###

This repository contains codes to generate results for manuscript submitted to Water Resources Research 
"Assessing the Trustworthiness of Crowdsourced Rainfall Networks: A Reputation System Approach" (2021WR029721)

###

1. Input data

# RSCRN input
0_KTXHOUST_meta.csv: metadata of PWSs
1_KTXHOUST_15min_rainfall.csv: 15 minute PWS rainfall of PWSs
2_HCFCD_15min.csv: 15 minute Harris County Flood Control District (HCFCD) rainfall data
3_HCFCD_meta: metadata of HCFCD rainfall stations

# PWS QC method input (generated from R code)
c1_HI_flags.csv: high influx flags for PWSs in cluster 1
c1_FZ_flags.csv: faulty zero flags for PWSs cluster 1
c1_SO_flags.csv: station outlier flags for PWSs in cluster 1
c2_HI_flags.csv: high influx flags for PWSs in cluster 2 
c2_FZ_flags.csv: faulty zero flags for PWSs in cluster 2
c2_SO_flags.csv: station outlier flags for PWSs in cluster 2


2. Code description

Python codes were written in Python 3.x version. Please follow the order because each step uses the output from previous steps.
This code will generate results for cluster 1. To generate results for cluster 2, change every "c1" to "c2" and "445" to "560".

0_cluster.py:

Input: PWS metadata (0_KTXHOUST_meta.csv) and PWS 15 min rainfall (1_KTXHOUST_15min_rainfall.csv)
Output: Clustered sub-dataset (ID_list_c1.csv)

1_consensus_and_score.py:
Input: PWS 15 min rainfall (1_KTXHOUST_15min_rainfall.csv)
Input 2: Clustered subdataset (ID_list_c1.csv)

Output: RSCRN trust score, alpha and beta parameters, weight, robust weight
(RSCRN_var_ff_new_string.csv, var = score, alpha, beta, weight, ff = forgetting factor)

2_find_storm_event.py
Input: HCFCD rainfall (2_HCFCD_15min.csv) and metadata (3_HCFCD_meta.csv)
Output: analyzed storm events (df_storm_event.csv)

3_plot_trust_score_evolution.py:
Input 1: storm event (df_storm_event_st.csv) (for cluster 1 and 2, st = 445 and 560 respectively) 
Input 2: Clustered subdataset (ID_list_c1.csv)
Output: trust score evolution for analyzed storm events (trust_score_evolution.png)

4_comparison_with_PWS_QC.py:
'''
Input 1: PWS rainfall (1_KTXHOUST_15min_rainfall.csv), 
         PWS QC results (HI_flag, FZ_flag, SO_flags.csv)
Input 2: Clustered subdataset (ID_list_c1.csv)
Output 1: Overall assessment of RSCRN and PWS QC methods (df_summary.csv)
Output 2: RSCRN trust scores and PWS QC method assessment for each storm event and each PWS (e.g., Figure 4 in the manuscript)

5_RMSE_comparison.py:
Input 1: PWS rainfall (1_KTXHOUST_15min_rainfall.csv), 
Input 2: HCFCD rainfall (2_HCFCD_15min.csv) and metadata (3_HCFCD_meta.csv)
Input 3: PWS QC results (HI_flag, FZ_flag, SO_flags.csv)
Output : RMSE comparison table (df_RMSE_storm_comparison.csv) (Table 5 and 6 in the manuscript)


### PWS QC method ###
This study compares the RSCRN trust scores with an a PWS Quality Control method (de Vos et al., 2019).
The R codes used to generate the results in this study can be found in the folder "PWS_QC_method".
These codes were modified from the original codes to run PWS data for the case study in this paper. 
Original codes can be found from https://github.com/LottedeVos/PWSQC
For details about the PWS QC method please refer to the original manuscript at https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019GL083731