Sampling triage
Rank sites when crews, lab capacity, boats, or field windows are limited.
For agencies, labs, researchers, and water-data teams
Turn water data into decisions you can defend.
MBAL, the Monterey Bay AI Lab, builds leak-safe forecasting, triage, and evidence systems for coasts, rivers, lakes, reservoirs, estuaries, and shellfish waters. Every model has to beat the simple rule first.
Use case
Sample smarter
Data stance
Timestamp ready
Evidence rule
Baseline first
Delivery shape
Pilot to product
What partners get
A practical data-science partner for monitored water systems.
MBAL is locally anchored in Monterey Bay and Elkhorn Slough, then built for broader water systems. We help teams turn historical monitoring, sensor, weather, flow, toxin, oxygen, and field-action data into tools that say where to sample, what to watch, and which claims survive.
The output is not an AI demo. It is a tested decision surface, a clear baseline comparison, a validation report, and a next-step plan your scientists, operators, or customers can challenge.
Early warning and watchlists
Flag likely onset, recurrence, or threshold crossing without pretending uncertainty vanished.
Signal audits
Test whether a proposed driver actually beats local history, season, persistence, or current practice.
Data readiness
Find timestamp, unit, station, leakage, and source-quality problems before they become product risk.
Pilot path
Four weeks to know whether your data can support a decision product.
Start with one water system, one field decision, and one real baseline. MBAL evaluates the data, builds the first leak-safe benchmark, reports what works and what fails, then scopes the smallest useful product.
Pick the decision
Sample now, defer, rank sites, flag onset, audit a driver, or expand monitoring.
Share the evidence
Observed time, available time, location, units, station IDs, labels, and action history.
Beat the baseline
Rain rule, persistence, seasonal floor, local memory, or the rule your team uses now.
Decide what ships
Dashboard, API, ranked worklist, monitoring plan, evidence report, or stop-work null.
Current proof
Results that show what MBAL can do.
These highlighted cards come from the current evidence library. They show useful lift, defined scope, and the kind of decision products MBAL can build with partner data.
Data readiness
Know what your monitoring data can support.
Fetchers pull source-native data into curated layers, then promotion checks decide what can support a named result.
Source inventory
Current source rows from the evidence library.
Modeling toolkit
Use the simplest method that wins.
MBAL tests baselines, tree models, rankers, transfer methods, time-series models, and source audits against the same decision target.
Water domains
One lab pattern, many water systems.
Each domain gets the same treatment: define the field decision, compare against the obvious simple rule, then publish the status without varnish.
Beaches and estuaries
Enterococcus, rainfall, discharge, tides, and advisory triage.
Rivers and stormflow
Gauge-linked runoff, turbidity, bacteria transport, and event onset.
Lakes and reservoirs
Bloom watchlists, cyanobacteria screens, oxygen stress, and sampling budget.
Shellfish and harmful algal blooms
Toxin recurrence, algae precursors, and next-visit risk ranking.
Hypoxia and oxygen
Low-oxygen onset, stratification, and short-lead monitoring triage.
Toxicity and wildlife pressure
Ambient toxicity screens, marine mortality signals, and source quality gates.
Results ledger
Every audited result, with status and evidence.
AP average precision, useful for rare-event ranking
FIB fecal indicator bacteria
DA domoic acid, a shellfish and wildlife toxin
HGBT histogram gradient boosted tree model