Welcome to the Harris Geospatial product documentation center. Here you will find reference guides, help documents, and product libraries.


Harris Geospatial / Docs Center / Using ENVI / Crop Science / Hotspot Analysis

Hotspot Analysis

Hotspot Analysis

Use Hotspot Analysis to identify areas in an image that are relatively different than the rest of the image. ENVI calculates Getis-Ord Gi* local statistics (Getis and Ord, 1992) to group neighboring pixels of similar value into clusters. The algorithm evaluates each pixel and its surrounding pixels within a specified distance to classify the pixel as "hot" or "cold" (statistically significant clusters of high or low values, respectively) or neutral (not statistically significant). Hotspot analysis can be used to look for variations in crop health throughout an area.

This tool is part of ENVI Crop Science, which requires a separate license and installation. You can also write a script to run hotspot analysis using ENVIAgSpectralHotspotAnalysisTask.

See the following sections:

Before You Begin


Two tools are available for hotspot analysis: The Hotspot Analysis tool takes a single-band raster as input, and the Spectral Hotspot Analysis tool takes a multi-band image as input and requires you to choose a spectral index to compute.

Before creating a spectral index image, the pixel values in the multi-band input image should be converted from DN values to radiance or top-of-atmosphere reflectance using the ENVI Radiometric Calibration tool. If you have the ENVI Atmospheric Correction module licensed and installed, you can further correct the data to apparent surface reflectance.

To use Hotspot Analysis for assessing crop health at a finer scale (for example, within a field), you should mask out all non-vegetation pixels such as soil and man-made objects.

You should also spatially subset the image to only include your area of interest. If using Hotspot Analysis for a single agricultural field, create a spatial subset that includes the extent of the field, or mask out the rest of the image outside of the field extent. The following section describes how to subset by region of interest (ROI).

Example of Subsetting by ROI

You can draw an ROI around a field, mask out the pixels outside of the field, then spatially subset the image to include the extents of the field. Follow these steps:

  1. Right-click on the input multi-band image in the Layer Manager and select New Region of Interest.
  2. Draw a polygon ROI around the field of interest. Right-click and select Complete and Accept Polygon.
  3. From the ENVI menu bar, select File > Save As > Save As (ENVI, NITF, TIFF, DTED).
  4. In the File Selection dialog, select the input image but do not click OK yet.
  5. Click the Mask button.
  6. Select the ROI that you just created, and click OK.
  7. Click the Spatial Subset button.
  8. Click the Subset by ROI button .
  9. In the ROI Selection dialog, select the ROI that you created and click OK.
  10. Select an output filename and location for the spatially subsetted image, and click OK.

The following image shows an example. Pixels outside of the field are set to values of NoData.

Background


Getis-Ord Gi* statistics (Getis and Ord, 1992) are typically used in vector feature space to look for statistically significant trends and anomalies. However, the same equations can be applied to raster data on a pixel-by-pixel basis. For rasters, hotspot analysis looks at each pixel and its surrounding pixels within a specified distance and compares them to the entire scene.

The result of Getis-Ord Gi* analysis is an array of Z-scores, one for each pixel, which is the number of standard deviations that the pixel and its neighbors are from the global mean. High Z-scores indicate more intense clustering of high pixel values, indicating hot spots. Low Z-scores indicate more intense clustering of low values, indicating cold spots. Individual pixels with high or low values by themselves might be interesting but not necessarily significant.

Run Hotspot Analysis


Choose one of the following options.

Hotspot Analysis

Use this option to run hotspot analysis using a single-band raster, without specifying a spectral index image.

  1. From the Toolbox, select Crop Science > Hotspot Analysis.
  2. Select a single-band raster for input.
  3. Select a Search Distance, which is the value used in the Getis-Ord calculation to determine how pixels are clustered. If the input raster has a valid spatial reference, then distance is measured in meters. Otherwise, it is measured in pixels. See the Examples section below for a comparison of different distance values. The input distance determines the spatial resolution over which the clustering occurs.
  4. Click the Browse button next to Output Raster, and select an output filename and directory.
  5. Select the Display result option to display the output image when processing is complete.
  6. Enable the Preview check box to see a preview of the settings before you click OK to process the data. The preview is calculated only on the area in the Image window and uses the resolution level at which you are viewing the image.
  7. Click OK. ENVI adds the resulting output to the Data Manager and Layer Manager, and it displays the output in the Image window.

