Vector
- Points, lines, polygons
- Separate objects with attributes
- Useful for occurrences and boundaries
Today’s questions
- How does a raster represent continuous environmental variation?
- How do resolution, extent, alignment, and NoData affect an analysis?
- Which environmental covariates make biological sense for our study?
Types of spatial data
Vector are shapes (boundaries or objects)
Rasters are georeferenced numerical arrays
Raster
- Regular grid of cells
- One value per band per cell
- Useful for climate, elevation, and land cover
Raster anatomy
A cell is an area, not a point
6.4°C
Every location inside one cell is represented by the same stored value.
The value may be observed, interpolated, modeled, classified, or summarized.
Metadata
Five properties define the grid
CRSHow coordinates locate the raster on Earth
ResolutionCell width and height in coordinate units
ExtentThe outer bounds of the raster
AlignmentWhether cell edges and centers coincide
NoDataCells that do not contain a valid measurement
Shape tells us the number of bands, rows, and columns.
Resolution
Finer cells retain more spatial variation
Actual WorldClim BIO1 values, aggregated to illustrate coarser grids
Scale
2.5 arc-minutes is angular resolution
0.04167° × 0.04167°
At Maine’s latitude, a cell is roughly 3 km east–west and 4.6 km north–south.
- Cell width changes with latitude
- Many occurrence points share one climate value
Extent and clipping
Keep the study area, preserve the grid
Global raster8,640 columns × 4,320 rows
→
Maine boundaryVector mask from Maine GeoLibrary
→
Maine rasterAbout 101 columns × 109 rows
Clipping changes extent and shape. It should not change CRS or resolution.
Alignment
Same resolution does not guarantee same cells
- Same CRS
- Same resolution
- Same extent or compatible overlap
- Same cell origin and boundaries
If grids are offset, row 20 / column 30 refers to different ground locations.
NoData
Missing is not zero
0°C
A valid temperature
0 mm
A valid precipitation value
NoData
No valid value for the cell
Think about a few ways we could deal with No Data?
Bioclimatic variables
Derived to represent biological constraints
BIO1Annual mean temperature
BIO5Max temperature of warmest month
BIO6Min temperature of coldest month
BIO12Annual precipitation
BIOCLIM: Refers to the specific metrics (Bio 1 through Bio 19) derived from monthly temperature and rainfall to represent annual trends, seasonality, and extreme environmental limits (e.g., Annual Mean Temperature or Precipitation of the Wettest Quarter).
A climatic extreme may be more limiting than an annual average.
The study area
Climate surfaces clipped to Maine
WorldClim 2.1 BIO1 and BIO12 at 2.5 arc-minute resolution
Read and inspect
Open the raster before analyzing it
bio1 = rxr.open_rasterio(
"bioclim/wc2.1_2.5m_bio_1.tif",
masked=True,
)
print(bio1.shape)
print(bio1.rio.crs)
print(bio1.rio.resolution())
print(bio1.rio.bounds())masked=True converts the file’s NoData code to missing values.
Clip
Use the state polygon as a mask
bio1_maine = bio1.rio.clip(
maine_boundary.geometry,
maine_boundary.crs,
drop=True,
from_disk=True,
)
bio1_maine = bio1_maine.squeeze()from_disk=True avoids loading the entire global raster into memory.`squeeze()` removes the one-element band of the bioclim variable to simplify extraction and plotting.Point extraction
Connect every occurrence to a raster cell
BIO1 = 6.4
Nearest-cell extraction
- Read point longitude and latitude
- Find the nearest cell center
- Copy the cell value to the record
- Raster grid cells have _ONE_ environmental value and sampled points do not interpolate given their position within the cell
Extract
Select values for many points at once
point_x = xr.DataArray(points.geometry.x, dims="record")
point_y = xr.DataArray(points.geometry.y, dims="record")
bio1_values = bio1_maine.sel(
x=point_x,
y=point_y,
method="nearest",
)
points["bio1_mean_temp_c"] = bio1_values.to_numpy()Interpret missing values
Why might a valid point receive NoData?
- Coastline detail differs between raster and boundary
- The nearest cell center falls over ocean
- Coordinate uncertainty places the point near a cell edge
- The source raster has no valid estimate at that location
Inspect the spatial pattern of missing values before deciding what to do.
Selecting predictors based on ecological meaning: a matter for debate
Should one choose predictors from hypotheses? Or just use all available predictors?
Activity
Mean temperature may constrain seasonal activity.
Tolerance
Warmest or coldest extremes may limit survival.
Moisture
Precipitation may relate to breeding habitat and desiccation risk.
Selection of environmental predictors is still a topic of debate. In the prior two papers we read the authors either used all available predictors or removed correlated predictors. What do you think is the best method and how would you evaluate/justify this method?
Correlation of bioclim variables
Some bioclim variables are strongly correlated (encode similar gradients)
| BIO1 | BIO5 | BIO6 | BIO12 | |
|---|---|---|---|---|
| BIO1 | 1.00 | 0.91 | 0.95 | −0.18 |
| BIO5 | 0.91 | 1.00 | 0.76 | −0.09 |
| BIO6 | 0.95 | 0.76 | 1.00 | −0.21 |
| BIO12 | −0.18 | −0.09 | −0.21 | 1.00 |
Illustrative matrix: students will calculate correlations from their extracted table.
Whether correlation among environmental predictors is a 'problem' is very method dependent. It's important to be aware of the constraints of the SDM method you are using, and to account for predictor correlation if necessary.
Time
Occurrence dates and climate normals answer different questions
WorldClim
1970–2000 average conditions
GBIF records
Many dates, often more recent
Extraction describes the site’s historical climatic context. It does not recreate weather on the observation date.