Time Series Analysis in Remote Sensing Using Satellite Imagery
The use of time series analysis in remote sensing is a sophisticated geospatial method applied to monitor, measure, and analyze changes on the surface of the Earth using satellite images. In contrast to conventional methods based on a single satellite image analysis, time series analysis involves the evaluation of multiple satellite observations collected at various dates to detect temporal patterns, seasonality, trends, and abrupt environmental changes.
Due to the widespread use of multispectral, hyperspectral, SAR, and high-resolution satellite imagery, the use of time series analysis has become a crucial element of modern geospatial science and environmental monitoring. Modern satellite programs, such as Landsat, Sentinel-1, Sentinel-2, and MODIS, provide persistent observations and data for wide-scale environmental monitoring and scientific research.
Integration of remote sensing, statistical, machine learning, and geospatial methods allows the extraction of the necessary information about vegetation, land-use changes, agricultural productivity, urbanization, deforestation, water resources management, and climate processes.

What Is Time Series Analysis in Remote Sensing?
Time series analysis in remote sensing involves the study of geospatial observations obtained at several time intervals to determine the changes in environmental and surface properties over time.
Every satellite image contains one observation about the surface condition acquired on a certain acquisition date. Once these observations are arranged in chronological order and consistent processing is done, the resulting set of data can be analyzed in order to determine the spatial and temporal variation.
For example, a time series analysis of Sentinel-2 images can be applied to monitor the health of the vegetation in agriculture throughout the growing season by computing NDVI values at each available observation.
A standard time series of remote sensing data can be expressed as follows:
\[ X(x,y,t,b) \]
Where:
\(x,y\) are spatial coordinates.
\(t\) is the acquisition date.
\(b\) is the spectral band or any other derived variable.
\(X\) is the measured pixel value or geophysical parameter.
Satellite Data Sources for Remote Sensing Time Series Analysis
The quality and time resolution of the time series analysis are highly dependent on the satellite data source, sensor capabilities, revisit rate, spatial resolution, and wavelength range.
There are several satellites that have provided suitable data for temporal analysis.
The Landsat Satellites
The Landsat 8 and Landsat 9 satellites have the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS), allowing the collection of multispectral and thermal information.
Key features are:
Spatial resolution: 30 meters for all multispectral bands.
Panchromatic resolution: 15 meters.
Thermal resolution: 100 meters, resampled to 30 meters.
Landsat 8 and 9 combined nominal revisit period: approximately 8 days.
Historical observations: Several decades of images from the Landsat archives.
The Landsat images are especially valuable for long-term land cover classification, vegetation change detection, urbanization, surface water analysis, and environmental studies over time.
Sentinel-2 Satellite Imagery
The Sentinel-2 mission involves two satellites with the MultiSpectral Instrument (MSI).
Some of its technical details are:
13 spectral bands.
Spatial resolution of 10, 20, and 60 meters.
Combined nominal revisit of approximately 5 days at the equator.
Visible, near-infrared, red edge, and short-wave infrared bands.
The Sentinel-2 images are commonly used for vegetation monitoring, agricultural crop classification, water quality assessment, burnt area analysis, and land cover change detection.
Sentinel-1 Synthetic Aperture Radar
In contrast to optical instruments, SAR sensors actively emit microwave pulses and record the resulting backscatter signal, thereby enabling data acquisition in day and night conditions as well as through clouds.
Significant features are:
C-Band Synthetic Aperture Radar.
Dual polarization mode: VV and VH polarization states.
Commonly observed resolution of around 10 meters for IW GRD data products.
Ability to detect surface roughness, moisture content, vegetation canopy, and other scatterers.
Sentinel-1 time series are especially useful for flood observation, soil moisture estimation, infrastructure deformation measurements, forest management, and agriculture monitoring.
MODIS Satellite Data
The MODIS datasets are ideal for regional and global temporal analysis due to the high frequency of observations.
Some applications include:
Monitoring vegetation productivity.
Studying seasonal temperatures of the land surface.
Drought monitoring.
Studying snow coverage.
Studying global land cover changes.
Ecosystem analysis on a large scale.
Despite having lower spatial resolution than Sentinel-2 and Landsat, MODIS is useful for analyzing rapid changes in the environment due to its repeated observations.
Key Components of Satellite Imagery Time Series Analysis
A successful time series analysis pipeline includes many processing steps. All of them are critical to ensure consistency and reduce potential errors in the resulting analysis.
Acquisition of Satellite Images
The initial step of the workflow includes obtaining satellite images for the designated region of interest during the specified period.
