From m@p|no|@10 @end|ng |rom gm@||@com  Fri Apr 17 19:10:08 2026
From: m@p|no|@10 @end|ng |rom gm@||@com (=?UTF-8?Q?Manuel_Sp=C3=ADnola?=)
Date: Fri, 17 Apr 2026 11:10:08 -0600
Subject: [R-sig-Geo] h3sdm: H3 hexagonal grids for polygon-based species
 distribution and landscape modeling in R
Message-ID: <CABkCotSm13WTNsrB4T+2WSb=YG2_HhKHVi8iaFNYVb6cOxCSVw@mail.gmail.com>

Dear list members,

I am pleased to announce that h3sdm is now available on CRAN.

h3sdm takes a different approach from traditional raster-based SDMs:
instead of working at the pixel level, it uses H3 hexagonal grids as the
unit of analysis. Environmental and land cover information is aggregated
within each hexagon, and information theory (IT) metrics are computed to
characterize spatial heterogeneity and landscape composition at the polygon
level.

This polygon-based framework makes the package suitable not only for
species distribution modeling, but also for habitat and landscape analysis
at regional or local scales ? watersheds, protected areas, or biological
corridors ? where pixel-level approaches are often too fine-grained or
noisy for ecological interpretation.

Modeling is built on tidymodels, allowing users to fit any classification
or regression algorithm available in that ecosystem ? from GAMs to random
forests or gradient boosting ? within the same reproducible workflow.

Key features:
- H3 hexagonal grid integration
- Extraction of continuous and categorical environmental variables per
hexagon
- Calculation of information theory metrics to characterize within-polygon
heterogeneity
- tidymodels-based modeling pipeline, open to any supported algorithm

Install from CRAN:

  install.packages("h3sdm")

GitHub: https://github.com/ManuelSpinola/h3sdm

Feedback and contributions are welcome.

Best regards,

Manuel Spinola

-- 
*Manuel Sp?nola, Ph.D.*
Instituto Internacional en Conservaci?n y Manejo de Vida Silvestre
Universidad Nacional
Apartado 1350-3000
Heredia
COSTA RICA
mspinola at una.cr <mspinola at una.ac.cr>
mspinola10 at gmail.com
Tel?fono: (506) 8706 - 4662
Sitio web institucional: ICOMVIS
<http://www.icomvis.una.ac.cr/index.php/manuel>
Sitio web personal: Sitio personal <https://mspinola-sitioweb.netlify.app>
Blog sobre Ciencia de Datos: Blog de Ciencia de Datos
<https://mspinola-ciencia-de-datos.netlify.app>

	[[alternative HTML version deleted]]


From n|ko@@tz|ok@@ @end|ng |rom gm@||@com  Sun Apr 26 13:59:30 2026
From: n|ko@@tz|ok@@ @end|ng |rom gm@||@com (Nikolaos Tziokas)
Date: Sun, 26 Apr 2026 14:59:30 +0300
Subject: [R-sig-Geo] =?utf-8?q?How_to_apply_a_geometric_anisotropic_filte?=
	=?utf-8?q?r_1/cos=C2=B2=28theta=29_to_a_raster=3F?=
Message-ID: <CAGg5H0auRT8Z59Y7v1U9pHbAXdY7dJij0VOkuAGPyMezGRk1rg@mail.gmail.com>

I am downscaling (i.e., increasing the spatial resolution) VIIRS nighttime
lights imagery. The transfer function between the coarse and fine
resolution is approximated by a Gaussian filter whose width

In the cross?track direction (left?right), the sigma scales as

?(?) = ?_0/cos(^2)?

where *?* is the per?pixel viewing angle. The filter should act only
horizontally (along x), pooling all y?values within the column uniformly.
My earlier attempt simply multiplied the raster values by 1/cos(^2)? but I
now realise the operation is a *convolution* with a 1?D Gaussian kernel
whose width depends on the column?s *?*.

Given a value raster x and an angle raster theta with identical dimensions,
what is the correct and efficient way in R (using terra) to apply a 1?D
Gaussian blur on each row, where the kernel?s standard deviation is
determined by theta at the central pixel?s column?

*Current (incomplete) attempt:*

library(terra)
# ---- 1. Create example rasters ----
set.seed(42)
nrows <- 5
ncols <- 200

x <- rast(nrows = nrows, ncols = ncols, vals = runif(nrows * ncols))
theta_vals <- rep(seq(20, 26, length.out = ncols), each = nrows)
theta <- rast(nrows = nrows, ncols = ncols, vals = theta_vals)
# ---- 2. Spatially varying Gaussian blur function ----# sigma0 =
standard deviation of Gaussian at nadir (in pixel units)# theta:
raster of angles in degrees (same dimensions as x)
anisotropic_blur <- function(x, theta, sigma0 = 10) {
  # Convert to matrix (rows = y, cols = x)
  mat_x <- as.matrix(x, wide = TRUE)
  mat_angle <- as.matrix(theta, wide = TRUE)  # same dimensions
  nr <- nrow(mat_x)
  nc <- ncol(mat_x)
  mat_out <- matrix(NA_real_, nr, nc)

  # For each row, apply convolution with column-dependent sigma
  for (r in 1:nr) {
    row_vals <- mat_x[r, ]
    for (c in 1:nc) {
      theta_c <- mat_angle[r, c]               # angle at this pixel (degrees)
      sigma <- sigma0 / (cos(theta_c * pi/180)^2)
      halfwin <- ceiling(3 * sigma)            # kernel half?width
      col_idx <- max(1, c - halfwin) : min(nc, c + halfwin)
      dist <- abs(col_idx - c)
      w <- exp(-0.5 * (dist / sigma)^2)
      w <- w / sum(w)
      mat_out[r, c] <- sum(w * row_vals[col_idx])
    }
  }
  # Return as raster with same properties
  rast(mat_out, crs = crs(x), ext = ext(x))}
# ---- 3. Apply the filter ----
x_blurred <- anisotropic_blur(x, theta, sigma0 = 10)
# ---- 4. Plot side?by?side ----
par(mfrow = c(1, 2))
plot(x, main = "Original predictor")
plot(x_blurred, main = expression("Filtered: Gaussian "*sigma*" =
"*sigma[0]/cos^2*theta))

 I need to actually perform a 1?D Gaussian convolution along rows with a
column?varying (\sigma). What is an idiomatic way to do this in terra?

	[[alternative HTML version deleted]]