Advanced Remote Sensing: Lab 2 Surface Temperature Extraction from Thermal Remote Sensing Data

Goals and Background

The main goal of this lab exercise is to equip us students with the skills of extracting land surface temperature information from thermal bands of satellite images and account for variations in land surface temperature over space. In order to accomplish this there are 3 main objectives for this lab:

1 Learn about spectral emittance collected by sensors and visually identify variations in relative land surface temperature

2) Build a model to quantitatively estimate land surface temperature from thermal bands 

3) Combining simple models to create larger more complex models

Methods

We made heavy use of ERDAS Imagine 2015 in this lab to aid in the creating and running of models to extract and compensate for errors in the imagery that are used by various forms of atmospheric interference. These models allow use to work with corrected images with the majority of atmospheric interference removed, increasing the accuracy and quality if the data.

Visual identification of relative variations in land surface temperature 

The first portion of the lab is dealing with comparing low gain and high gain bands to look at slight variations that occur between the two in radiometric qualities of the same study area. We compared spectral reflectance bands in the imagery to thermal bands. Spectral reflectance are just what they sound like, they are reflecting solar energy that is collected by the sensor. Thermal imagery works differently in that what the sensor is collecting is emittance. Objects absorb solar energy or heat during the day and as the sun goes down and is less intense these objects emit that energy or give it off as they cool, this is what a thermal camera is collecting. Reflectance are much more intense and easier for the sensor to collect and can be collected really any time there is enough light for the sensor to gather the data. With thermal emittance there are times of the day which are much better for data collection such as early evening or even night time. Thermal data collection does not require the sun to be out because it is collecting the energy given off by objects. 

We study a thermal aerial image of the Eau Claire area to determine places of low, medium and high emmittance. The rate of emittance is determined by an objects thermal inertia. Objects that heat up and cool down quickly have low thermal inertia and the opposite is also true. Water bodies have a high thermal inertia because it takes them a long time to heat up but they then store that heat for a long time. In our imagery water bodies are then a cool feature giving off low amount of heat energy. An example of medium emittance or moderately warm object is vegetation. High emittance are coming from concrete or asphalt which heats up and cools down much more quickly that other surfaces because of low thermal inertia.

Part 2: Calculation of land surface temperature from ETM+ image

Section 1: Conversion of Digital Numbers (DN) to at-satellite radiance

The first step to calculating land surface temperature in a ETM+ image is to convert the Digital Number. To do this conversion a mathematical equation is used which is L(lambda)= Grescale*DN+ Brescale. Before you can fill out this equation you have to solve for Grescale. It also has a formula. Grescale = (LMAX-LMIN)/(QCALMAX-QCALMIN). Brescale is found by looking at the LMIN value. All of these values needed to fill into the equations are found in the meta data of the imagery. Open the meta data in WordPad or a similar program and it will look like Figure 1. Filling in these equations gives the user values which are then put into model maker in ERDAS Imagine 2015.  

Figure 1 This is the meta data table from which values for calculating Grescale, Brescale and other equations come from.

Open the model tool and insert  the input original image, the function or equation that is going to create the new image and then name the new image and select where it will be saved. Figure 2 is the model and function used to create a new image in this first step of land surface temperature calculation which will show the at-satellite radiance or emittance instead of the the DN values.

Figure 2 This is the model and function for converting the DN values to the at-satellite emittance values.

Section 2: Conversion of at-satellite radiance to blackbody surface temperature

Now that we have the radiance values for the imagery the next step is to convert those values through use of another model into blackbody temperature. This is done to correct the temperature values which will be different at the satellite, or the true/kinetic temperature, compared the temperatures on the earths surface. The equation used to convert radiance values to the surface temperature is: Tb= (K2/ln((K1/L))+1)). The K1 and K2 values are the thermal band caliration constants for the ETM+ and TM satellites. Once you have the values for the equation a new model is created with the radiance image from step one as the input, a new function which we just created and an output image of the ground temperature in the imagery. Figure 3 is the model and Figure 4 below in the results section is the resulting image. Figure 5, also found in results, is the same image brought into ArcMap where you can use the select tool to find the temperature of different items in the image. For example we found the Chippewa Valley Regional Airport and took the temperature of the concrete in the runway. Areas of red are higher temperature than areas in oranges and yellows. 

Figure 3 This is the model and function for converting the at-satellite emittance to the blackbody surface temperature. 

 Part 3: Calculation of land surface temperature from TM image

This third potion of the lab is basically a repeat of the first two steps only instead of running the models separate we are going to combine them and run the process from start to finish in one larger model. Figure 6 is the large combined model (Figure 6) The first image input is the Landsat TM image. Next is the function to convert the DN to at-satellite values just like Figure 2. This creates an output but instead of an actual separate output image we create it as a temporary image which becomes the input for the next function. This function is the same as Figure 3 converting the radience temperature to blackbody or surface temperature. The final output is the surface temperatures for the Landsat TM image (Figure 7) in results, which can be brought into ArcMap to look at surface temperature just as we did with Figure 5. 

Figure 6 This is the combined model making use of a temporary image to create the surface temperature final image,

Part 4: Calculation of land surface temperature from Landsat 8 image

This portion of the lab is making use of the skills learned in the First 3 part to create a surface temperature map of Eau Claire and Chippewa Counties. We used a LandSat 8 Thermal IR image collected on May 23, 2014 at 9:48 am. Using the same procedure as Part three we calculated surface temperature. Before we ran the model however we modified the Landsat 8 image we are using so that it focused specifically on Eau Claire and Chippewa counties not the entire image. This is done by using the subset tool which allows you to bring in a shape file of the counties and basically extract those areas out of the larger image and run the analysis just on that subset part of the imagery. Once that is done we created our model. Figure 8 is the model we created again making use of the temporary image just like we did in Part 3. Figure 9 is the final map created in ArcMap using the final surface temperature image created by the Figure 6 model.

Figure 8 This is the complete model for converting the the LandSat 8 subset image to surface temperature values. Figure 9 is the final map created in ArcMap.  

Results

In this lab we learned the process of taking raw thermal imagery from multiple sattelites and convert it to surface temperature vales. These new images can be used to accurately examine the temperature of surface objects. The images below are the resulting images from the models run during the lab.

Part 2
Section 2

Figure 4 This is the converted blackbody surface temperature image displayed in ERDAS Imagine. 

Figure 5 This is the same image as Figure 4 but it is brought into ArcMap and has a color ramp assigned to the thermal values in the image.  The red areas are higher temperature and the oranges and yellows are cooler areas. Water bodies are easy to pick out as they have the lowest temperature and stand our as yellow in the imagery.

Part 3

Figure 7 This is the final surface temperature image from the LandSat TM image using the combined model.

Part 4

Figure 9 This is the final surface temperature map created from the LandSat 8 image from May 23, 2014. Blues are cooler areas and the yellows to reds are warmer. All temperature are in Kelvin so simple conversion can be done to find Fahrenheit temperatures. 

Sources

Landsat satellite image is from Earth Resources Observation and Science Center, United States Geological Survey. Area of interest (AOI) file is derived from ESRI counties vector features