Predictive Analytics on Carbon-Awareness of LAX Logistics
Dr. Ming Wang · Brian Ta · Daniel Ramirez
CSULA Department of CIS | SAIES Research Organization | NSF Grant Project
Most enterprise carbon accounting tools — including SAP Green Ledger (GA December 2024) — are compliance tools. They tell you what you already emitted. The gap nobody has filled: when a supply chain decision is being made in real time — hold this order, reroute it, or expedite it — carbon is completely invisible at that moment.
CAPE changes that. When an order is flagged as high risk for late delivery, CAPE calculates the downstream carbon cost of that delay before the fulfillment decision is made, not after.
Research question: Can integrating ERP transactional data with carbon emissions records produce a leading indicator of carbon exposure — surfacing risk before it becomes emissions?
Short answer: Yes. Late order risk is a statistically meaningful predictor of CO₂e penalties. 60.8% of all direct (Scope 1) emissions in the ERPsim dataset come from inventory overstock — not from moving goods. That's the carbon cost of delayed decisions.
Project Link: https://cape-dashboard.streamlit.app/
| Metric | Value | Significance |
|---|---|---|
| Total CO₂e | 421,694 kg | Full simulation carbon footprint |
| Scope 1 Direct | 234,250 kg | 55.5% of total emissions |
| Overstock CO₂e | 142,500 kg | 60.8% of all Scope 1 — not from shipping |
| Avg Overstock Penalty | 1,827 kg/period | Carbon cost of late order buildup |
| High Risk Periods | 8 of 38 | Periods scoring above 0.6 threshold |
| Highest Risk Period | R3-S6 (score: 0.834) | Carbon intensity: 0.146 kg CO₂e/$ |
| LAX Peak Month | March 2021 | 254,057 tons — real-world validation anchor |
CAPE Risk Score = (Carbon Intensity Scaled × 0.70) + (Overstock CO₂e Scaled × 0.30)
Periods scoring above 0.6 are flagged as High Risk and surfaced as alerts in the dashboard. Carbon intensity is weighted more heavily (70%) as a forward-looking signal; overstock penalty (30%) is a lagging indicator of accumulated fulfillment failures.
Provided by Dr. Ming Wang through SAP University Alliance access.
| Dataset | Rows | Key Fields |
|---|---|---|
| Sales.xlsx | 2,568 | SIM_ROUND, SIM_STEP, NET_VALUE, COST, QUANTITY |
| Carbon Emissions.xlsx | 2,397 | SCOPE, TYPE, TOTAL_CO2E_EMISSIONS, ORIGIN, DESTINATION |
| Inventory.xlsx | 6,888 | INVENTORY_OPENING_BALANCE, PLANT, STORAGE_LOCATION |
| Financial Postings.xlsx | 2,424 | GL_ACCOUNT_NAME, AMOUNT, DEBIT_CREDIT_INDICATOR |
| Purchase Orders.xlsx | — | Order-to-receipt timing, lateness proxy |
| Stock Transfer.xlsx | — | Inter-DC movements, Scope 1 emissions trigger |
| ERPSIM.xlsx | — | Competitive pricing rounds, market share by team |
The core technical contribution is joining Sales and Carbon Emissions on SIM_ROUND + SIM_STEP — a join not previously performed in ERPsim academic literature. This produces 160,132 rows connecting every sales transaction to its corresponding carbon record across 38 matching simulation periods.
1,712 records of monthly air freight tonnage at LAX from 2006–2023 (data.lacity.org). Serves as real-world empirical grounding for the freight mode-switching logic: when ground shipments are delayed, some percentage escalate to air freight, which carries significantly higher carbon intensity per ton-mile. LAX peak (March 2021: 254,057 tons) aligns with CAPE's highest-risk simulation periods.
| Component | Technology |
|---|---|
| Data processing | Python, pandas |
| Analytical queries | DuckDB |
| ML models | scikit-learn (Random Forest) |
| Dashboard | Streamlit |
| Visualization | Plotly |
| Deployment | Streamlit Cloud |
CAPE/
├── Home.py # App entry point
├── pages/
│ ├── 1_CAPE_Carbon.py # Carbon risk scores, overstock analysis, LAX validation
│ ├── 2_Control_Tower.py # Order risk intelligence, carbon alerts
│ └── 3_Sales_Intelligence.py # Brian's dashboard (in progress)
├── data/ # ERPsim datasets and LAX cargo data
├── ml/ # Pre-trained Random Forest models
├── control_tower/ # Order risk scoring module
├── requirements.txt
└── CAPE_Analysis.ipynb # Exploratory analysis and data join notebook
| Phase | Status | Description |
|---|---|---|
| 1 — Data join + carbon risk scores | ✅ Complete | Sales × Carbon join confirmed; CAPE Risk Score implemented; LAX validation done |
| 2 — Sales Intelligence | 🔄 In progress | Merging Brian's competitive pricing dashboard as a Sales tab |
| 3 — Control Tower integration | 🔄 In progress | Integrating Random Forest order risk model as ML input to CAPE scores |
| 4 — Behavioral study | ⏳ Planned | Human-AI feedback loop study with ERPsim game participants |
| 5 — Paper + SAP presentation | ⏳ Planned | NSF research presentation (late June/early July 2026); SAP University Alliance submission |
| SAP Green Ledger | CAPE | |
|---|---|---|
| Orientation | Backward-looking | Forward-looking |
| Function | Records carbon per financial transaction | Predicts carbon before the order is late |
| Scope tracking | Scope 1/2/3 compliance reporting | Carbon intensity as a real-time risk signal |
| Decision support | Audit-ready ESG reporting | Alerts at the moment a fulfillment decision is made |
CAPE is the predictive layer that feeds what Green Ledger will eventually need. They are complementary, not competing.
A 2025 systematic literature review identified ERP-native carbon prediction at the operational decision level as critically under-researched. SAP holds a patent on carbon-aware inventory optimization at the planning level — a different problem. No published paper has used the ERPsim carbon emissions table for ML-based predictive scoring. The combination of order risk scoring, carbon exposure forecasting, and human-AI behavioral analytics does not exist in the current literature.
| Name | Role |
|---|---|
| Daniel Ramirez | CIS — app development, ML, data engineering, LAX domain expertise |
| Brian | Finance/Supply Chain — sales dashboard, business framing, competitive analysis |
| Dr. Ming Wang | Faculty advisor — CIS Dept. Chair, CSULA; SAP University Alliance access |
Affiliated with: CSULA MBDS · SAIES Research Organization · NSF Undergraduate Research Program (Spring/Summer 2026)
Live Dashboard: https://cape-dashboard.streamlit.app