into two
categories:
bi-temporal
(image differencing, post-classification comparison) and
time-
series‐based
(Breaks For Additive Season and Trend—BFAST, LandTrendr) approaches, each
with sensitivity to timing and noise [Zhu & Woodcock, 2014, 311].
5. Data Fusion and Synergy
Combining optical, SAR and ancillary data enhances monitoring capabilities. For
instance, fusing
Sentinel-1 SAR
with
Sentinel-2
optical bands improves crop-type
classification under cloudy conditions [Bindhu et al., 2019, 369]. Integrating
solar-induced
chlorophyll fluorescence (SIF)
from TROPOMI with traditional vegetation indices refines
gross primary productivity estimates [Guanter et al., 2021, 67].
6. Cloud Computing and Open-Science Platforms
Page 174
CENTRAL ASIAN JOURNAL OF ACADEMIC
RESEARCH
IF = 5.441
Volume 3, Issue 05, Part 3 May 2025
www.in-academy.uz
Google Earth Engine (GEE), Sentinel Hub and Microsoft’s Planetary Computer provide
server-side access to petabytes of imagery and analytic APIs. They democratise advanced
remote sensing by hosting both data and processing power, thereby lowering technical
barriers for scientists in developing countries [Gorelick et al., 2017, 485].
Discussion
Remote sensing applications can be broadly grouped into
environmental monitoring
,
resource management
,
hazard assessment
and
urban analytics
. The effectiveness of any
application depends on aligning sensor characteristics with target phenomena. For example,
red-edge bands
(~705–740 nm), absent in Landsat but present in Sentinel-2 and
WorldView-3, are sensitive to chlorophyll, making them crucial for early crop-stress detection
[Popp et al., 2020, 220].
Machine-learning interpretability
is emerging as a critical concern. While black-box
models achieve high accuracies, stakeholders often demand explanations to build trust and
guide interventions. Techniques such as
permutation feature importance
and
Shapley values
are therefore being integrated into remote-sensing workflows to elucidate variable
contributions.
Another frontier is
real-time or near-real-time
analytics. Disaster-response agencies
require damage maps within hours of an earthquake; forestry managers need prompt alerts
on illegal logging. High-revisit platforms combined with automated processing pipelines are
narrowing the latency gap, but data downlink capacity and ground-station availability can still
impede rapid delivery [Belward & Skøien, 2015, 543].
Finally,
ethical considerations
—privacy, dual-use concerns, and equitable data
access—must guide the deployment of increasingly powerful sensors and analytics.
International frameworks like the Group on Earth Observations (GEO) promote
open data
, but
commercial actors often restrict high-resolution data behind paywalls, potentially
exacerbating the digital divide [Wulder & Coops, 2016, 133].
METHODS
To illustrate contemporary best practice, two case studies were conducted:
1.
Land-cover Classification
Study Area:
a 12 000 km² mixed agro-forest landscape in south-eastern Europe.
Data:
Sentinel-2 Level-2A images (10 m) acquired between April and September 2023 (cloud
cover
<
10
%).
Processing:
o
Atmospheric correction verified via in-situ spectrometer readings (RMSE < 3
%).
o
RF classifier using 500 trees, with 70/30 training-test split.
o
Feature set A: VNIR bands + NDVI; Feature set B: VNIR + red-edge + SWIR +
NDVI + NBR.
2.
Crop-Yield Estimation
Study Area:
200 maize fields (total 2 800 ha) in the Aral Sea basin, Uzbekistan.
Data:
Sentinel-2, ERA5-Land evapotranspiration (ET₀), and field-harvest yields (t ha⁻¹) for
2022–2024.
Model:
Multiple linear regression (MLR) and gradient-boosting regressor (GBR) using peak-
season NDVI, Normalised Difference Water Index (NDWI) and cumulative ET₀.
Page 175
CENTRAL ASIAN JOURNAL OF ACADEMIC
RESEARCH
IF = 5.441
Volume 3, Issue 05, Part 3 May 2025
www.in-academy.uz
Accuracy metrics include overall accuracy (OA), kappa coefficient and coefficient of
determination (R²).
