Model outputs and results

These predictions and diagnostics pertain to "Virtual gauges: the surprising potential to reconstruct continuous streamflow from strategic measurements" by Vlah et al. 2023. Check out the zips of input data, figures, and isolated composite discharge series alongside this item. Follow the README files in our code on GitHub (https://github.com/vlahm/neon_q_sim) to reproduce these outputs. <br> Structure of this archive . ├── composite_series/: the primary outcome of this effort; discharge series for each NEON site, spliced from best estimates │ │ ~~CSV STRUCTURE~~ │ │ datetime: date and time, UTC │ discharge_Ls: estimated discharge, liters per second │ discharge_lower95_Ls: 2.5% prediction interval │ discharge_upper95_Ls: 97.5% prediction interval │ source (categorical): source of predictions │ NEON: NEON continuous discharge product DP4.00230.001 │ Linreg_scaled: linear regression on specific discharge (i.e. scaled by watershed area) │ Linreg: linear regression on absolute discharge │ generalist: LSTM broadly trained on CAMELS and MacroSheds data │ specialist: || and finetuned on NEON discharge │ generalist_pgdl: LSTM broadly trained on CAMELS, MacroSheds, and NHM data │ specialist_pgdl: || and finteuned on NHM discharge │ ├── gap_stats.csv: statistics on data gaps, some used in the paper ├── lm_out/: outputs from linear regression on absolute discharge │ ├── results.csv: model scores and specific linear regression methods used for each site │ ├── fit/: misnomer; observations vs. predictions │ │ │ │ ~~CSV STRUCTURE~~ │ │ │ │ site_code: four-letter NEON site ID │ │ datetime: date and time in UTC │ │ Q_neon_field: NEON field-measured discharge observations in L/s; product DP1.20048.001 │ │ Q_predicted: predicted discharge in L/s │ │ : USGS and/or MacroSheds donor gauge discharge in L/s │ │ season: seasonal categorical variable │ │ dummy: before/after dummy for 2016 wildfire at site LECO │ │ │ ├── predictions/: predictions for 2015-2022, where available │ │ │ │ ~~CSV STRUCTURE~~ │ │ │ │ datetime: date and time in UTC │ │ Q_predicted: predicted discharge in L/s │ │ Q_pred_int_2.5: 2.5% prediction interval │ │ Q_pred_int_97.5: 97.5% prediction interval │ │ : USGS and/or MacroSheds donor gauge discharge in L/s │ │ season: seasonal categorical variable │ │ dummy: before/after dummy for 2016 wildfire at site LECO │ │ │ └── summary/: lm summary output │ ├── lm_out_specQ/: same as above, but for specific discharge (i.e. scaled by watershed area) │ ├── results_specificq.csv │ ├── fit/ │ ├── predictions/ │ └── summary/ ├── lstm_out/: outputs from ensembled LSTM models │ ├── param_search_skill.csv: performance of each LSTM trained during the parameter search phase │ ├── results.csv: performance of each ensembled LSTM │ ├── fit/: misnomer; observations vs. predictions │ │ │ │ ~~CSV STRUCTURE~~ │ │ │ │ site_code: four-letter NEON site ID │ │ datetime: date and time in UTC │ │ Q_neon_field: NEON field-measured discharge observations in L/s; product DP1.20048.001 │ │ Q_predicted: predicted discharge in L/s │ │ │ └── predictions/: predictions for 2015-2022 │ │ ~~CSV STRUCTURE~~ │ │ date: date of prediction │ Q_predicted: predicted discharge in L/s │ Q_pred_int_2.5: 2.5% prediction interval of ensemble │ Q_pred_int_97.5: 97.5% prediction interval of ensemble │ Q_neon_release2023_qcpass: NEON continuous discharge without dischargeFinalQF == 1 or dischargeFinalQFSciRvw == 1 │ ├── lstm_runs/: NeuralHydrology run directories, including final models and optimizer states ├── models_used_to_build_composite_series.csv: Table S1 ├── neon_wateryear_assess/: evaluations of each site-wateryear of continuous NEON discharge against field discharge └── score_table.csv: Table 6 <br> <br>