Local slope signals for clearer community preparedness.
PRAHARI proposes rugged, low-cost ESP32 sensor nodes placed near vulnerable hillsides (such as Wayanad's fragile ghats) to monitor hyper-local soil saturation, tilt velocity, and precipitation alongside macro-level advisories from the India Meteorological Department (IMD) and State Disaster Management Authorities (SDMAs).
Live Telemetry & Simulation
Seamlessly toggle between live api.prahari.space telemetry and synthetic scenario sandboxes.
Sense → Assess → Communicate
Dual capacitive soil probes, ADXL345 3-axis motion, and automated risk synthesis.
Dual Review Thresholds
60% & 85% review milestones—clearly marked unvalidated, pending local field calibration.
Supports real-time feed from https://api.prahari.space with instant fallback sandbox.
Bridging Regional Forecasts with Slope-Level Reality
How PRAHARI proposes to complement district-level meteorological advisories with hyper-local, community-accessible slope signals in vulnerable sectors like Wayanad.
Why District-Level Alerts Need Slope Context
In mountainous terrain (such as Wayanad, Uttarakhand, or the Western Ghats), the India Meteorological Department (IMD) and State Disaster Management Authorities (SDMAs) provide invaluable regional weather bulletins (e.g. orange or red monsoon alerts spanning hundreds of square kilometers).
However, whether a specific slope slips depends on hyper-localized geotechnical conditions: localized rain shadow cloudbursts, pre-existing groundwater saturation, slope angle of repose, drainage blockages, and micro-movements in soil strata. A village 5 km away may remain completely stable while an adjacent slope reaches critical saturation.
What PRAHARI Proposes
PRAHARI (meaning Sentinel or Watchman) is an open IoT slope sentinel framework developed by <HYPNEXIS>:
- Continuous Edge Monitoring: Measures subsurface soil moisture across dual depths (S1 & S2), rainfall presence, tilt displacement, and micro-vibrations.
- Contextual Assessment: Evaluates local readings alongside official IMD/SDMA regional alerts to estimate a composite risk score (0–100).
- Live Backend Integration: Connects to
https://api.prahari.spaceto stream live ESP32 prediction logs. - Community Empowerment: Equips Village Disaster Management Committees (VDMC) and panchayat focal points with local observational indicators.
🛰️ LIVE BACKEND MODE (api.prahari.space)
LIVE TELEMETRY TREND • Risk Score (%)
Real-time rolling observation window showing up to the last 40 telemetry samples.
HISTORICAL SENSOR TELEMETRY LOG
0 readings loaded| Timestamp | AI Probability | S1 (%) / ADC | S2 (%) / ADC | Rain | Vibration /10 | Tilt P/R | Mic dBFS | Mic Hz* | GPS | Wi-Fi RSSI | SIM Signal |
|---|---|---|---|---|---|---|---|---|---|---|---|
Connecting to backend at https://api.prahari.space...
|
|||||||||||
Proposed Edge Sensor Node Architecture
A rugged, off-grid IoT node engineered with accessible, low-power components to ensure hillside durability, battery longevity, and continuous multi-modal telemetry.
The Exposed Ridge • Fragile Western Ghats
In Wayanad's steep mountain tracts, heavy monsoonal precipitation rapidly supersaturates thin topsoil layers resting on steep, impermeable metamorphic bedrock. Traditional district weather radar tells you rain is falling, but only localized slope telemetry can detect when water begins trapped in the shear zone.
Off-Grid ESP32 Architecture
Anchored directly on high-risk terrain without relying on mains electricity or cellular towers. A dual-core 32-bit Xtensa MCU samples all geotechnical channels, applies median noise filtration, and operates for 72+ continuous hours of heavy overcast via a 10W monocrystalline solar panel and 6000mAh LiFePO4 battery pack.
Dual-Depth Pore-Water Infiltration
Surface soil drains and dries rapidly in tropical heat, but deep subterranean saturation is what triggers slope shear failure. PRAHARI deploys dual corrosion-resistant capacitive probes buried at 30 cm (S1) and 80 cm (S2) to monitor the downward hydraulic wetting front and pore-water pressure spikes in real time.
