The
Effects of Precipitation on NDVI Values in the Caballo Watershed during the
Summer Months
Precipitation
Data: The precipitation
data file format is netCDF. In this case, each file contains daily data for
one month out of a particular year. Gridded
precipitation data exists for 1950-2013 across North America for a total of 756
files (~0.25 gigabytes per file) across the entire time domain. Python, particularly Xarray,
was employed to manipulate the netCDF files. Using Xarray I took
spatial and temporal averages followed by a re-merging of the files into one netCDF file. After
this step, a file of containing the monthly average precipitation (in mm/day)
for each summer month (June, July and August) for 2002 and 2013 exists. Additionally, an average was taken (over the
study area) across three months to yield a JJA average precipitation composite.
NDVI Data: In order obtain NDVI values across the
study region, Landsat 8 images were downloaded from 2013 for June, July, and
August that contained less than 20% cloud cover. Landsat 7 images from 2002 were also
downloaded over the same months with the same query as the Landsat 8
method. Once the images were obtained,
Landsat images were brought into ArcGIS where NDVI was calculated using the
raster calculator by subtracting the Red from the Near Infrared divided by their
sum. This step yields NDVI values from
-1 to +1 across the study area. The
clipping tool isolates NDVI values across the region of interest (Caballo
Watershed).
Projection Used: (WGS 1984 UTM
Zone 13) Used to preserve the area since I was analyzing one watershed and my
study area was not laterally extensive.
Ordinary Least
Squares (OLS): 1,000
random samples were established within the Caballo Watershed using the “generate
random points” tool. Values were
extracted at each of the 1,000 points for both precipitation and NDVI. Then, the NDVI and precipitation attributes
tables were joined. Once joined,
precipitation was the principle explanatory variable using the Ordinary Least
Squares (OLS) method. Moran’s I was also
calculated to determine if the data were spatially autocorrelated.
Here is a list of
some of the Tools used:
|
ArcGIS Tools |
Python Packages |
|
Project |
Xarray |
|
Raster calculator |
Pandas |
|
Netcdf to raster |
Numpy |
|
Clip |
NCO |
|
Extract to mask |
Matplotlib |
|
Value to point |
Seaborn |
|
Create random points |
|
|
Raster to point |
|
|
Kriging |
|
|
OLS |
|
|
Moran’s I |
|
|
Cell Statistics |
|
|
Join |
|