Hexagon Geospatial
MENU

Spatial Modeler Tutorials

Learn more about our products, find answers, get the latest updates, and connect with other Hexagon Geospatial product users, or get support from our professional service team.
Showing results for 
Search instead for 
Do you mean 

Biophysical Composition Index (BCI) for mapping impervious surface and other landcover classes

by Technical Evangelist ‎02-15-2019 11:44 AM - edited ‎02-18-2019 07:16 AM (261 Views)

Download models

 

Description:

 

In the paper "BCI: A biophysical composition index for remote sensing of urban environments" (C DengC Wu - Remote Sensing of Environment, 2012 - Elsevier), the authors propose a Biosphysical Composition Index (BCI) following the mechanism of Ridd's conceptual Vegetation – Impervious – Soil triangle model (Ridd, 1995) in order to effectively represent major biophysical compositions in an urban environment. With the BCI, impervious surfaces are expected to have positive and relatively high values; vegetation is expected to be differentiated from other landcovers through its negative and low values; and bare soil is expected to have a value of near zero, and can be separable from impervious surfaces.

 

The left-hand 2D View shows a Landsat 8 image, the right-hand View shows the results of the BCI analysis
BCI_Results.PNG

 

The major objective of developing the BCI index is to derive a simple and convenient spectral enhancement approach that can highlight the contrasts among three major urban biophysical compositions, namely vegetation, impervious surfaces, and soil. Deng and Wu's analysis of results suggests that BCI has a significant and positive correlation with urban imperviousness, and a significant but negative association with vegetation abundance at various resolutions. More importantly, BCI showed promise in discriminating bare soil from impervious surfaces.

 

Deng and Wu's analysis of a "Landsat ETM+ BCI image indicates that:

 

  • Bright impervious surfaces, including concrete roads and bright roofs (e.g. glass, metal, and plastic), have the highest and positive values, and are characterized as a white tone. 
  • Dark impervious surfaces, including asphalt roads, parking lots, and dark roofs, have the second highest and positive values, and are indicated as a light gray tone. 
  • Further, soils and mixed land covers (including mixtures among vegetation and soil, vegetation and imperviousness, and all three major components) have a BCI value near to zero, and are shown with a tone of medium gray. 
  • Dark and bright vegetation, e.g. trees, senescent grasses and developing green grasses, etc., has the lowest and negative values (less than around −0.1), and is displayed as dark gray and black."

Deng and Wu's BCI algorithm is based on a normalised Tasseled Cap transformation and subsequent index, which has been implemented in the Spatial Model provided here.

 

BCI_v16_5_1.gmdx
BCI_Model.PNG

 

Data Requirements and Assumptions

 

Input imagery should be corrected to reflectance (TOA or BOA) and be from a sensor supported by the Tasseled Cap transformation. TOA correction can be performed using Spatial Models such as the one for Landsat 8 I bundled with the models download on this page (l8_toa_reflectance_v16_5_0.gmdx)

 

If the input image does not already have NoData defined, there is an option to set zeros to NoData. But if you've corrected to Reflectance, hopefully you've already also identified NoData.

 

Deng and Wu's paper suggests masking out water prior to executing the analysis. Doing so largely depends on whether you consider water bodies (lakes, streams, rivers, etc) as being impervious surfaces (which you can strongly argue they are). They certainly generally fall in the "low albedo, impervious surface" portion of the BCI continuum.

 

I'd also recommend masking out clouds. Otherwise they will be identified at the high end of the BCI (high albedo, impervious surface). Snow and ice may also be confused as bright impervious surfaces.

 

The model has been set up so that all three output filenames are optional. Obviously, if you dont' specify any of the three, the model will not produce any outputs! If you want to produce the Clustered BCI product you must also specify a filename for the BCI image. If you wish to produce just the Thresholded BCI (with no BCI or Clustered BCI) you can. 

 

The Spatial Model enables the BCI to be converted from a continuous index result to a classified image in two possible ways: The (optional) Thresholded BCI requires user-specified thresholds to divide the BCI values into the four categories mentioned above; The Clustered BCI uses the K-Means unsupervised machine learning algorithm to divide the BCI into the four categories. The advantage of Thresholded BCI is that Preview can be used to set the desired threshold values in real-time. The advantage of Clustered BCI is that it does not require additional user inputs. 

 

Input parameters:

 

Raster Input: Name of the input multispectral image (which has been corrected to reflectance)

Sensor: Name of the sensor which captured the Raster Input. This must be a sensor for which Tasseled Cap coefficients exist.

Stats Skip Factor: Specify a skip factor to be used by the Spatial Model when calculating the ranges of the Tasseled Cap layers. A larger value will speed up the calculation at the expense of possible inaccuracy in the ranges over which the data is normalised. A value of 1 ensures correct statistics but at the cost of calculation time.

Set Input Zeros to NoData?: A boolean yes/no value. If the Raster Input does not have NoData already specified this option can be set to True and locations where DN values are 0 will be considered NoData locations.

BCI: Name of the output file to store the continuous Biophysical Composition Index (BCI) values. The field can be left blank, in which case the BCI is not written to disk.

Thresholded BCI: Name of the output file to store class values derived from the BCI values (by using the threshold values specified belwo). The field can be left blank, in which case the Thresholded BCI is not calculated and not written to disk.

Vegetation Threshold: BCI values less than this value will be considered to be part of the "Dark and bright vegetation" class

Mixed Threshold: BCI values less than this value (but larger than the Vegetation Threshold) will be considered to be part of the "Soils and Mixed Land Covers" class

Dark Impervious Trheshold: BCI values less than this value (but larger than the Mixed Threshold) will be considered to be part of the "Dark impervious surfaces" class. Values larger than this value will be considered "Bright impervious surfaces".

Clustered BCI:  Name of the output file to store class values derived from the BCI values (by using a K-Means clustering algorithm). The field can be left blank, in which case the Clustered BCI is not calculated and not written to disk. If specified, the BCI filename should also be specified.

 

 

BCI_Run.png

   

 

 

Courses
Contributors