Use-Case: Public Forest

Istanbul, Türkiye

Executive Summary: Enhancing Wildfire Detection in Istanbul

Protecting forests in and around Istanbul involves more than just scale; it requires addressing complexity. The Prince Islands, particularly Heybeliada, represent a high-risk interface where forested areas and settlements are closely interwoven. Seasonal tourism increases pressure, and logistics are constrained by island geography. Monitoring every blind spot continuously is challenging. Cameras need a clear line of sight and dense placement to eliminate blind spots. While satellite monitoring aids regional situational awareness, it lacks the immediacy required for localized early warning.

In this context, the Istanbul Regional Directorate of Forestry (Istanbul OGM) initiated a field pilot on Heybeliada to enhance early-stage awareness and improve readiness during peak season. ForestGuard was deployed as an additional digital layer to support early warning and operational decision-making. The pilot aimed not to replace existing response assets like patrols, watchtowers, aerial response, or public safety measures, but to provide a practical early-stage risk layer that offers faster, localized signals. This helps teams prioritize where to look and when to act.

To address this gap, ForestGuard started a public-sector proof of concept (PoC) with Istanbul OGM on Heybeliada in April 2024. The PoC focused on early-stage wildfire risk detection before flame visibility and aimed to produce operationally usable risk states that guide teams toward the right locations at the right times. Public interviews and press coverage featuring Istanbul OGM leadership, including Ikram Celik (Head of the Wildfire Fighting Branch), discussed the rationale and operational value of the Heybeliada deployment.

1) The Problem We Targeted

Istanbul faces a classic wildfire challenge amplified by density. Forested zones exist close to urban life and tourism activity. Even when the overall forest area is limited compared to rural regions, the consequences can be severe because ignition can occur near settlements and spread quickly. Heybeliada is particularly demanding due to several risk multipliers:

• Wildland Urban Interface Conditions: Forest and residential areas are in close proximity.

• Seasonal Intensity: Visitor activity increases the probability of ignition sources.

• Island Logistics: Response and resupply options are constrained compared to mainland operations.

• Monitoring Limitations: Cameras require a clear line of sight and dense coverage to eliminate blind spots. Satellite monitoring is valuable for regional situational awareness but does not provide the continuous, localized ground-truth signals needed for operational early warning.

  
  

The pilot was designed around a practical question: Can Istanbul OGM gain earlier, location-specific awareness before flame visibility and translate that into actions that reduce escalation risk?

2) Pilot Objectives

ForestGuard and Istanbul OGM aligned on clear operational objectives:

• Early-Stage Wildfire Risk Detection: Before flame visibility, using multi-parameter sensing and AI-powered anomaly analytics.

• Simple Operational Risk Indicator: Helps teams prioritize attention and patrol.

• Reliable Field Operation: Ensures functionality through peak season conditions.

• Documentation and Learnings: Supports scaling decisions across other islands and high-risk zones in Istanbul.

  
  

3) Deployment Overview

ForestGuard was deployed on Heybeliada as a seasonal public-sector field pilot initiated with Istanbul OGM. The goal was to test whether earlier, location-specific awareness could strengthen readiness in a high-risk island environment. The pilot began in April 2024 and focused on creating a practical early warning layer in an area where forested zones, settlements, and visitor activity exist in close proximity. Instead of attempting blanket coverage across the island, the deployment concentrated on selected high-risk points where earlier signals could provide the greatest operational value.

Placement decisions were made collaboratively with Istanbul OGM, utilizing local operational knowledge of terrain, ignition exposure, and areas where traditional monitoring has blind spots. This was crucial because Heybeliada is not a simple monitoring environment. Vegetation, topography, and settlement patterns create pockets where line-of-sight tools may not provide consistent early awareness. By placing sensors at multiple risk points, the pilot was designed to support more localized monitoring of environmental risk indicators during the season and help teams understand where attention should be directed first.

The deployment utilized off-grid capable configurations designed for outdoor conditions and continuous seasonal operation. The pilot was structured not as a stand-alone replacement for existing assets but as an additional layer that could work alongside patrols, watchtowers, and other response resources. This approach made the project operationally realistic for a public entity: starting with a focused deployment at critical points, validating how the signals perform in real field conditions, and generating the evidence needed for future scale decisions across other islands and high-risk zones in Istanbul.

4) How the Pilot Worked Operationally

ForestGuard is designed to create an early-stage risk layer that is easy to use under real operational pressure:

• Field Nodes: Monitor environmental signals and send them to the platform.

• AI-Powered Anomaly Analytics: Converts patterns into a risk indicator.

• Alerts and Risk Changes: Can be reviewed on a map-based view and used to guide patrol decisions.

  
  

The pilot emphasized operational practicality:

• Avoid Overwhelming Teams: Prevent constant high-severity alerts.

• Provide Early Signals: Guide targeted checks.

• Create a Learning Loop: Understand thresholds and site behavior over time.

  
  

The PoC used a tiered risk logic designed to help grid operators act without overwhelming teams:

• Low Risk: Early anomaly signals, used to build awareness and trend monitoring.

• Medium Risk: More persistent or correlated signals, triggering closer attention.

• High Risk: Actionable escalation, triggering targeted patrol or readiness actions.

• Highest Risk State: Urgent escalation, triggering immediate operational response.

• Direct Fire Case: Confirmed early-stage fire detection event.

  
  

The key was not the volume of signals but the escalation funnel that filters attention and prevents alert fatigue.