The choice of imagery is determined by the goals of the research, desired spatial resolution, revisit period, sensors used, and historical data availability.
For instance, for agricultural analysis, one may need satellite images captured every 5-10 days using Sentinel-2 satellites during the vegetation period.
Satellite images can be accessed via:
USGS EarthExplorer.
Copernicus Data Space Ecosystem.
NASA Earthdata.
Google Earth Engine.
Microsoft Planetary Computer.
When choosing satellite images, the date of acquisition, cloud cover, sensor calibration, level of processing, and spatial extent should be taken into account.
Radiometric and Atmospheric Correction
The satellite sensors detect the electromagnetic radiation in digital form. Nevertheless, this digital data may suffer from atmospheric scattering, absorption, sensor properties, and lighting condition influences.
Radiometric and atmospheric corrections are used to make sure that the data collected on different dates are comparable.
Typical pre-processing techniques are:
Radiometric calibration: Conversion of raw digital counts to physical quantities – radiance and reflectance according to the specific instrument and product.
Atmospheric correction: Compensation of atmospheric influence and derivation of surface reflectance from optical data.
Geometric correction: Alignment of the satellite image with its geographical location.
Cloud and shadow masking: Elimination of clouds, cloud shadows, cirrus, and any other atmospheric contamination.
The Sentinel-2 data are atmospherically corrected in the Level-2A products. The Level-2 products for Landsat Collection 2 provide surface reflectance and surface temperature products.
Image Registration and Spatial Alignment
Temporal satellite imagery needs to undergo geometric alignment prior to any pixel-wise change detection.
A small amount of misalignment could cause change to occur where there is none, especially at edges and corners.
Image registration is the process of matching up corresponding geographic points within several images.
Some processes involved include:
Orthorectification.
Transformation of coordinate referencing systems.
Matching ground control points.
Sub-pixel image registration.
Resampling and grid alignment.
All images used for pixel-wise change detection need to have the same coordinate referencing system, geographical coverage, resolution, and alignment.
Nearest-neighbor, bilinear, and cubic convolution are among the resampling processes used. The most suitable technique depends on the nature of the imagery and the analysis.
Temporal Compositing and Gap Filling
Satellite imagery often suffers from gaps as a result of cloud presence, sensor constraints, or other gaps during observation.
Temporal compositing is a process whereby a combination of several images that have been observed within a specified time interval is used to create an average image.
There are various compositing methods, such as:
Median compositing.
Mean compositing.
Maximum compositing.
Minimum compositing.
Pixel quality-based compositing.
For example, the monthly median NDVI can mitigate the effect of clouds and abnormalities in the data.
Other methods of gap-filling include filling in gaps with the help of temporal, spatial, or mathematical modeling.
However, it should be noted that interpolated data must be differentiated from satellite observations.
Spectral Indices for Satellite Time Series Analysis
These indices convert satellite data on reflectance into variables representing particular surface features.
Analysis of the spectral indices acquired during different periods of time allows us to conclude environmental changes.
Normalized Difference Vegetation Index (NDVI)
NDVI is one of the most common indices for vegetation monitoring.
It reflects the difference between the near-infrared and red bands.
\[ NDVI=\frac{NIR-Red}{NIR+Red} \]
where:
NIR stands for the near-infrared band.
Red stands for the red band.
NDVI ranges from -1 to +1.
Positive values usually correspond to green vegetation, and lower values correspond to sparse vegetation, bare ground, etc. The interpretation of NDVI values depends on surface and observation conditions.
Time series analysis of NDVI allows us to detect:
seasonal vegetation growth.
vegetation phases of crops.
drought-related stress on vegetation.
deforestation.
land-cover changes.
Enhanced Vegetation Index (EVI)
EVI increases vegetation sensitivity in densely vegetated regions and minimizes some atmospheric and soil background effects.
\[ EVI=G\frac{NIR-Red}{NIR+C_1Red-C_2Blue+L} \]
where,
\(G\) – is the gain factor.
\(C_1\) and \(C_2\) – are aerosol resistance coefficient.
\(L\) – is the canopy background adjustment.
NIR, Red, Blue – are spectral reflectance.
The commonly used MODIS version has \(G=2.5\), \(C_1=6\), \(C_2=7.5\), and \(L=1\).
EVI time series are useful for monitoring dense forest ecosystems, vegetation productivity, and crop growth.
Normalized Difference Water Index (NDWI)
NDWI is used to detect surface water features.
One of the NDWI formulations:
\[ NDWI=\frac{Green-NIR}{Green+NIR} \]
This formulation is suitable for detecting open water bodies.
