Carson Stam and Dr. Ryan Jensen, Geography

As the principal habitat of the human race, urban areas require a high degree of attention and research. The United Nations has estimated that slightly over 50% of the world’s population now lives in urban areas. This number is projected to increase steadily in the coming decades. The ability to understand our habitat as a species is key to understanding our future. An important component of any urban area is the urban forest. McPherson and Luttinger (1996) stated that “[r]esearch is revealing that with proper management and care, urban forests can contribute to the economic vitality and the quality of life in cities” (p. 53). This research sought to further explain this relationship using a regression analysis between urban forest health and single family homes’ value using remote sensing technology. A secondary goal was to determine whether vegetation indices derived from hyperspectral data are better for estimating home value than indices derived from multispectral data. The hypothesis used in this research was that home values had a strong, positive correlation with vegetation health.

“Remote sensing is the art and science of obtaining information about [the Earth] without being in direct physical contact with [the Earth]” (Jensen, 2000, p. xiii). This is done using Earth orbiting satellites or airplanes equipped with sensors that can measure reflected electromagnetic energy from the surface of the earth. These measurements are useful indicators of plant health because of known interactions between plants and electromagnetic energy during photosynthesis. Photosynthesis determines how much energy from the sun is absorbed and reflected in different portions of the electromagnetic spectrum. Chlorophyll a and b absorb red and blue light and absorb a relatively smaller amount of green light. Because of this relatively smaller amount of green light absorbed, healthy green vegetation appears green to our eyes (Jensen, 2005, p. 304). This is the most obvious interaction between vegetation and radiant flux because it occurs in the visible portion of the electromagnetic spectrum. As plants senesce or come under environmental stress, chlorophyll production decreases. The decrease in the amount of chlorophyll present in a plant causes much higher reflectance of red light and higher than normal absorption of near-infrared energy. This causes plants to appear yellow or red. Because the amount of energy reflected and absorbed in specific portions of the electromagnetic spectrum is known, equations can be calculated which quantify vegetation health using remotely sensed measurements. These equations, known as vegetation indices, allow values to be compared through space and time.

For this research, 98 individual homes across Provo, UT, were randomly selected for study. Home property values were collected from “Zillow Real Estate” online. Two different remotely sensed image datasets were obtained. The first was a multispectral (coarse spectral resolution) dataset obtained from the National Agriculture Imagery Program. The second image dataset was a hyperspectral (fine spectral resolution) dataset collected by the AISA+ sensor. Both data sets were collected during “leaf-on” season. Polygons were created for each home’s area of vegetation using “head’s-up” digitization. Roofs, driveways, sidewalks, decks, and any other
non-vegetation land covers were not included in the polygon to protect the integrity of the vegetation index values computed for each home. Reflectance values were extracted for each home’s polygon from both the multispectral and hyperspectral image datasets. For each home, one vegetation index was calculated from the multispectral imagery and seven vegetation indices from the hyperspectral imagery. These values were then compared to their corresponding property values using a regression analysis.

The results of the regression analysis showed that there was a significant relationship between home value and the vegetation index calculated from the multispectral imagery. The R2 value produced was 0.056 at a significance level of 0.02, meaning that at the α = .05 level this relationship was statistically significant. The vegetation index that was calculated from the multispectral imagery is known as the Normalized Difference Vegetation Index (NDVI). NDVI looks at the ratio of red-light reflectance (healthy plants reflect little at this wavelength) and near-infrared reflectance (healthy plants reflect most of the energy at this wavelength).

None of the previously defined vegetation indices that were computed from the hyperspectral imagery produced significant relationships with home value. However, I was able to create an equation from two of the wavelengths in the hyperspectral imagery that did produce significant results. When the bandwidths of 666 nanometers (nm) and 696 nm were algebraically compared with a simple ratio (reflectance at 666 nm / reflectance at 696 nm) a significant relationship was created using a quadratic regression model. The R2 value produced was 0.093 at a significance level of 0.01, meaning that at the α = .05 level this relationship was statistically significant.

While these R2 values seem small, we need to consider them in context. Variation in vegetation indices account for a minimum of 5.6% (NDVI calculated from multispectral imagery) to a maximum of 9.3% (my own vegetation index calculated from hyperspectral imagery) of the variation in home value in Provo, Utah. These are large percentages considering all other factors which affect a home’s value that were not included in the regression models such as number of bedrooms, number of bathrooms, square footage, and year built. The percentage of variation that vegetation indices explain when compared to these ordinary explanatory variables is small; however, even small percentages of the variation in a $200,000 home is something that homeowners should consider when wanting to sell their home.

The conclusion of this research is that my hypothesis was correct. This is shown by the significance levels tested at the α = .05 level. This paper does not suggest using vegetation indices as a stand-alone predictor of home value; however, vegetation indices can be useful when used in conjunction with other variables (i.e. square footage) to model home value. The secondary goal, determining whether hyperspectral data would be better than multispectral data for modeling home value, was also achieved. It appears that vegetation indices can be derived from hyperspectral data that account for 3.7% more variation than indices derived from multispectral data.
As the principal habitat of the human race, urban areas require a high degree of attention and research. The United Nations has estimated that slightly over 50% of the world’s population now lives in urban areas. This number is projected to increase steadily in the coming decades. The ability to understand our habitat as a species is key to understanding our future. An important component of any urban area is the urban forest. McPherson and Luttinger (1996) stated that “[r]esearch is revealing that with proper management and care, urban forests can contribute to the economic vitality and the quality of life in cities” (p. 53). This research sought to further explain this relationship using a regression analysis between urban forest health and single family homes’ value using remote sensing technology. A secondary goal was to determine whether vegetation indices derived from hyperspectral data are better for estimating home value than indices derived from multispectral data. The hypothesis used in this research was that home values had a strong, positive correlation with vegetation health.

