NASA Reference Number: 1995-GlobalCh00404

Faculty Advisor: Dr. Susan L. Ustin

Institution: Department of Land, Air and Water Resources
University of California, Davis
Davis, CA 95616
 

Signatures:

Date: March 12, 1997
 

Academic Progress

In addition to the work described below, I submitted the paper "Geostatistical scaling of canopy water content in a California salt marsh", received it for revision, and forwarded the revised paper back to the editors at Landscape Ecology. I have also worked on a geographic information systems model of vegetation distributions in a nearby salt marsh. This work will be presented in abstract form at the annual Ecological Society of America meeting this summer.
 

Research Progress

The rationale for this project is that changes in climate affect two of the main drivers of earth systems: temperature and precipitation. Because these two drivers are fundamental to almost all ecological processes, it has often been suggested (and it seems obvious) that climate change will influence ecosystems in some way; however the details of that influence are more difficult to specify because there are so many possible changes. This study suggests that instead of looking for the response of detailed processes, maybe we should use an integrated approach. If we are trying to identify landscapes that are influenced by climate change, we should use an integrated measure that does not depend on specific mechanisms, but which might be a common expression of many mechanisms. Once we have identified affected landscapes, then scientists can study those locations in more detail to account for the observed changes.

Landscape structure is potentially such an integrated measure. We know that plant species distributions have changed dramatically in the past because of climate change. It seems reasonable to expect that anthropogenic climate change in are own era will cause similar changes (Dyer, 1995). The early signs of changes in vegetation distribution are likely to be contraction or expansion of patches, movement of boundaries between vegetation zones, changes in overall biomass and cover, i.e. changes in landscape structure. The landscape ecology literature is rich with a number of landscape metrics for quantifying landscape structure (Riitters, et al, 1996; Qi and Wu, 1996; Hunsaker et al, 1994; Turner et al, 1991; Musick and Grover, 1991). Recent reviews have suggested a small number of these metrics which best describe variation in landscape structure (Riitters et al, 1995).

The data I am bringing to bear on this problem are described in Table 1. Most of my efforts over the last year have been in assembling, analyzing, and beginning to compare these datasets. Because I am looking at so many sites in so many different conditions of climate, land use and topography, developing generalized analytical methods and implementing them by computer has been essential. In the course of this work, I have dramatically increased my skills and expertise in a variety of computer software packages, including ARC/INFO, IDL, ENVI, and particularly perl, which has been my basic coding tool. The specific steps and methods used have been outlined in Table 2 and are summarized below.

Table 1. Data Sources

Landscape Structure data – Remotely sensed AVIRIS quicklooks and selected full AVIRIS scenes

Soils data

The landscape structure metrics I have implemented are those recommended by Riitters et al (1995) (perimeter-area ratio, perimeter-area ratio normalized to a square, fractal dimension, contagion), supplemented by a couple of other methods not considered by those authors (scale based on an autocorrelation threshold, angular second moment, inverse difference moment), and by some basic statistics on the number of patches, the number of small patches, and the proportional coverage of each class. Some of these landscape structure metrics are designed for analysis of continuous data (provided by the quicklooks); others for pixels aggregated into classes by some rule. Remote sensing data is advantageous in this case because its underlying data structure is a continuum of gray levels from 0 to 255, but it can be classified according to simple rules into patches corresponding (roughly) to vegetation and soil.

The classification in this case takes advantage of the difference in reflectance between vegetation and soil in the quicklook band (~700 nm). In this band, vegetation is typically darker than soil. I confirmed this pattern by taking five calibrated AVIRIS data cubes from different climate types and vegetation communities in California and classifying pixels into vegetation, soil or water using the Spectral Angle Mapper algorithm. These classifications were qualitatively verified by field work conducted by the CSTARS laboratory at these sites. These classified images were compared to their corresponding quicklooks to derive gray level thresholds which distinguish vegetation from soil, with a small intermediate class. Because these classes have not been rigorously confirmed by field reconnaissance in all the images, they are referred to as Class-V, Class-S and Class-I, so that it is clear they are based only on a simple remote sensing interpretation. An example of a quicklook and its classified product are shown in Figure 1.

