From Data to Insight: Navigating the Ecological Landscape Through Analysis

“In the forest of facts, we seek the trail of truth. Our models are maps, our code the compass.”

---

Over the past few weeks, our research team has worked through a series of ecological data workflows—exploring PCA ordination, NDVI symbology, spatial clipping, and ordinal regression modeling. These weren’t just technical exercises. Each step was shaped by a deeper question: how can data guide better decisions for species, ecosystems, and landscapes under threat?

---

🔍 Grounded in Ecological Meaning

Analysis in conservation must be biologically literate. When modeling elephant habitat, for example, we asked:

• How far is too far from water?
• What elevation zones correlate with dung presence?
• Which topographic features matter most?

We didn't treat PCA loadings or regression outputs as abstract patterns. Every plot and table was scrutinized for its ecological plausibility.

---

📊 Multivariate Analysis with Clarity

We used PCA to reduce dimensionality in habitat variables and visualize ecological gradients. Each site was labeled and grouped by dung presence.

fviz_pca_biplot(pca_result,
                geom.ind = "point",
                col.ind = df$feces,
                addEllipses = TRUE,
                label = "var") +
  geom_text_repel(aes(label = df$location), size = 3.5)

To increase interpretability, we annotated each plot with relevant axes labels (e.g., "Jarak Ke Air"), adjusted margins, and grouped by meaningful site categories.

---

🌿 Raster Processing & NDVI in QGIS

Satellite data preprocessing included band stacking, NDVI computation, and masking with AOI boundaries. While Python scripts could automate this, QGIS offered greater transparency during exploration.

Typical workflow:

• Stack B04 and B08 from Sentinel or Landsat.
• Calculate NDVI using Raster Calculator:

(B08 - B04) / (B08 + B04)

• Apply AOI mask with Raster > Extraction > Clip Raster by Mask Layer.
• Style using custom NDVI symbology (.qml) for reproducibility.

---

📈 Ordinal Regression for Ranked Habitat Suitability

We modeled multi-class outcomes (e.g. low–medium–high suitability) using ordinal logistic regression. Category thresholds were annotated to interpret transitions.

library(MASS)
model <- polr(suitability ~ distance_water + elevation + cover, data = df, 
  Hess = TRUE)
summary(model)

To explore how prediction categories change across a variable:

library(effects)
plot(Effect("distance_water", model), main = "Effect of Water Distance")

And for a quick confusion matrix comparing predicted vs actual:

table(Predicted = predict(model), Actual = df$suitability)

These results were visualized with faceted bar plots and transition point annotations to communicate uncertainty and overlap between classes.

---

🧬 From Workflow to Reusable System

Our approach isn’t just about one paper or project. We’re assembling a modular pipeline for ecological modeling—components that can be reused across studies:

• Reproducible PCA biplots with ggplot2 and factoextra.
• Ordinal regression diagnostics and visualizations.
• NDVI pipelines in both R and QGIS.
• Spatial masking, stacking, and standard styling for rapid visual output.

These tools aren't abstract—they're tuned for field-ready, stakeholder-relevant conservation decisions.

---

🌱 What’s Next?

We're continuing to translate ecological questions into actionable insights. Upcoming Notivra posts will explore:

• Annotating category transition points in ordinal models.
• Automating spatial processing with R (sf, terra, stars).
• Building multilingual dashboards for communicating results to communities and policymakers.

---

At Notivra, we turn messy ecological data into meaningful decisions—without losing sight of the forest for the stats.

---

📝 Written by the Notivra Team. Need help building your conservation analysis workflow? Reach out at https://notivra.com.