“Remote sensing is the art and science of obtaining information about [the Earth] without being in direct physical contact with [the Earth]” (Jensen, 2000, p. xiii). This is done using Earth orbiting satellites or airplanes equipped with sensors that can measure reflected electromagnetic energy from the surface of the earth. These measurements are useful indicators of plant health because of known interactions between plants and electromagnetic energy during photosynthesis. Photosynthesis determines how much energy from the sun is absorbed and reflected in different portions of the electromagnetic spectrum. Chlorophyll a and b absorb red and blue light and absorb a relatively smaller amount of green light. Because of this relatively smaller amount of green light absorbed, healthy green vegetation appears green to our eyes (Jensen, 2005, p. 304). This is the most obvious interaction between vegetation and radiant flux because it occurs in the visible portion of the electromagnetic spectrum. As plants senesce or come under environmental stress, chlorophyll production decreases. The decrease in the amount of chlorophyll present in a plant causes much higher reflectance of red light and higher than normal absorption of near-infrared energy. This causes plants to appear yellow or red. Because the amount of energy reflected and absorbed in specific portions of the electromagnetic spectrum is known, equations can be calculated which quantify vegetation health using remotely sensed measurements. These equations, known as vegetation indices, allow values to be compared through space and time.

For this research, 98 individual homes across Provo, UT, were randomly selected for study. Home property values were collected from “Zillow Real Estate” online. Two different remotely sensed image datasets were obtained. The first was a multispectral (coarse spectral resolution) dataset obtained from the National Agriculture Imagery Program. The second image dataset was a hyperspectral (fine spectral resolution) dataset collected by the AISA+ sensor. Both data sets were collected during “leaf-on” season. Polygons were created for each home’s area of vegetation using “head’s-up” digitization. Roofs, driveways, sidewalks, decks, and any other non-vegetation land covers were not included in the polygon to protect the integrity of the vegetation index values computed for each home. Reflectance values were extracted for each home’s polygon from both the multispectral and hyperspectral image datasets. For each home, one vegetation index was calculated from the multispectral imagery and seven vegetation indices from the hyperspectral imagery. These values were then compared to their corresponding property values using a regression analysis.

The results of the regression analysis showed that there was a significant relationship between home value and the vegetation index calculated from the multispectral imagery. The R2 value produced was 0.056 at a significance level of 0.02, meaning that at the α = .05 level this relationship was statistically significant. The vegetation index that was calculated from the multispectral imagery is known as the Normalized Difference Vegetation Index (NDVI). NDVI looks at the ratio of red-light reflectance (healthy plants reflect little at this wavelength) and near-infrared reflectance (healthy plants reflect most of the energy at this wavelength).

None of the previously defined vegetation indices that were computed from the hyperspectral imagery produced significant relationships with home value. However, I was able to create an equation from two of the wavelengths in the hyperspectral imagery that did produce significant results. When the bandwidths of 666 nanometers (nm) and 696 nm were algebraically compared with a simple ratio (reflectance at 666 nm / reflectance at 696 nm) a significant relationship was created using a quadratic regression model. The R2 value produced was 0.093 at a significance level of 0.01, meaning that at the α = .05 level this relationship was statistically significant.

While these R2 values seem small, we need to consider them in context. Variation in vegetation indices account for a minimum of 5.6% (NDVI calculated from multispectral imagery) to a maximum of 9.3% (my own vegetation index calculated from hyperspectral imagery) of the variation in home value in Provo, Utah. These are large percentages considering all other factors which affect a home’s value that were not included in the regression models such as number of bedrooms, number of bathrooms, square footage, and year built. The percentage of variation that vegetation indices explain when compared to these ordinary explanatory variables is small; however, even small percentages of the variation in a $200,000 home is something that homeowners should consider when wanting to sell their home.

The conclusion of this research is that my hypothesis was correct. This is shown by the significance levels tested at the α = .05 level. This paper does not suggest using vegetation indices as a stand-alone predictor of home value; however, vegetation indices can be useful when used in conjunction with other variables (i.e. square footage) to model home value. The secondary goal, determining whether hyperspectral data would be better than multispectral data for modeling home value, was also achieved. It appears that vegetation indices can be derived from hyperspectral data that account for 3.7% more variation than indices derived from multispectral data.

References

• Jensen, J. R. (2000). Remote sensing of the environment: An earth resources perspective. Prentice- Hall, Upper Saddle River, New Jersey, USA.
• Jensen, J.R. (2005). Introductory digital image processing: A remote sensing perspective. Prentice- Hall, Upper Saddle River, New Jersey, USA.
• McPherson, E.G. & Luttinger, N.S. (1996). The critical role of urban forest research. Western Arborist, 22(4), 53