Micro-Radian Angular & Tremor Telemetry
Rigidly coupled to an anchored slope datum, the ADXL345 3-axis accelerometer tracks sub-degree gravitational tilt deviations (Δpitch and Δroll) caused by plastic soil creep, while simultaneously capturing high-frequency micro-tremors from fracturing rock strata before macroscopic slope rupture occurs.
ESP32 Dual-Core MCU
Dual-core 32-bit Xtensa LX6 @ 240MHz with ultra-low-power co-processor. Supports deep sleep states (~15 µA), integrated ADC channels, I2C/SPI interfaces, and proposed long-range radio / LoRa telemetry.
ADXL345 3-Axis Accelerometer
High-resolution (13-bit) measurement up to ±16g. Mounted rigidly on an anchored slope datum to measure static gravitational tilt changes (<0.1°) and high-frequency tremor bursts from internal shear movement.
Dual Capacitive Soil Moisture Probes
Corrosion-resistant capacitive dielectric probes installed at two distinct slip depths (S1 at 30 cm and S2 at 80 cm) to observe pore-water infiltration gradients and shear planes without electrode electrolysis.
Tipping Bucket & Rain Presence
Self-emptying dual-bucket mechanism with magnetic reed switch paired with optical surface rain detection. Generates hardware interrupts on the ESP32 per 0.2 mm of precipitation to calculate instantaneous rate and wet/dry surface state.
DHT22 Temperature & Humidity
Calibrated capacitive humidity sensing element and thermistor. Evaluates ambient vapor saturation, temperature inversion, and evapotranspiration rates to provide atmospheric context to drying cycles.
BMP180 Barometric Air Pressure
Piezoresistive pressure sensor reading 300 to 1100 hPa. Rapid barometric pressure plummets often herald approaching localized mountain convective squalls and cloudbursts before rain strikes.
Solar Harvesting & LiFePO4 Energy System
10W monocrystalline photovoltaic panel paired with an MPPT charge controller and a 6000 mAh Lithium Iron Phosphate (LiFePO4) cell pack. Engineered to deliver 72+ hours of autonomous operation during continuous monsoon overcast, with thermal resilience across mountain freezing and summer heat.
Sense → Assess → Communicate
A three-stage data pipeline designed for bandwidth efficiency, edge resilience, and village-level accessibility.
SENSE
The ESP32 node awakes from deep sleep every 60 seconds (or immediately upon tilt/rain interrupt). It captures synchronized readings across all sensor channels:
- Samples accelerometer (ADXL345) for static pitch/roll and vibration energy.
- Reads dual capacitive probes for volumetric soil moisture (S1 & S2).
- Polls tipping bucket count, BMP180 barometric delta, and DHT22 climate metrics.
- Applies edge-level median filtering to discard mechanical sensor noise.
ASSESS
The backend at api.prahari.space runs the proposed Random Forest scoring algorithm to combine multi-dimensional slope parameters:
- Cross-references cumulative rainfall against soil moisture saturation curves.
- Checks rate-of-change on tilt angles (Δtilt/Δtime) to identify plastic creep.
- Ingests regional IMD / SDMA advisory status to weigh background meteorological risk.
- Outputs a normalized 0–100 composite risk score with confidence intervals.
COMMUNICATE
Information is delivered in simple, actionable formats to local stakeholders without requiring complex technical interpretation:
- Presents data on a low-bandwidth local village dashboard (like this one and prahari.space).
- Flags initial review states (60% Advisory / 85% Critical Review) for village volunteers.
- Prompts local disaster committees to inspect drainage channels and road culverts.
- Logs verifiable telemetry for post-monsoon geotechnical calibration.
Under the Hood: The 3D Folding Telemetry Architecture
Explore how raw ESP32 byteframes fold into multi-variable ML tensors and unfold into community village alerts.