5) Results from the Six-Month PoC

The Heybeliada PoC aimed to produce two types of results:

1. Operational Validation: Confirming that the system can run in real outdoor conditions and support Istanbul OGM workflows.

2. Practical Risk and Escalation Picture: Helping teams focus attention and prioritize patrol and response actions.

   
   

5.1 Field Operation Validation

During the PoC period, Istanbul OGM and ForestGuard validated the following in field conditions:

• Devices operated reliably in outdoor conditions on the island.

• Data transmission and monitoring workflows were validated in daily operation.

• Operators could review risk status on the platform and use it for decision support.

• Thresholds and site behavior were observed and tuned over time.

  
  

If you want to publish quantitative values later, insert them here as a short KPI table. Suggested KPIs are uptime during the PoC period, alert delivery latency, and number of sensors active.

5.2 Risk Detections and Escalation Funnel

ForestGuard employs a tiered risk workflow designed to avoid alert fatigue. The Heybeliada PoC deliberately separated what is visible on the dashboard from what is pushed as a notification to first responders. Low and medium risk events were visible on the dashboard for situational awareness and trending but did not generate notifications. High and above events triggered notifications, with clear operational instructions aligned with Istanbul OGM workflows.

During the PoC, ForestGuard detected the following risk events:

• Low Risk Cases: 1000+ low-risk cases were recorded. These events were visible on the dashboard but did not trigger notifications.

• Medium Risk Cases: 500+ medium-risk cases were recorded. These events were visible on the dashboard but did not trigger notifications.

• High Risk Notifications: 150+ high-risk events triggered notifications to first responders to be cautious in the area.

• Highest Risk Notifications: 52 events triggered notifications to first responders to check the perimeter.

• Direct Fire Risk Notifications: 6 direct fire risk events triggered notifications to first responders to respond immediately.

  
  

This escalation funnel is crucial operationally. It shows that the system does not treat every weak signal as an emergency. Most events remain in low and medium risk for awareness only, while notifications are reserved for the levels that require action.

---

Additional Proactive Weather Condition Reminder

In addition to event-based risk states, ForestGuard was configured with a proactive reminder rule for extreme conditions. When temperatures rise above 35℃ and relative humidity drops below 20%, the system notifies first responders. This allows them to take precautionary actions such as watering the perimeter and increasing readiness.

---

6) How Our Technology Helps Public Entities

Public entities operate wildfire readiness under constraints that private operators rarely face. Budgets are seasonal and fixed. Response teams are distributed across districts. Decisions must be defensible, documented, and consistent with established procedures. At the same time, wildfire risk is increasingly dynamic, driven by weather extremes, visitor pressure, and fast-changing local conditions. ForestGuard is designed to help public entities manage this reality by adding an operational early warning layer that fits into existing prevention and response structures.

6.1 Early Awareness Where Traditional Monitoring Has Blind Spots

Cameras are effective when they have a clear line of sight, dense coverage, and reliable infrastructure. However, in many public forests and interface zones, those conditions are not universally available. Terrain creates hidden pockets, and vegetation can obstruct visibility. Installing and maintaining cameras across every blind spot becomes costly and slow. Satellite monitoring supports regional situational awareness but is not a continuous, local ground-truth system for operational early warning. It is most valuable as a broad layer rather than a tool that can guide minute-by-minute field actions. ForestGuard fills this gap by placing sensors at critical points where teams must act quickly, especially in zones with older flora and areas that are hard to monitor using other technologies.

6.2 Turning Environmental Signals into Actionable Priorities

Public response teams cannot treat every weak signal as an emergency. If notifications are too frequent, teams stop trusting them. Conversely, if notifications are too rare, early opportunities are missed. ForestGuard uses a tiered risk workflow that separates situational awareness from operational alerting. In the Heybeliada PoC, low and medium risk events were available on the dashboard for trending but did not generate notifications. Notifications were reserved for higher risk states that require action. This approach supports real operations in three ways:

• It keeps first responders focused on the small number of cases that matter most.

• It reduces alert fatigue and maintains trust in the system.

• It creates a clear, repeatable logic for patrol prioritization and readiness.

  
  

6.3 Supporting Preparedness Actions, Not Only Event Response

Public wildfire management is not solely about responding to events; it is about readiness during extreme conditions. ForestGuard supports this by enabling proactive rules based on environmental thresholds. In the Heybeliada deployment, a reminder alert was configured for high temperatures and relative humidity below 20 percent. This notification helps first responders take precautionary actions such as watering the perimeter, increasing patrol readiness, and preparing resources before an event escalates.

6.4 Improving Coordination and Documentation

Public entities require documentation and repeatability. When an alert triggers action, the chain of events should be explainable. ForestGuard supports operational logging by providing a time-stamped view of risk states and notifications. This can improve after-action reviews and help agencies refine their response playbooks over time. It also aids communication with municipal leadership and other stakeholders about why certain areas were prioritized during high-risk days.

6.5 Scalable Model for Districts, Islands, and High-Risk Zones

Public deployments are most successful when they scale in phases. The practical path is to start with the highest-risk points, validate the workflow with the teams who will use it, and then expand coverage in a structured way. The Heybeliada PoC demonstrated that a focused deployment at critical points can produce operational value while generating the learning needed to plan the next phase. This model applies to other public use cases such as:

• Peri-urban forest interfaces near dense residential areas.

• High visitation recreation zones.

• Strategic forest corridors connected to infrastructure.

• Islands or districts where logistics constrain response speed.

  
  

The outcome is not only improved detection; it is a repeatable operating model for early warning, prioritization, and readiness.