Tumbesian dry forest in southern Ecuador and northern Peru faces a number of threats from factors such as climate change, wildfire, and intensive livestock grazing that can result in degraded conditions. Large-area monitoring is needed to support national and international watershed projects seeking to preserve the forest and thereby maintain a fresh water supply, plant and animal diversity, and other resource values. The work is part of a Fulbright Scholar’s project entitled “Advancing integrated geospatial data and technologies for environmental management, monitoring, and decision making in southern Ecuador” at the Universidad Técnica Particular de Loja (UTPL). Students and professionals were trained to use cost-free satellite image processing tools and satellite data for environmental mapping and monitoring. In addition, the team assessed machine-learning applications for developing satellite-based species diversity metrics in dry topical forest.
As a proof-of-concept experiment, scientists used total station tree coordinates (d.b.h. ≥ 5 cm) from a 9-hectare permanent plot located on the Arenillas Ecological Reserve (REA) in southwest Ecuador. They simulated random 0.10-ha model training (n = 50) and validation (n = 60) plots to extract tree diversity metrics (including inverse Simpson’s diversity metric) and combined them with multidate Sentinel-1 polarization (e.g., VH and VV) and Sentinel-2 spectral bands, indices, and canopy metrics. Simulated training data were used to develop random forest (Rf) and boosted regression (Xgb) tree models that were validated from test sample data. Initial results indicated that higher spatial resolution Sentinel-2 blue and red bands (10 m) were important for model performance as well as for predicted canopy metrics such as FCOVER (fractional cover) or CW (canopy water content) that also likely depend on 20-m shortwave infrared spectral regions. Study results will help establish a set of satellite-based forest mapping products and monitoring approaches that can be used together with a broad network of permanent forest plots maintained by the UTPL Biology Department for effective management of these important resources.
Dry forest study site located inside the Arenillas Ecological Reserve (REA) in southwest Ecuador (left). REA forest plot (9 ha) and inverse Simpson’s tree species diversity model validation from a, b) random forest (Rf) regression tree estimates and predicted diversity grids summarized within additional 0.20 ha equidistant validation plots and c, d) xgboost (Xgb) regression tree estimates and predicted diversity grids summarized within 0.20 ha simulated validation plots (right).