Another vegetation water content index using NIR and SWIR band formulation. Don't confuse these two formulations since they measure different properties of the surface.
NDWI time series help in:
Reservoir monitoring.
Assessment of lake areas.
Flood extent monitoring.
Mapping wetlands.
Studying seasonality of surface water.
Normalized Burn Ratio (NBR)
NBR is an index that helps to map burn scars and vegetation disturbance caused by fires.
\[ NBR=\frac{NIR-SWIR2}{NIR+SWIR2} \]
Through comparison of pre- and post-fire NBR values, it is possible to derive a differenced Normalized Burn Ratio (dNBR).
\[ dNBR=NBR_{pre-fire}-NBR_{post-fire} \]
Technical Workflow for Time Series Analysis Using Python
Python provides a flexible environment for processing satellite imagery, extracting temporal features, implementing statistical models, and visualizing geospatial changes.
Several open-source libraries support different stages of the remote sensing workflow.
Python library | Technical application |
Raster data reading, writing, and geospatial processing | |
Geospatial raster and vector data operations | |
NumPy | Numerical computation and array manipulation |
Multidimensional geospatial data processing | |
Vector data analysis | |
Pandas | Temporal data organization and manipulation |
Matplotlib | Time series visualization |
SciPy | Statistical analysis and signal processing |
Scikit-learn | Machine learning and classification |
Statsmodels | Regression and statistical time series modeling |
Deep learning and sequential model development | |
Earth Engine Python API | Cloud-based satellite imagery analysis |
The Role of Cloud Computing in Large-Scale Satellite Time Series Analysis
The Importance of Cloud Computing in Time Series Analysis of Satellite Imagery
With the constant increase of satellite observation archives, the need for scalable infrastructure for geospatial computing has become more urgent.
Processing decades of multispectral imagery on a desktop computer may turn out to be a costly procedure when processing a large geographical area.
The use of cloud-native solutions will allow addressing these shortcomings because it is based on the availability of distributed processing, scalable storage, parallel computation, and vast catalogs of satellite imagery.
There are various solutions such as Google Earth Engine, Microsoft Planetary Computer, and Amazon Web Services that facilitate various approaches to geospatial data processing at scale.
Cloud-based processing can incorporate:
Cloud Optimized GeoTIFF (COG) data sets;
Spatiotemporal Asset Catalog (STAC) metadata;
Zarr multidimensional arrays;
Distributed raster processing;
Parallel ML workflows;
Cloud-native geospatial databases;
Automated pipelines for satellite data processing.
For large raster archives, scalable imagery hosting and cloud geospatial analysis will make the processing easier.
Future of Time Series Analysis in Remote Sensing
The future of time series analysis in remote sensing depends largely on the development of artificial intelligence, multi-sensor data fusion, cloud-native technology, and increased frequency of satellite acquisitions.
Several trends will enhance the possibilities of time series analysis in remote sensing:
AI-powered temporal modeling: More advanced architectures in deep learning allow for better recognition of complex patterns and disturbances in the environment.
Multi-sensor data fusion: Fusion of optical, SAR, thermal, and hyperspectral sensors allows for a greater comprehension of the Earth's surface.
Real-time monitoring: Automated satellite data streams provide an opportunity to observe phenomena like floods, wildfires, crop stress, and land cover disturbance in near real-time.
Foundation models for Earth observation: The pretrained models in geospatial analysis help with feature extraction, classification, and representation learning in satellite imagery.
Cloud-native geospatial infrastructure: Technologies such as COG, STAC, Zarr, and distributed computation facilitate time series analysis.
These trends transform satellite imagery from a set of separate images to a continuous flow of geospatial intelligence.
The time series analysis of remote sensing based on satellite images offers a strong methodology that enables a better understanding of environmental changes and monitoring of surface conditions and geospatial dynamics.
The utilization of Landsat, Sentinel-1, Sentinel-2, and MODIS satellite image processing, combined with the use of spectral indices, statistical tools, machine learning, and deep learning, allows for extracting temporal features from satellite images.
Nevertheless, obtaining correct results requires proper image preprocessing, spatial and temporal alignment, harmonization, and validation of models.
As the amount of satellite data and cloud computing capabilities grow further, the importance of time series analysis will increase in agriculture, forestry, urban planning, disaster management, climatology, and geospatial intelligence.
For more information or any questions regarding the LizardTech suite of products, please don't hesitate to contact us at:
Email: info@geowgs84.com
USA (HQ): (720) 702–4849
(A GeoWGS84 Corp Company)




Comments