Associated files and scripts for the manuscript "An open-source image classifier for characterizing recreational activities across landscapes" by Samantha G. Winder, Heera Lee, Bumsuk Seo, Emilia H. Lia, & Spencer A. Wood (In review).
Includes scripts used for the analysis of recreational activities in Flickr photos in the Mt Baker-Snoqualmie National Forest, USA.
Scripts for training, applying, and evaluating the classifier. Scripts 1 and 2 have been abstracted in the recCNNize repository. Author: Bumsuk Seo
1_Middlefork_Training_CNN.py - Build our classifier model based on pre-trained InceptionResNetV2.
2_Middlefork_BatchTagging.py - Applies model to Middle Fork region.
3_1_SamplePhotosForEvaluation.R - Sample tagged photos for manual evaluation.
3_2_Manual_evaluation_MiddleFork_MountainLoop.R - Creates confusion matrices and other evaluation metrics for the image classifier.
Calculating activity photo user-days. Author: Sama Winder
1_addingUserData.R - Adds user id and date taken to predictions from the recCNNize model. Reads in FlickrMiddlefork_predicted.shp and FlickrMBSMtLoop_predicted.shp (the final outputs from the CNN model), joins them to user data, and writes out <file>_predicted_users.geojson.
2_creating_gridded_apuds.R - reads in *predicted_users.geojson and intersects with a grid, then calculates Activity photo user-days (APUD) per grid cell and writes out <area>_activities_by_grid_dddd.geojson
Scripts that work with the survey data, and compare it to the recCNNize predictions. Author: Sama Winder
1_subsettingSurveyDatatoActivities.R - reads in raw survey data, renames and combines some columns to make comparable with CNN outputs. Creates survey_activities_data_20210510.csv, as well as some other files that are divided into sites
2_ComparingSurveytoModelUserDays.R - creates scatterplot and calculates correlation between APUD and survey activities across the Middle Fork (also has some older code that does similar things at a "site" level)
Scripts to relate landscape characteristics to activity diversity. Author: Sama Winder
1_transforming_and_clipping_predictors.R - reads in raw predictor (landscape characterstics) data and standardizes them for use in the preference model by transforming, flattening, validating, and clipping them to the relevant area. Writes out individual geojsons for each predictor.
2_preparing_predictors.R intersects various predictors with the <area>_activities_by_grid_dddd.geojsons, then write them all out in a single Predictors_dddd.geojson which includes the activities numbers as well.\
3_preference_model.R reads in Predictors_dddd.geojson, runs 2 negative binomial models (one per region), and creates a coefficient plot showing them against each other.
Makes maps used in the manuscript. Author: Sama Winder
flickr_map_mf_mtnloop.R - Creates Figure 1, showing the locations of the study areas and all Flickr photos included in the study.
mapping_activity_diversity.R Creates Figure 5 - maps of activity diversity across both regions. Reads in <area>_activities_by_grid_dddd.geojson for both regions, combines, then creates a map of activity diversity.