Assimilating IoT rain data into SWMM

I’m running a real-time PySWMM setup where 12 LoRaWAN rain gauges stream 5‑min totals over MQTT, and an ensemble Kalman filter nudges Green–Ampt parameters and Manning n on the fly; the pipeline is Dockerized with Kafka and pushes SWMM outflows as boundary conditions to a 2D HEC‑RAS grid. Anyone tried a SWMM+EnKF loop and seen stability or latency hiccups during >30 mm/hr bursts, or have bounds/regularization tricks to keep the filter from diverging?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍‍​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠‌‌⁠⁠‌⁠‌​‌‍⁠⁠‌⁠​​‌‍‍‌‌‍​⁠​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‌​⁠‍​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌​​‍‌‍​‌‌‌​‌‌⁠‍‌‌​‌​‌‍​‌‌⁠​‍‌‍⁠‌‌‍⁠‍‌‍⁠⁠‌⁠​‌‌‍​⁠​⁠​‌​⁠‍‌​⁠‌‌‌⁠‌‌​‍​‍‌⁠⁠‌​​

Saw instability >30 mm/hr; freeze Manning n, limit EnKF step, median-filter ‘5‑min totals’ before nudging.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍‍​‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‌​​⁠‌‍​⁠‌​​⁠‍‌​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‌‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠​‍‌​⁠‌‌‍​‌​⁠​‌‌‍⁠‍​⁠‌​‌​‌‌‌‌‍​‌​‌‌‌​​‌‌‍‍​​⁠‌⁠‌‍‍⁠​‍⁠‌‌⁠‍‌‌​​‍​‍​‍‌⁠⁠‌​