The Limits of Rain: Why Precipitation Alone Fails to Explain Shrubland Encroachment Course: Universi

The Limits of Rain: Why Precipitation Alone Fails to Explain Shrubland Encroachment
Course: University Scholars Program
Instructor: Dr. Jane Southworth | Spring 2025
Introduction
Namibia’s expansive savannas are undergoing a thorny transition, with invasive brush thickets now covering approximately 45 million hectares of the dryland African country (Birch et al., 2016). The Otjozondjupa region is among the most severely affected, where shrub densities in some areas exceed 25,000 individual bushes per hectare (SAIEA, 2015). This transformation threatens the region’s rangelands, diminishing livestock carrying capacity and reducing the economic viability of pastoral livelihoods (de Klerk, 2004). Bush encroachment is also linked to biodiversity loss and increased evapotranspiration, further exacerbating local water stress and environmental degradation (Birch et al., 2016). Given the substantial ecological and economic harm that encroachment causes, targeted bush management strategies to trim back the thicket are imperative. However, precise indicators factors driving shrubland expansion remain elusive.
This study aims to critically evaluate the role of precipitation in shrubland expansion by analyzing land cover changes in Otjozondjupa from 2014 to 2024 using Random Forest classification of Landsat 8 imagery and CHIRPS precipitation data. By examining spatial and temporal correlations between precipitation patterns and shrubland dynamics, this research assesses whether annual, wet-season, and dry-season precipitation trends can reliably predict shrub encroachment. Identifying the limitations of precipitation as a sole explanatory factor will refine bush management strategies and potentially suggest other variables driving shrubland persistence and expansion. Understanding these dynamics is crucial for developing targeted, evidence-based solutions that optimize rangeland productivity while mitigating the adverse effects of unchecked bush encroachment.

Methods
Data Collection and Preprocessing
This study employed a supervised classification approach using the Random Forest (RF) algorithm to assess shrubland area growth and decline throughout the Otjozondjupa region from 2014 to 2024. Surface reflectance imagery sourced from the Landsat 8 collection in Google Earth Engine (GEE) was processed to generate annual median composites which serve as the base for landcover analysis (USGS, 2025). The ESA WorldCover 10m dataset in GEE served as the primary reference for the study area’s land cover classification (Zanaga et al., 2022). This dataset classifies all of Earth’s landcover under 10 different major classes, notably the study area is only comprised of 4 types (shrubland, grassland, built-up areas, and bare/sparse vegetation).
The RF classification scheme trains itself off labeled data, identifies key shared characteristics between data types, and then produces novel landcover classifications based on its training data. To train the RF classifier, 5,000 randomly sampled points were extracted from the ESA WorldCover 2021 dataset, with an equal number of samples allocated to each class to prevent model bias. Once trained on the labeled 2021 Otjozondjupa landcover type data, the classifier was applied retrospectively to classify land cover for each year from 2014 to 2024, ensuring consistency in classification across the study period.

Precipitation Data Analysis
To explore the relationship between precipitation and shrubland expansion/decline, precipitation data from GEE’s CHIRPS Daily dataset gathered for each study year were aggregated to derive mean annual rainfall (Funk et al., 2015). This process created yearly average precipitation map layers depicting the distribution of rainfall (in milliliters) across the study area. By overlaying the CHIRPS precipitation layers with the classified landcover data, rainfall averages for each area can be identified. The mean precipitation for each year within a two-year period was computed by totaling the annual precipitation for each of the land categories and dividing it by the region’s pixel count. These yr1 and yr2 values were then averaged to represent the typical precipitation conditions associated with each category of year-to-year shrubland change.

Long-Term Analysis of Shrubland Stability and Dynamics
To assess decade-long trends in shrubland persistence and transitions, the classified landcover layers were masked to only include areas labeled as shrubland. These annual shrubland extents were then composited by stacking the shrubland layers into a multi-band image, where each band corresponded to a specific year’s shrubland area between 2014 and 2024. Using this composite dataset, pixels were categorized into three groups. The first group, stable shrubland, represents pixels that were labeled as shrubland across all years. The stable non-shrubland group included all pixels never categorized as shrubland in the study period. The final category, dynamic shrubland, consists of all remaining pixels and identifies areas which fluctuated between shrubland and other land cover types (primarily grassland) from 2014-2024.

Results
The classification of shrubland areas across Otjozondjupa from 2014 to 2024 revealed distinct regions of stability and transition. Stable shrubland pixels were predominantly concentrated in the northern and southern parts of the study area, with scattered stable shrublands across the east (Figure 1). Conversely, the dynamic shrubland region spanned the entire study area, forming a clear transitionary zone separating the permanently bush-encroached territory. Stable non-shrubland areas represented a minuscule amount of the study area, mainly composed of built-up regions (roads, towns) and bare rock.

