A comparison of moisture between two distinct agricultural settings (Forested vs. Open agriculture) - 2023

This study was a preliminary investigation aimed at exploring the viability of a potential idea for my master’s thesis—specifically, whether agroforestry practices in tropical regions might offer greater resilience to drought than open-field monocultures. By comparing soil moisture dynamics between two distinct agricultural settings in Costa Rica—forested agriculture (Site A) and open-spaced agriculture (Site B)—the study sought to understand how topographic variation and vegetation cover influence moisture retention.

Using Sentinel-2 data, land use and land cover were classified with 97.32% accuracy through a Random Forest model. Moisture levels were quantified using the Normalized Difference Moisture Index (NDMI), and spatial regression models—OLS, GWR, and MGWR—were applied to assess how topographic variables (slope, aspect, elevation, and flow direction) correlate with moisture patterns.

Although NDMI values showed only marginal differences between the two sites, topographic factors were found to interact with moisture levels in complex, spatially variable ways. The MGWR model was the most effective in capturing these nuances by allowing relationships to vary across space and scale.

Key Challenges

  • Persistent cloud cover prevented time-series analysis, limiting insights to a single date.
  • Site selection was not optimal; terrain differences between sites complicated direct comparison.
  • Lack of local knowledge (e.g., crop types, irrigation practices) made NDMI interpretation difficult.
  • The 20m resolution of Sentinel-2 may have been too coarse to capture fine-scale terrain effects.

Despite these limitations, the study was highly valuable in shaping my understanding of the topic, highlighting methodological considerations and identifying areas needing refinement. It underscores the importance of temporal data, careful site selection, and local context when investigating the potential role of agroforestry in climate resilience.

Overview of the general location of the two study areas, highlighting the difference in the disparity forest cover, located in the Alajuela Province, Costa Rica. Made using ArcGIS Pro.

Dislayed output of the random forest model for the supervised classification of our study areas, defined by five distinct landcover types. Made using R.

Map depicting terrain dynamics of elevation, slope, aspect, and flow direction across Site A (Forested Agriculture) and Site B (Open Agriculture). Made using ArcGIS Pro.

Comparison of the Normalized Difference Moisture Index (NDMI) between two agricultural sites: a) Forested Agriculture and b) Open-Spaced Agriculture. Made using ArcGIS Pro.

Spatial Variation of Local Coefficient Estimates from Fixed MGWR Analysis (significant values with p ≤ 0.05 highlighted). Made using ArcGIS Pro.

Dynamics of local coefficient across agricultural sites as analysed by the fixed MGWR model. Direction trends for each site are presented in panels a to d, while their relative magnitudes are illustrated in panels e-h. The coefficient estimates shown in panels e-h were computed to their absolute value for easier comparison, where the average of the predictor estimates are shown between each site locations. Made using R.