Riparian Data Analysis using GLMs and Sensitivity Analysis
Description
Periphytic biomass In each of the 30 catchments, a 150-m stream reach was selected for periphytic biomass experiments and partitioned into four sections of 30 meters. To estimate periphytic biomass, referred to as the total biological material residing on submerged surfaces in streams, acetate plates (10 cm x 15 cm) were installed in four pools per stream for approximately 45 days. The periphytic biomass was obtained by the sum of chlorophyll-a concentration, organic matter and inorganic matter. The extract was measured in a spectrophotometer at 750 and at 665 nm, according to Steinman, et al. (2017). Precombusted filters were used to determine biomass through the ash-free dry mass (AFDM). The remaining material after the filter combustion was considered the periphytic inorganic matter. There were a total of 87 samples of periphytic biomass. Our aerial survey included samples of early-stage (n = 29), mid-stage (n = 16), edge-dominated forest classes (n = 30), and old-growth forest sites within a national park, which served as the control (n = 12). LiDAR data collection The aerial survey was conducted with a DJI Matrice 300 RTK Remotely Piloted Aircraft (RPA) equipped with a Zenmuse L1 LiDAR sensor. Data acquisition was performed between 9h and 16h, with the autonomous flight programmed to take place at a speed of 8 m/s at 60 meters above surface level (AGL), utilizing a "terrain follow" flight mode to maintain a consistent distance from the terrain's surface, configured in non-repetitive mode with 2-return. The collected data was exported to DJI Terra, version 3.15.24 (DJI Enterprise, Shenzhen, China), software was used to generate the 3D point cloud. The transects were executed alongside the streamline, producing a high-density point cloud (~100 ± 50 pts/m2). A D-RTK base station provided real-time kinematic (RTK) corrections for precise georeferencing, with a RMSE of 9.6 cm. The output from this procedure was a series of files containing xyzi (easting, northing, elevation, intensity) information for each laser pulse return collected from the sampled riparian ecosystem. The four reach sections and their limits were geo-located using a pair of Spectra Geospatial SP-60 GNSS receivers in RTK mode and used to assess the relationship between periphyton and forest structure. We emphasize that, although we aimed for a 30-m reach section. The fitted generalized linear model for periphytic biomass was analysed to consider the relative importance of each forest structure variable on the model output. The relative importance of each forest structure variable on the model’s output was then conducted using the eFAST analysis. The eFAST sensitivity analysis was conducted using 1,000 sampling points. First-order indices were computed based on variance decomposition, enabling the identification of each parameter’s contribution to output variability.
Files
Steps to reproduce
ReadMe - Riparian Data Analysis using GLMs and Sensitivity Analysis Project Description This project aims to identify and quantify the effects of forest structure on periphyton growth in small streams located in the Atlantic Forest, Brazil. It was developed to explore potential functional indicator thresholds for implementing best management practices in restoring riparian ecosystems. This analysis involves fitting regression (GLM) models to stream data, evaluating multicollinearity, performing cross-validation, and conducting sensitivity analysis using the eFAST method. Implemented in R, the study provides insights into how environmental variables influence stream ecological responses. Usage Instructions Data Access: Data should be stored in an Excel file (data.xlsx) located at D:/Riachos/artigo2/Rsteps/. Required Packages: RStudio with the following packages: readxl, car, sjPlot, DHARMa, caret, sensitivity. Run Script: Load and execute the script in RStudio, either step-by-step or in blocks. Data Structure The dataset contains predictors: h: forest maximum height l: canopy heterogeneity x: leaf area index of understory a: plot size time: a categorical variable indicating sampling times Response variables include: cl: chlorophyll-a ptot: total periphyton biomass pin: inorganic matter por: organic matter The data are stored in .xlsx format. Workflow Overview Methodology Data Loading: Read data from Excel and convert time into a factor. Model Fitting: Adjust multiple GLMs for different response variables using Gamma distribution with a log link. Model Evaluation: Check for multicollinearity with Variance Inflation Factor (VIF), analyze residuals with DHARMa, and perform ANOVA. Cross-Validation: Implement 25-fold cross-validation with caret, generate performance metrics, and save results. Sensitivity Analysis (eFAST): Identify the influence of continuous predictor variables on model outputs using the eFAST method. Experimental Procedure For each specified formula, a GLM is fitted. Evaluation metrics include hypothesis tests, residual diagnostics, and variable importance. Cross-validation results are stored in CSV and RData files. Sensitivity analysis helps quantify the contribution of predictors like h, l, x, and a. Variables Description Predictors: h: Forest maximum height l: Canopy heterogeneity x: Leaf area index of understory a: Plot size time: Sampling time (factor) Response variables: cl: Chlorophyll-a ptot: Total periphyton biomass pin: Inorganic matter por: Organic matter