These landscape structure data are being compared to the estimated climate at each site. To estimate the climate, I decided to follow the practice of many ecologists in calculating the average, long term, climatic water balance for each site. Vegetation at the regional scale has been shown to be strongly influenced by the amount of water surplus, water deficit and their seasonal timing (Stephenson, 1990; Major, 1977). I have obtained mean monthly temperatures and total precipitation from the Global Historical Climatological Network, available for free over the Internet. This database has over 6000 precipitation stations and over 4000 temperature stations, with a large concentration of stations in North America, each with a historical record of at least 10 years. Using an average of the five closest stations within 100 km of each site, I calculated the potential and actual transpiration at each site using the method of Thornthwaite and Mather (1955), which requires knowledge of only the mean monthly temperature, latitude and available soil water capacity. Using this calculations with the average precipitation, it is possible to use a mass balance technique to calculate the annual water deficits and surpluses, given an estimate of the available soil water capacity

In the past scientists have often assumed a uniform level of available soil water capacity for all sites (Eagleman, 1976; Major, 1977), but today it is possible to get a geographically specific estimate using the Natural Resources and Conservation Service STATSGO database. By overlaying each quicklook polygon over the appropriate soil coverage, I calculated an area-weighted average available water capacity for each site. Soil water capacity has a strong influence on the annual water deficits and surpluses because it determines both the amount of water stored in the soil and the proportion of actual to potential evapotranspiration, making estimations of this parameter an important step to more accurate water balance calculations. A paper describing these methods is in preparation since this method is general to any location in the contiguous Unite States. A representative water balance diagram is shown in Figure 2.

Topographic data (digital elevation models) at the 1:250,000 scale are also available for free over the Internet from the US Geological Service (USGS). These data come as ARC/INFO triangular irregular networks, which are converted to raster (GRID) format, then clipped with the quicklook coverage. The minimum, maximum, and mean elevations, aspects and slopes were determined for each site. Most of the contiguous United States have digital elevation models at this scale. An example of the topographic data available is shown in Figure 3.

Land use data, also from the USGS, are available for in ARC export format. These coverage describe land use in terms human land use (urban, rural, agricultural, abandoned, etc.) and current vegetation cover (deciduous forest, grassland, shrubland, etc.). The land use for each site is summarized by area for each site. These land use data have been commonly used in past landscape structure studies (e.g. Riitters et al, 1995; Hunsaker et al, 1994) An example of the land use data is shown in Figure 4.

An outline of the work completed so far is shown in Table 2. Using these techniques and data sources, I am currently working with a subset of 51 quicklooks from the contiguous United States to debug the processing and to insure that all the data are coordinated. I am also beginning to explore the multivariate statistical properties of the data. A preliminary listing of selected statistics for these scenes is provided in Table 3.

Table 2. Outline of Work Completed from April 1996 - February 1997.
1. Developed classification algorithm.

1. Used 5 calibrated AVIRIS images in California in different climates
2. Developed image based referenced endmembers for soil, water and veg for each image based on average of 30 pixels each.
3. Calculated Spectral Angle between each reference endmember and image pixels for each image.
4. Developed threshold to classify pixels into soil, veg and water.
5. Compared classified endmember maps from full image to quicklooks.
6. Developed gray level thresholds to separate soil from veg. Water couldn’t be consistently distinguished.

1. Projected starting and ending coordinates to Albers State plane projection used by soils, land use and DEM databases.
2. Based on number of lines in each run, known number of pixels and lines per quicklook, and pixel size, used trigonometry to calculate corners of each quicklook.
3. Developed criteria to reject some quicklooks based on inconsistencies in number of lines reported.

1. Wrote programs to calculate adjacency matrices for quicklook and classified quicklooks, then calculate contagion, texture measures.
2. Also calculated autocorrelograms, patch shape and number metrics, fractal dimension, patch perimeter area ratio.
3. Statistics calculated for quicklook and classified image and at patch, class and landscape levels.
4. Validation of the code using pseudo-quicklooks with known spatial relationships.