Spectral Hotspot Analysis

Use this option to run hotspot analysis using a spectral index image.

Data should be in units of reflectance prior to creating spectral indices. Most UAV and airborne sensors used for precision agriculture measure reflectance in visible and near-infrared wavelengths, so no further calibration should be required. Some spectral indices (for example, Anthocyanin Reflectance Index 1 and 2) require the data to be in units of reflectance, ranging from 0 to 100 percent (scaled from 0 to 1). Use the ENVI Spectral Math tool if necessary to scale the reflectance data to this range. See the Spectral Indices topic in ENVI Help for more information.

  1. From the Toolbox, select Crop Science > Spectral Hotspot Analysis.
  2. Select an input multispectral raster. See Before You Begin for considerations in selecting an input image.
  3. From the Spectral Index drop-down list, select a spectral index to compute on the input raster. The spectral index image will be analyzed for hotspots. The list includes only those indices that can be computed on the input raster.
  4. Select a Search Distance, which is the value used in the Getis-Ord calculation to determine how pixels are clustered. If the input raster has a valid spatial reference, then distance is measured in meters. Otherwise, it is measured in pixels. See the Examples section below for a comparison of different distance values. The input distance determines the spatial resolution over which the clustering occurs.
  5. Click the Browse button next to Output Raster, and select an output filename and directory.
  6. Select the Display result option to display the output image when processing is complete.
  7. Enable the Preview check box to see a preview of the settings before you click OK to process the data. The preview is calculated only on the area in the Image window and uses the resolution level at which you are viewing the image.
  8. Click OK. ENVI adds the resulting output to the Data Manager and Layer Manager, and it displays the output in the Image window.

 

Interpret the Results


A color slice is applied to the image to create color gradients for specific Z-scores. In the Layer Manager, these gradients are labeled with plus and minus symbols to represent higher and lower Z-scores, respectively. The resulting hotspot image shows statistically significant positive values in green and negative values in red. The following image shows an example:

 

When using a broadband greenness vegetation index, red is less vigorous and green is healthier. However, no matter how healthy or vigorous the vegetation in an area of interest, you will always see hot and cold spots in the results. This is because you will always have values that are above and below the mean.

You may find regions that appear unhealthy but are green in the hotspot result. This is because small local variations do not qualify as hot or cool in the results when the search distance is set to larger values.

Examples


This section provides some different examples of using hotspot analysis, each with a different Search Distance value. The choice of value depends on what you are using hotspot analysis for. Enter a smaller value (for example, the pixel size of the input image) to evaluate crop health on a smaller scale such as individual rows of crops. Or, enter a larger value to assess the general crop health across an entire field.

The following example shows an agricultural field extracted from a multispectral image at 1-meter resolution (courtesy of the National Ecological Observatory Network).

An Enhanced Vegetation Index (EVI) image was created from the multispectral image, followed by hotspot analysis on the EVI image using two different Search Distance values. This example illustrates how you can assess crop health over detailed areas, or over a larger extent.

 

The next example shows an agricultural field extracted from a hyperspectral image at 1-meter resolution (courtesy of Rochester Institute of Technology).

If you have access to high-resolution hyperspectral imagery, you can compute narrowband indices that indicate vegetation stress. Then run Hotspot Analysis on the spectral index images. These examples show which areas contain less water, have lower nitrogen in their leaves, and have lower levels of chlorophyll. They use a Search Distance value of 4 meters.

References


Getis, A., and J. K. Ord. "The Analysis of Spatial Association by Use of Distance Statistics." Geographic Analysis 24, no. 3 (1992): 189-206.

Mitchell, A. The ESRI Guide to GIS Analysis, Volume 2: Spatial Measurements and Statistics. Esri Press, 2005.

Ord, J. K., and A. Getis. "Local Spatial Autocorrelation Statistics: Distributional Issues and an Application." Geographical Analysis 27, no. 4 (1995): 286-306.



© 2017 Exelis Visual Information Solutions, Inc. |  Legal
My Account    |    Buy    |    Contact Us