ESP32 Binary Struct (C/C++)
sizeof = 32 BytesPacked fixed-width struct transmitted over LoRa/SPI. Maximizes radio battery lifetime by avoiding verbose JSON at the physical node.
typedef struct __attribute__((packed)) {
uint16_t magic_header; // 0xAA55 (Sync Frame)
uint32_t timestamp_sec; // Epoch GPS/NTP
int16_t accel_x_mg; // ADXL345 Tilt X
int16_t accel_y_mg; // ADXL345 Tilt Y
int16_t accel_z_mg; // ADXL345 Vibration
uint16_t soil_vwc_s1; // Topsoil Probe (30cm)
uint16_t soil_vwc_s2; // Deep Bedrock (80cm)
uint16_t rain_pulse_cnt; // Tipping Bucket 0.2mm
int16_t temp_centic; // DHT22 x100 (°C)
uint8_t humidity_pct; // Relative Humidity
uint32_t pressure_pa; // BMP180 Barometric
uint16_t crc16_checksum; // CRC-CCITT Integrity
} prahari_packet_t;ML Geotechnical Input Vector
Tensor Shape: [1, 5]Synthesized non-linear feature array ingested into the 100-tree Random Forest classifier to compute localized failure probability.
// Geotechnical Feature Vector normalized to [0.0 - 1.0]
const FeatureVector = {
pore_saturation_ratio : 0.824, // 35% Weight
cumulative_24h_rain_mm: 78.500, // 28% Weight
tilt_angular_rate_deg_h: 0.142, // 22% Weight
rms_vibration_energy_g: 0.038, // 10% Weight
barometric_rate_hpa_h : -1.80 // 5% Weight
};
// Ensemble Multi-Tree Inference Call
let composite_risk = evaluateRandomForest(FeatureVector);
// Output: 74.2 / 100 (Advisory Review Tier)Village Advisory JSON Payload
Payload = 168 BytesStandardized lightweight JSON published to MQTT brokers and village dashboard displays for clear local stakeholder comprehension.
{
"system": "PRAHARI",
"node_id": "CHOORALMALA-MAST-01",
"status": "ADVISORY_REVIEW",
"risk_score": 74.2,
"threshold": ">= 60% UNVALIDATED",
"village_advisory_en": "Inspect road culverts and natural stormwater chutes.",
"village_advisory_ml": "ഓടകളിലെ തടസ്സങ്ങൾ പരിശോധിച്ച് നീക്കം ചെയ്യുക.",
"attribution": "<HYPNEXIS>"
}Proposed Random Forest Risk Synthesis
Why an ensemble of decision trees is proposed to model non-linear geotechnical slope interactions from multi-modal sensor inputs.
Why Random Forest for Landslide Risk?
Slope failure is rarely caused by a single isolated variable. For example:
- High Rainfall + Dry Sandy Soil: Water drains quickly; immediate risk remains low.
- Moderate Rainfall + 90% Saturated Soil: Pore-water pressure spikes; soil shear strength collapses.
- Micro-Tilt Change Alone: Could be wind buffeting or animal contact; requires corroborating moisture and vibration signals to confirm ground movement.
A Random Forest ensemble model handles these non-linear cross-correlations exceptionally well. By aggregating predictions across multiple decision trees trained on diverse subsets of geotechnical variables, it prevents false triggers caused by any single faulty sensor while remaining lightweight enough for edge or gateway micro-inference.
Proposed Feature Weight Distribution
Relative importance proposed for the Random Forest model (subject to geotechnical field calibration):
How the Proposed Ensemble Evaluates Multi-Sensor Inputs
Hydrological Focus
Evaluates cumulative rain vs soil drainage capacity and pore pressure build-up.
Geomechanical Focus
Evaluates accelerometer static tilt angles (Δθ) and high-frequency tremor spikes.
Atmospheric Storm Focus
Correlates barometric pressure collapse with sudden precipitation bursts and humidity.
Random Forest Score
Averages individual tree predictions to generate an interpretable 0–100 composite risk score.
Proposed Initial Review Thresholds
PRAHARI introduces a dual-tier review threshold logic for community preparedness. Crucially, these are unvalidated initial baselines that require site-specific local field calibration.
Elevated Awareness Threshold
Proposed trigger when sustained rainfall combined with rising soil saturation passes preliminary stability thresholds.
- Local disaster committee members check drainage paths & debris basins.
- Verify latest regional IMD radar & SDMA meteorological bulletins.
- Node telemetry transmission frequency increases from 15 min to 2 min.
- Community focal points confirm communication channels with village elders.
Critical Technical Review Threshold
Proposed trigger when extreme soil saturation occurs concurrently with detected micro-tilt displacement or anomalous vibration spikes.
- Flag urgent technical inspection condition for local disaster committee.
- Direct visual verification of hillside cracks and tension scarps by trained volunteers.