Figure 1. Land cover trends in Otjozondjupa from 2014 to 2024, highlighting stable shrubland, stable non-shrubland, and areas with temporary shrubland cover.

Precipitation Patterns and Long-Term Shrubland Stability
A two-factor ANOVA was conducted to determine whether stable and dynamic shrubland regions experienced significantly different annual precipitation patterns. Results indicated no significant difference (p = 0.9188), suggesting that both stable and dynamic shrubland areas received comparable levels of annual precipitation over the study period. Similarly, when analyzing dry-season precipitation separately, no significant difference was found between stable and dynamic shrubland regions (p = 0.3259), indicating that seasonal drought conditions did not disproportionately affect one category over the other. Furthermore, wet-season precipitation patterns also showed no significant variation between these two categories (p = 0.9653).

Year 1 vs. Year 2 Precipitation Trends on Shrubland Area Change
A parallel regression analysis was conducted to evaluate whether yr2 precipitation in newly shrubbed areas influenced the size of the expansion. This analysis yielded a marginally non-significant result (p = 0.0617, r = 0.6089, R² = 0.3708), suggesting that higher precipitation in year-two may be associated with an increase in newly established shrubland, albeit without reaching strict statistical significance. The parallel between increase in mean precipitation and shrubland expansion is highlighted by increases of shrubland in 2016-2017 and 2019-2020 coinciding with higher rainfall levels. However, the analysis of year-one pre-growth precipitation as a predictor for new shrubland expansion did not reveal any significant effect (p = 0.4396, r = -0.2763, R² = 0.0764).


Figure 2. Composite map created from by calculating each pixel’s average annual precipitation


Discussion
Despite scrutinizing the annual, wet-season, and dry-season precipitation trends across Otjozondjupa, no clear-cut relationship emerged between rainfall variability and shrubland persistence. Regions which frequently transitioned between grassland and shrubland shared near-identical precipitation trends as nearby stable-shrubland areas. These findings reveal that rainfall likely only plays a supporting role in driving shrubland encroachment, rather than being the single dominant predictor of expansion.
Interestingly, while yr1 precipitation trends proved ineffective at forecasting next-year shrubland expansion or decline, a marginal relationship emerged when examining yr2 rainfall. Areas that experienced higher precipitation in yr2 showed weak but suggestive trends toward increased shrubland expansion. Additionally, higher yr2 rainfall appeared to mitigate shrubland loss. Though these correlations are not strictly statistically significant, they suggest that while precipitation alone does not trigger encroachment, it may promote the short-term growth of shrubland species and prompt temporary landcover transitions.
While rainfall may influence short-term shrubland fluctuations, it seems insufficient to fully predict encroachment, as other ecological forces also shape vegetation patterns. Effective bush management strategies should therefore integrate multiple ecological factors rather than relying solely on rainfall-based predictions.
References
Birch, C., Harper-Simmonds, L., Lindeque, P., & Middleton, A. (2016). Benefits of bush control in Namibia. The Economics of Land Degradation. https://www.eld-initiative.org/fileadmin/pdf/ELD-CS_namibia_04_web_300dpi.pdf.
de Klerk, J. N. (2004). Bush encroachment in Namibia: Report on phase 1 of the bush encroachment research, monitoring and Management Project. Ministry of Environment and Tourism.
Funk, C., Peterson, P., Landsfeld, M., Pedreros, D., Verdin, J., Shukla, S., Husak, G., Rowland, J., Harrison, L., Hoell, A., & Michaelsen, J. (2015). The Climate Hazards Infrared Precipitation with Stations—A new environmental record for monitoring extremes. Scientific Data, 2, 150066. https://doi.org/10.1038/sdata.2015.66
SAIEA. (2015). Strategic Environmental Assessment of Large-Scale Bush Thinning and Value Addition Activities in Namibia. https://the-eis.com/elibrary/sites/default/files/downloads/literature/SAIEA%20-%20Bush%20Thinning%20Stragetic%20Environmental%20Assessment%202016.pdf.
University of Arizona . (2024, June 8). Senegalia mellifera. University of Arizona Campus Arboretum. https://apps.cals.arizona.edu/arboretum//taxon.aspx?id=1252#:~:text=are%20present.%20(-,Subsp.,(1%2C%203).%20(
USGS. (2025). USGS Landsat 8 level 2, Collection 2, tier 1. Google Earth Engine Data Catalog. https://developers.google.com/earth-engine/datasets/catalog/LANDSAT_LC08_C02_T1_L2.
Zanaga, D., Van De Kerchove, R., Daems, D., De Keersmaecker, W., Brockmann, C., Kirches, G., Wevers, J., Cartus, O., Santoro, M., Fritz, S., Lesiv, M., Herold, M., Tsendbazar, N. E., Xu, P., Ramoino, F., & Arino, O. (2022). ESA WorldCover 10 m 2021 v200. https://doi.org/10.5281/zenodo.7254221