Abstract
Water is a crucial pre-requisite for all human activities. Due to growing demand from population and shrinking supply of potable water, there is an urgent need to use computational methods to manage available water intelligently, especially in developing countries like India where basic data and physical infrastructure are often inadequate. In this context, we present a dataset of water pollution containing quantitative and qualitative data from a combination of modalities: real-time sensors, lab results, and estimates from people using mobile apps. The data, accessible via an API-enabled cloud platform, covers more than 60 locations and consists of both original high-frequency sensor data and reconciled lab results. Furthermore, we discuss an application of this data to understand the spatio-temporal pollution impact of the Ardh Kumbh 2016, a massive religious gathering where hundreds of millions of people converge on the river. This multi-modal approach provides unprecedented detail to help authorities manage ongoing events and plan for future environmental protection.
Introduction
Water management in developing nations faces a significant "data poverty" crisis. While government agencies collect environmental data, it is often fragmented, inaccessible to the public, or lacks the temporal resolution needed to understand short-term, high-impact events. The Ganga River, a lifeline for millions, faces extreme stress during "Kumbh" festivals—gatherings where tens of millions of people perform ritual bathing and offer materials like milk, flowers, and ashes to the water. Standard monitoring stations are often too geographically sparse to capture the hyper-local environmental impact of these events, leaving policymakers and citizens without clear information regarding immediate risks to health and ecology.
Methodology
To address these gaps, a multi-tiered data collection strategy was integrated into a cloud platform called BlueWater:
- Real-Time (RT) Sensing: Using portable multi-sensor kits, the team collected high-frequency data (every second) on parameters including pH, electrical conductivity, dissolved oxygen (DO), and turbidity. These portable units allowed for high spatial coverage along a 76km stretch, reaching areas fixed stations miss.
- Crowdsourcing (Neer Bandhu): A mobile app, Neer Bandhu ("Water Friend"), was launched to allow citizens to submit qualitative data, such as photos of trash, water color, and smell—factors that sensors cannot easily quantify but that are critical for public health awareness.
- Data Integration: The platform reconciles these diverse inputs with traditional lab results from over 60 locations.
- Analysis Techniques: Researchers calculated the Weighted Arithmetic Water Quality Index (WQI) and employed K-means clustering to categorize river stretches into "good" and "poor" quality zones based on the integrated data.
The Multi-Modal Dataset
The resulting dataset provides a comprehensive view of the Ganga, Yamuna, and Hindon rivers during the critical period of late 2015 to early 2016. It includes over 10,500 real-time sensor points, 117 reconciled lab test results, and 78 crowdsourced qualitative reports. This combination allows for a more holistic understanding of river health than any single data source could provide.
Key Insights and Results
Spatial and Ritual Impacts
The study tracked water quality over a 76.1 km stretch from Rishikesh to Roorkee. A key finding was the "One Bank" phenomenon: at Haridwar, the bank used for religious rites (Har-ki-Pauri) showed significantly higher pollution levels than the opposite bank. Interestingly, while Dissolved Oxygen (DO) levels remained within safe limits, electrical conductivity and pH spiked during peak bathing days. This suggests that the primary impact of the gathering is the introduction of dissolved solids and organic additives (milk, ghee, ashes) rather than immediate oxygen depletion.
Actionable Data via GangaWatch
By making the data accessible via APIs and the GangaWatch mobile app, the researchers have enabled the development of decision-support tools for the public. For example, pilgrims can check which ghats are safest for bathing on a given day based on real-time WQI values.
Clustering and WQI Trends
K-means clustering successfully identified "poor quality" zones specifically around high-density bathing areas during the Ardh Kumbh. WQI values fluctuated significantly according to the festival's peak bathing schedule; on certain peak days, water quality at specific ghats reached "Very Poor" or "Unsuitable for Drinking" classifications.
Conclusion and Future Work
While the current dataset of 10,000+ points is a significant step forward, it remains sparse relative to the massive scale of the Ganga basin. Future efforts will focus on improving data reconciliation between different agencies and addressing the limitations of current real-time sensors, which cannot yet measure biological parameters like Fecal Coliform (FC) in real-time. The ultimate goal is to use this multi-modal data to train machine learning models capable of predicting pollution spikes, allowing authorities to proactively manage water flow from upstream dams to "flush" the river during peak festival periods.
Summary updated