- Consult official district disaster administration for official instructions.
- Prepare standby shelter kits in accordance with standard village safety protocols.
MANDATORY NOTICE: LOCAL FIELD CALIBRATION IS REQUIRED
The 60% and 85% thresholds shown above are unvalidated mathematical benchmarks created for hackathon demonstration. No two hillsides behave identically:
Slope angle of repose (e.g., 28° vs 55°), soil porosity, hydraulic permeability, bedrock depth, shear strength parameters (cohesion c and internal friction angle φ), and historical rainfall-trigger thresholds vary dramatically across Wayanad, Uttarakhand, Himachal Pradesh, and Sikkim.
These thresholds must NEVER be deployed in the field without empirical geotechnical soil sampling, multi-season baseline data logging, and validation alongside certified disaster management geologists.
Proposed Pilot Implementation Roadmap
A staged, responsible progression from breadboard bench verification to multi-slope academic pilot review.
Interactive Week-by-Week Deployment Timeline
Step through each milestone week with folding transitions to observe hardware tests, geotechnical validations, and deliverables.
Bench Demo & Hardware Validation
Assembling breadboard prototypes with ESP32, ADXL345, DHT22, BMP180, and dual capacitive probes. Rigorous validation of sensor communication over I2C/SPI buses, synthetic signal calibration, interrupt handling for tipping buckets, and power profiling under simulated deep sleep cycles.
Controlled Field Test
Deployment of a single prototype node on an uninhabited, accessible hillside slope testbed. Evaluation of IP66 weatherproofing against heavy rain, solar panel MPPT harvesting through cloud cover, telemetry packet reliability, and baseline noise analysis of diurnal temperature fluctuations on accelerometer tilt.
Pilot at Three Slopes
Deploying 3 sensor nodes across diverse slope profiles (e.g., road-cut slope, agricultural terraced slope, and forest embankment) in collaboration with a partner village community. Establishing local LoRa mesh communication, conducting village focal-point orientation workshops, and gathering feedback on dashboard clarity.
Evidence Review & Geotechnical Correlation
Rigorous empirical review comparing captured sensor trends against official IMD monsoon precipitation records, observed micro-slips, and geotechnical soil surveys. Tuning the Random Forest model feature weights, measuring false-positive rates, and submitting transparent findings to State Disaster Management Authorities (SDMAs).
Safety, Boundaries & Limitations
PRAHARI is strictly an academic hackathon proposal. Responsible deployment mandates complete transparency regarding what this system is—and what it is not.
Not a Public Warning Service
PRAHARI is not authorized, certified, or intended to operate as a public early-warning siren or disaster broadcasting platform. It must never be relied upon as the primary safety mechanism for human settlements.
Does Not Issue Evacuation Orders
Evacuation orders are exclusively the legal, administrative, and operational prerogative of designated civil authorities (District Collectors, SDMAs, NDMAs). PRAHARI will never issue evacuation commands or trigger autonomous alarms.
Does Not Connect to Real Hardware Yet
While the live mode connects to https://api.prahari.space, all data should be treated as experimental research packets. No certified safety-critical life-support operations are automated.
Must Not Replace Official Guidance
In all meteorological events and landslide alerts, citizens and administrators must adhere strictly to advisories from the India Meteorological Department (IMD), National Disaster Management Authority (NDMA), and local administration.
Summary of Explicit Project Boundaries
| Capability / Feature | Current Status in Hackathon Project | Explanation & Boundary |
|---|---|---|
| Live ESP32 Telemetry Feed | ACTIVE CLIENT INTEGRATION | Website polls api.prahari.space directly; includes instant simulator fallback when physical nodes sleep. |
| Prediction Accuracy Claim | NO VALIDATED ACCURACY | Machine learning model is a conceptual proposal; no validated statistical accuracy is claimed. |
| Lead Warning Time (e.g. 30-min) | NO TIME GUARANTEE | We do not claim a fixed warning window. Landslide kinetics vary from gradual creep to abrupt rupture. |
| Automated SMS / Sirens | NOT IMPLEMENTED | No automated public alarms or SMS dispatches to avoid risk of panic from uncalibrated test data. |
| Real-World Field Deployment | EXPERIMENTAL PILOT ONLY | Project is in the prototype and field test phase according to the proposed roadmap. |