TAMARAW: Temperature- and Mosquito-Adjusted Risk & Alert Workflow

TAMARAW is our Philippines dengue modeling project. It couples a climate-forced vector-host transmission model with lightweight ETL so dengue risk can be monitored weekly at the city or barangay level and exported to dashboards or reports. The initial release provides a single-node SEI-SEIR demonstration driven by temperature and rainfall; the framework is designed to expand into a metapopulation with mobility, interventions, and observation models. The current demo implements temperature-dependent biting and extrinsic incubation, rainfall-driven adult recruitment, and a simple post-week-40 reduction in biting to mimic repellents or community action. All components are pure Python and run in seconds, making it easy to iterate with field teams.

Live Map: Philippines

Model Core (SEIR-SEI)

Humans follow a susceptible–exposed–infectious–recovered (SEIR) progression: \( S_h \to E_h \to I_h \to R_h \), while mosquitoes follow a susceptible–exposed–infectious (SEI) progression \( S_m \to E_m \to I_m \).

Human (SEIR):
\[ \begin{aligned} \frac{dS_h}{dt} & = -\beta\, a(T)\, m\, \frac{I_m}{N_m} S_h, \\ \frac{dE_h}{dt} & = \beta\, a(T)\, m\, \frac{I_m}{N_m} S_h - \sigma_h E_h, \\ \frac{dI_h}{dt} & = \sigma_h E_h - \gamma_h I_h, \\ \frac{dR_h}{dt} & = \gamma_h I_h. \end{aligned} \]
Mosquito (SEI):
\[ \begin{aligned} \frac{dS_m}{dt} & = \Lambda_m(R) - c\, a(T)\, \frac{I_h}{N_h} S_m - \mu_m(T,R)\, S_m, \\ \frac{dE_m}{dt} & = c\, a(T)\, \frac{I_h}{N_h} S_m - \sigma_m(T)\, E_m - \mu_m(T,R)\, E_m, \\ \frac{dI_m}{dt} & = \sigma_m(T)\, E_m - \mu_m(T,R)\, I_m. \end{aligned} \]

The temperature-dependent biting \( a(T) \), mosquito-to-human ratio \( m \), and the infectious fraction \( I_m / N_m \) together determine the force of infection from mosquitoes to humans. Rainfall enters through \( \Lambda_m(R) \), which modulates adult mosquito recruitment and thus scales the available vector population. Temperature also modifies extrinsic incubation \( \sigma_m(T) \) and mortality \( \mu_m(T,R) \), so the overall transmission potential reflects the combined, multiplicative influence of biting, recruitment, and survival. Lag times can be added as needed (e.g., 1–8 weeks for rainfall) to capture delayed effects.

Parameter Definitions

Symbol Description Units / Typical Range
\( \beta \)Transmission probability per bite– (dimensionless, often 0.2–0.5)
\( a(T) \)Temperature-dependent biting rate of mosquitoesbites·mosquito⁻¹·day⁻¹ (≈0.2–0.5)
\( m = N_m / N_h \)Mosquito-to-human population ratio– (typically 1–10)
\( N_m, N_h \)Total mosquito and human populationsindividuals
\( \sigma_h \)Rate of progression from exposed to infectious (humans)day⁻¹ (≈1/5–1/7)
\( \gamma_h \)Recovery rate of humansday⁻¹ (≈1/5–1/10)
\( \Lambda_m(R) \)Rainfall-dependent adult mosquito recruitment ratemosquitoes·day⁻¹
\( \sigma_m(T) \)Temperature-dependent extrinsic incubation rateday⁻¹ (≈1/5–1/10)
\( \mu_m(T,R) \)Temperature- and rainfall-dependent mosquito mortality rateday⁻¹ (≈0.05–0.2)
\( c \)Transmission probability from human to mosquito per bite– (≈0.3–0.7)

Data & ETL

Inputs

Temperature Rainfall Humidity Weekly dengue cases (DOH)

Start with daily T/R, aggregate to weekly means/sums, and create lagged features. Observation models (e.g., negative binomial with reporting fraction) plug in later.

Outputs

  • Weekly incidence and nowcasts
  • Scenario curves (e.g., repellents, source reduction)
  • PNG figures for web embedding
  • CSV exports for dashboards

Roadmap

References

  1. Ross-Macdonald style vector-host models for Aedes-borne disease.
  2. AI Low Voltage Innovations - dengue SEI-SEIR demo source code (dengue_model.py, run_ph_dengue_demo.py).