What Is a Satellite Image Time Series?
In a previous article, I discussed in detail everything you need to know about working with geospatial raster data. This type of data is associated with a specific geographical area. However, it's not possible to study how that area has changed over time by analysing a single raster file. This guide aims to introduce the reader to satellite image time series and explain why this data is incredibly important in addressing current and future challenges facing the Earth.
Introduction
Satellite Image Time Series (SITS) is probably the most important resource for studying how a particular area has changed over time:
A SITS is a set of satellite images taken from the same area at different times.
Each image is stored as a geographic raster file. The images are also sorted by date so that they can be analysed sequentially. What is shown here is a SITS of a crop area:
SITS images are usually taken at regular intervals. This helps to monitor changes at a constant rate. The time series above is made up of images taken every five days by the Sentinel-2 satellite, but some of them have been removed because they contain clouds, such as:
Cloud cover is a serious problem with satellite imagery and SITS because clouds reflect Sentinel-2 bands quite well.
Sentinel-2 captures 13 spectral bands, divided into visible bands (shorter wavelengths) and infrared bands (longer wavelengths). The visible bands are the only ones we can see and are made up of red, green and blue bands (visible bands). The other bands can pick up other information, but we can't see them. This is one of the reasons why SITS are not usually analysed by simply looking at them but by using different algorithms that can process all the bands.
Applications
Satellite image time series are a valuable source of information for a wide range of applications, especially for future global challenges. Here are some examples:
Agriculture
Satellite image time series can be incredibly useful in agriculture, providing information on crop health, growth and yield. By analysing changes in vegetation over time, farmers can make informed decisions about irrigation, fertilisation and other management practices. Key applications include crop monitoring, yield forecasting, irrigation management, pest and disease detection, and land use planning.
Forest monitoring
Satellite image time series can be a powerful tool for forest monitoring and management. By tracking changes in forest cover, health and composition over time, they can help identify areas of concern and inform management decisions. For example, SITS can be used to track changes in forest cover over time, helping to identify areas of deforestation and assess the extent and rate of forest loss. They can therefore be useful for improving reforestation, afforestation or forest restoration practices.
Climate monitoring
Satellite image time series can be used to monitor patterns that may be related to climate change and provide valuable information for climate research and policy. They can be used to monitor changes in sea level, sea ice and snow cover, to track the movement of glaciers and ice sheets, and to monitor changes in vegetation caused by climate change.
Urban planning
Satellite image time series can be used to monitor changes in land use and land cover over time, providing information on urban growth, expansion and land use change. This information can be used to plan and manage urban development, infrastructure and services. In addition, SITS can be used to monitor transport infrastructure such as roads, highways and railways and assess their performance and impact on urban areas. This could help to plan new transport infrastructure, optimise existing networks and assess the impact of transport on the environment.
Disaster response
Satellite imagery can be used to assess the extent and severity of damage caused by natural disasters such as earthquakes, floods, hurricanes and wildfires. SITS can be used to track the recovery of affected areas, prioritise response efforts and improve resource allocation.
The main idea behind each application is that satellite observations offer opportunities to understand how the Earth is changing, to identify the causes of these changes, and to predict future changes. As technology advances and more data becomes available, the possibilities for using satellite imagery to understand and address global challenges are virtually limitless.
SITS Analysis
From a very practical point of view, a SITS is usually stored as a collection of geographic raster files, where the filename contains the date the image was taken. The time series can be analysed using a variety of approaches, depending on the change you want to study. Most of these methods require the use of GIS software such as ArcGIS or QGIS and a good knowledge of the subject. Although this guide is an introduction to SITS, I can briefly present some techniques that you can try out:
Image differencing
One way to analyse change over time is to subtract the pixel values of two images taken at different times. This technique is called image differencing and can highlight changes in vegetation cover, land use and other features. It's a simple technique, but it may be sufficient for some simple tasks. However, depending on the chosen bands to compute the difference, image differencing could highlight changes because the images were taken by the satellite under different conditions, not because there have been real changes in the area. The code to subtract two rasters belonging to the same time series is very simple:
import rasterio as rio
first = rio.open("lombardia3/data2019/1208/20190308.tiff").read()
second = rio.open("lombardia3/data2019/1208/20190313.tiff").read()
diff = first - secondThe result will be something like this:
Yellow tints highlight parts of the raster that have changed more when compared with the previous image of the time series. Depending on the task, you can choose which bands to subtract. Moreover, a threshold can be set to prevent small changes from being displayed and only highlight larger changes.
Plotting Bands
Each pixel of a geographic raster covers an area of a certain square metre. The bands of a given pixel can be plotted to show how they have changed over time. This can be useful because it's possible to quantify how that area is changing by combining the bands. The following code takes all the images in a time series. It then plots the values of the different bands in the top-left pixel from the oldest image to the most recent:
In my case, the final line graph shows how the 13 bands of that pixel have changed from the first image in the time series to the tenth:
Furthermore, you can also use the band values to calculate various indices (reported here). These indexes can be computed on the same pixel for each image of the time series. For example, the Normalized Difference Vegetation Index (NDVI) measures the greenness and the density of the vegetation captured in a satellite image, and it's calculated in this way:
In my case (Sentinel-2 image) the NIR band is the 10th band and the red band is the 4th band. I can apply the above formula and plot how the NDVI changes over time:
3D Neural Networks
Deep learning approaches have been successfully applied to Remote Sensing applications. In particular, 3D neural networks (3D-NN) are well suited to making predictions from satellite image time series. 3D-NN is a type of neural network that takes a 3D input, and a SITS is a three-dimensional input because it's a collection of 2D images. For example, I. Gallo et al. made a 3D Feature Pyramid Network model to perform crop mapping of Sentinel-2 images time series [1]. The architecture they made takes as input 30 images, each having 48 × 48 pixels and 13 bands and returns a 48 × 48 segmentation mask. Each pixel of the output contains the class of the crop that grows in that location.
My recent studies have been towards Deep Learning models to analyse SITS. If you have specific questions feel free to ask them in the comments.
Concluding Remarks
Satellite image time series can be acquired from a variety of satellite platforms, including Landsat, MODIS and Sentinel. These images are typically acquired at regular intervals, ranging from daily to monthly, depending on the satellite platform and the specific application. The resulting image sequences can be analysed for a variety of purposes, including improving crop management, tracking forest health, monitoring the effects of climate change on the environment, planning urban infrastructure, and assisting areas affected by natural disasters. SITS can be analysed in different ways, depending on the change you want to highlight.
All images, unless otherwise noted, are by the Author.
[1] I. Gallo, R. La Grassa, N. Landro, M. Boschetti, Sentinel 2 Time Series Analysis with 3D Feature Pyramid Network and Time Domain Class Activation Intervals for Crop Mapping (2021)
