The Orbiting Carbon Observatory-2 (OCO-2) model intercomparison project (MIP) to study the impact of assimilating OCO-2 retrieval data into atmospheric inversion models
Everything you need to know about this platform
This interactive visualization displays atmospheric CO₂ concentration levels across Indonesia's municipalities (kabupaten/kota) from historical corrected satellite observations and future predictions. The data helps understand regional variations in greenhouse gas levels and climate patterns.
Historical Data: Derived from processed and corrected Level 2 (L2) total column CO₂ measurements from NASA's OCO-2 (Orbiting Carbon Observatory-2) and OCO-3 (Orbiting Carbon Observatory-3) satellites. These satellites use spectrometers to measure atmospheric CO₂ concentrations with high precision.
Prediction Data: Generated using Random Forest–based Hybrid Time Series Predicting. This combines the strengths of statistical time series methods with machine learning to predict future CO₂ concentration trends.
The color scale represents CO₂ concentration levels in parts per million (ppm). Green indicates lower concentrations, transitioning through yellow and orange to red for higher concentrations. The exact range is shown in the legend at the bottom of the sidebar.
Purple circle markers on the map represent fire hotspot detections from satellite thermal sensors. The size of each circle corresponds to the number of hotspots detected in that municipality. Fire hotspots are often associated with forest fires, agricultural burning, and peatland fires, which are significant sources of CO₂ emissions. The hotspot data helps identify correlations between fire activity and elevated atmospheric CO₂ concentrations.
Dynamic Scale (default) adjusts the color range based on the current date's data, providing better contrast between regions within each time period. This is especially useful for prediction data where variations are smaller.
Global Scale uses the same color range across all dates, making it easier to compare absolute values over time but potentially reducing visual contrast in periods with smaller variations.
Click the "Play" button or press the Spacebar to start an automatic animation that cycles through all available dates. This lets you see how CO₂ levels change over time across Indonesia. Press again to pause.
Click the "Export Data" button to download a CSV file containing CO₂ concentration and fire hotspot data for all municipalities on the currently selected date. The file includes province names, city names, CO₂ values, data source (historical or prediction), and fire hotspot counts.
The orange "PREDICTION" badge appears in the timeline when you're viewing predicted future data rather than historical observations. In the chart, prediction data is shown as a dashed orange line with a vertical marker indicating where historical data ends.
Click the "Hide/Show Hotspots" button in the Visualization panel or press the H key to toggle the fire hotspot overlay on the map. The chart also displays hotspot counts on a secondary Y-axis (right side) showing the temporal relationship between fire activity and CO₂ concentrations.
The wind overlay displays animated wind patterns from ERA5 reanalysis data on top of the CO₂ map. Colored particles show wind direction and speed—red indicates stronger winds while lighter colors indicate weaker winds. Click the "Hide/Show Wind" button in the Visualization panel or press the W key to toggle the wind layer.
Note: Wind data is available only for historical periods (2015–2024). When you navigate to the prediction period, a notification will appear indicating "No Wind Data for Prediction Period." This helps you understand meteorological influences on CO₂ distribution in the observed data.
YoY Change (Year-over-Year) shows the difference in CO₂ levels compared to the same month one year earlier.
Avg Growth (5yr) displays the average annual change over the past five years, helping identify long-term trends.
Avg Growth (All) shows the average annual growth rate since 2015, providing a comprehensive view of the overall trend from the baseline year.
Darker (red) regions have higher CO₂ concentrations, while lighter (green) regions have lower concentrations. Factors affecting CO₂ levels include industrial activity, population density, forest cover, and local geography.
Historical data shows concentrations ranging from approximately 392-429 ppm, while prediction data predicts a range of 418-433 ppm. These values are measured in the atmosphere and reflect both local emissions and global atmospheric CO₂ levels.
The AI Chat Assistant is here to help answering questions about CO₂ concentrations, fire hotspots, and data trends. Click the chat button (💬) to open the chat panel.
You can ask about specific cities/provinces, compare locations, analyze growth trends, and explore relationships between CO₂ levels and fire hotspots. Examples: "CO₂ growth in Jakarta 2020-2024", "Compare Surabaya vs Bandung", "Relationship between hotspots and CO₂", "Highest CO₂ province in 2023".
The AI recognizes Indonesian cities, provinces, and common abbreviations (e.g., "Jakarta", "Jabar" for Jawa Barat, "Kalbar" for Kalimantan Barat). It can also understand both Indonesian and English month names and years.