1. Identified ancillary datasets required: climate, topography, land use.
2. Climate: decided to represent climate based on water balance diagrams.

1. Requires calculation of potential ET, actual ET and estimates of precipitation and soil water capacity.
2. Downloaded monthly mean temperature and total monthly precipitation from the Global Historical Climatological Network.

1. averaged temp and precipitation values for each station (temp and prec stations slightly different).
2. projected coordinates to Albers projection.
3. calculated distance from quicklook center to station.
4. estimated monthly temp and prec at quicklook based on average of five closest stations within 100 km.

3. Downloaded soils maps and database from the STATSGO database to get available water capacity of soil

a. Calculated average soil water capacity for each soil unit by doing weighted averages for layers (weighted by depth of each layer). Calculated weighted average for soil unit based on percentage of each soil component.

b. by state, downloaded soils coverage and overlaid quicklook polygon. Averaged available water capacity based on areal extent of each soil unit.

4. Estimated potential ET and actual ET using Thornthwaite and Mather (1955) method.

1. potential ET is function of mean monthly temp and latitude.
2. latitudinal adjustment based on daylength/360 hours. Estimated daylength using method of Forsythe et al (1995).
3. actual ET is function of potential ET and the soil moisture ratio (prec + storage)/awc, but limited not to exceed potential ET or (prec + storage), whichever is less.

1. Used mass balance approach to calculate soil storage, deficit and surplus on a monthly basis.
1. Wrote program in IDL to plot water balance diagrams for display.

3. Topography: calculate slope and aspect using 1:250,000 DEMs available from USGS

1. Identify and download DEM files overlapping with site. Inventories downloaded.
2. Convert DEM lattices to raster format. Method developed.
3. Clip out area of quicklook and calculate min/max and average slope, aspect and elevation. Method developed.
4. in progress for preliminary (51) quicklook.

4. Land use: summarize landaus data from USGS/EPA land use data.

1. Identify and download land use coverages overlapping site. Inventories downloaded.
2. Convert from export format to ARC/INFO coverage. Method developed.
3. Clip out and average land use and vegetation cover type for quicklook areas. Method developed.
4. in progress for preliminary (51) quicklook.

4. Multivariate analysis of landscape structure metrics alone and with respect to ancillary datasets.

1. Preliminary analysis for 51 quicklooks.
2. Analysis of univariate and bivariate distribution of landscape statistics and ancillary data.

Future Work

Any patterns which I find between landscape structure and climate in the whole dataset will be tested by examining several year sequence of scenes at a few sites to see if the relationship holds predictably over single sites in time. Candidate sites for this test are the places AVIRIS regularly visits: Harvard Forest, MA, Blackhawk Island, WI, and Jasper Ridge, CA.

An important piece of work still to be undertaken is a sensitivity analysis of the methods used to analyze and summarize the data. I plan to revisit my classification method and quantify how small differences in the selected thresholds influence the landscape statistics. Similarly I would like to examine how the actual spatial location of the quicklooks influences the landscape structure. I will examine adjacent quicklooks on continuous runs to see how scene sampling influences the landscape structure. These sensitivity analyses will overlap nicely with planned investigations of how the spatial and spectral resolution of the remote sensing data influence our measurement of landscape structure. For a few sites where calibrated full AVIRIS data cubes are available, I will examine how spectral bandwidth and placement effect landscape structure, and through pixel averaging, how spatial grain affects the statistics.

From this work I expect to write at least two papers within the next year. The first paper will describe the method used here for deriving the climatic water balance over small regions, appropriate to remote sensing scenes. The second paper will summarize the results of my investigation into the relationship between landscape structure and climate and will include a description of the data sets used, the rationale for this approach, the results of the multivariate analyses, and a discussion of applicability of the method. A third paper, based my investigations of the full AVIRIS data cubes and the influence of spatial and spectral resolution on the analysis may also be possible, if novel results are found. These papers will form a major portion of my Ph.D. dissertation, which I will also write within the next year.
 

References