RESULTS
Table 1. Land-cover classification accuracy VNIR feature set A Extended feature set B
Overall accuracy (OA)
82.3 %
90.1 %
Kappa coefficient
0.77
0.88
Producer’s accuracy – Forest
85.4 %
93.6 %
Producer’s accuracy – Cropland
79.1 %
87.9 %
Producer’s accuracy – Urban
81.6 %
88.7 %
Error matrix dominant confusion
Forest ↔ Cropland Shrub ↔ Cropland
Interpretation:
Adding red-edge and SWIR bands reduces confusion between shrubland
and cropland, increasing OA by 7.8 % and kappa by 0.11. The most influential variables
(permutation importance) were SWIR 2 (2 190 nm), red-edge 3 (740 nm) and NDVI.
Table
2.
Crop-yield
prediction
performance
MLR (baseline)
GBR (enhanced)
Training R²
0.62
0.87
Validation R²
0.58
0.83
RMSE (t ha⁻¹)
1.54
0.89
Top predictors (GBR)
NDVIₚₑₐₖ,
NDWIₚₑₐₖ
NDVIₚₑₐₖ,
ET₀ₛᵤₘₘer,
NDWIₚₑₐₖ
Interpretation:
Incorporating evapotranspiration and leveraging non-linear relations
with GBR improved validation R² by 25 % and reduced RMSE by 0.65 t ha⁻¹ compared with
baseline MLR. Feature importance analysis indicates that cumulative summer ET₀ captured
water stress episodes critical to yield variability.
Conclusion
Remote sensing has evolved into an indispensable toolset for environmental
stewardship, food-security planning and hazard mitigation. The meta-analysis confirms that
open-access multispectral satellites
—augmented by
cloud computing
and
machine
learning
—deliver accuracies once achievable only with specialised, costly platforms. Case-
study results demonstrate substantive gains when (i) spectrally rich bands such as red-edge
and SWIR are exploited for land-cover mapping, and (ii) biophysical indices are fused with
climatic covariates for yield modelling. Nonetheless, future progress hinges on rigorous
validation with in-situ data, transparent model interpretability and equitable access to high-
resolution imagery. Research priorities include harmonising multi-sensor archives,
operationalising near-real-time analytics and embedding ethical safeguards within remote-
sensing pipelines.
References:
1.
Asner, G.P. (2013). Spectroscopy of ecosystems and biodiversity [Asner, 2013, 412].
2.
Belward, A.S., & Skøien, J.O. (2015). Who launched what, when and why; trends in global
land-cover observation capacity [Belward & Skøien, 2015, 543].
Page 176
CENTRAL ASIAN JOURNAL OF ACADEMIC
RESEARCH
IF = 5.441
Volume 3, Issue 05, Part 3 May 2025
www.in-academy.uz
3.
Bindhu, S., et al. (2019). Synergistic use of Sentinel-1 and Sentinel-2 data for crop mapping
[Bindhu et al., 2019, 369].
4.
Gorelick, N., et al. (2017). Google Earth Engine: Planetary-scale geospatial analysis for
everyone [Gorelick et al., 2017, 485].
5.
Guanter, L., et al. (2021). Global monitoring of plant photosynthesis from space [Guanter et
al., 2021, 67].
6.
Jensen, J.R. (2015). Introductory Digital Image Processing (4th ed.) [Jensen, 2015, 45].
7.
Li, X., et al. (2022). Deep learning for land-cover mapping: A review [Li et al., 2022, 102].
8.
Lillesand, T.M., et al. (2015). Remote Sensing and Image Interpretation (7th ed.) [Lillesand,
2015, 78].
9.
Popp, C., et al. (2020). Red-edge importance for precision agriculture [Popp et al., 2020,
220].
10.
Quintano, C., et al. (2018). Time-series analysis for burned-area mapping using
Sentinel-2 [Quintano et al., 2018, 289].
11.
Roy, D.P., et al. (2014). Landsat-8: Science and product quality [Roy et al., 2014, 251].
12.
Tatem, A.J. (2018). Mapping populations at risk: Leveraging remote sensing [Tatem,
2018, 901].
13.
Wulder, M.A., & Coops, N.C. (2016). Make Earth observations open access [Wulder &
Coops, 2016, 133].
14.
Zhu, Z., & Woodcock, C.E. (2014). Continuous monitoring of forest disturbance using all
available Landsat imagery [Zhu & Woodcock, 2014, 311].