If the question is too complex or outside the scope of the available data, the AI will provide the best possible answer based on the loaded CO₂ and hotspot information. For very specific technical questions, you may need to consult the original data sources.
Click the EN/ID button at the top of the chat panel to toggle between English and Indonesian. The AI will respond in your selected language, and the chat interface will display in that language as well.
The Ground Data button links to ground-based CO₂ measurement stations and observatories across Indonesia. These provide direct atmospheric measurements that complement satellite observations, offering validation and additional context for the CO₂ data.
The Modeling button opens the Numerical Modeling panel. It includes Forward Modeling (live CAMS forecast grid with playback, lead-time slider, opacity control, and fixed color legends) and an Inverse Modeling slot labeled "Coming Soon." Use it to explore forecast concentrations alongside the existing Random Forest hybrid predictions shown on the main map.
The OCO2 MIP Fluxes button displays the Orbiting Carbon Observatory-2 (OCO-2) model intercomparison project (MIP) data, showing top-down CO₂ budget estimates at a municipal level. You can select different assimilation scenarios (In situ, OCO-2 Land, combined approaches), flux variables (carbon loss, net biosphere exchange, net carbon exchange), and view annual or mean data. This helps understand how satellite data improves carbon flux estimates.
Forward Modeling visualizes 3-hour CAMS forecast grids (≈0.5° resolution) for CO₂, CH₄, and CO over Indonesia (map view focused on 90°E–145°E, 15°S–15°N). You can play/pause the time steps, scrub the lead-time slider to specific forecast hours, and adjust opacity. Legends are fixed for consistency: CO₂ 380–550 ppm, CH₄ 1700–2200 ppb, CO 0–1200 ppb, all shown with continuous gradients instead of categories.
CAMS forecast outputs are converted into a compact JSON grid (≈34 MB) to keep the browser responsive compared to raw CSV files (~2 GB). Values are normalized to consistent units (CO₂ in ppm; CH₄ and CO in ppb) and gridded boxes are rendered instead of individual dots to give smooth coverage. The original spatial extent is preserved while the Leaflet view zooms to Indonesia for quick navigation.
Historical CO₂ data from satellites is processed and made available with a typical latency of several weeks to months, depending on satellite pass availability and processing delays. Prediction data is periodically updated to reflect new model training runs. Check the data source documentation for the most current update schedules.
The visualization panel provides several overlay options to enhance your analysis:
Spacebar — Play/Pause the timeline animation
H — Toggle fire hotspot overlay on/off
W — Toggle wind overlay on/off
P — Toggle peatland areas on/off
M — Toggle map theme (Light/Dark)
C — Toggle color scale mode (Dynamic/Global)
? — Open this FAQ modal
This application works best in modern browsers (Chrome, Firefox, Safari, Edge) released within the last 2 years. The interactive map and charting features require JavaScript to be enabled. For the best experience, ensure your browser has sufficient memory and a stable internet connection.
The data is embedded in the HTML file, allowing the application to run entirely in your browser without relying on external servers for data queries. This provides fast performance and offline functionality. Maps and visualizations are rendered using Leaflet and Chart.js libraries.
Animating the entire map with a large number of regions can be computationally intensive. If playback is slow, try zooming into a specific region to reduce the number of visible features, or use a more powerful device. The dynamic color scaling helps by focusing on visible data ranges.
Try refreshing the page. If the issue persists, clear your browser cache and reload. Ensure JavaScript is enabled and that you're using a supported browser. Check your internet connection, as map tiles may need to be downloaded from Leaflet's servers.
Ensure you have an active internet connection, as the AI feature requires communication with the Application Programming Interface (API). If the error persists, the API service may be temporarily unavailable. Try again in a few moments. Check that your question is clear and includes relevant location or time details for better answers.
This typically occurs when satellite coverage is limited for a particular region or date. Equatorial regions like Indonesia may have cloud cover or orbital gaps that result in missing observations. Prediction data fills forward-looking gaps based on model predictions. Check adjacent dates for more complete coverage.
Sudden spikes or drops in CO₂ levels can be caused by extreme weather events, seasonal fires, industrial emissions, or data processing artifacts. Use the hotspot overlay to identify whether fires correlate with CO₂ increases. Conversely, unexpected decreases might indicate lower human activity or seasonal patterns. The AI chat can help analyze these anomalies.
Ensure that you have permission to access the data and that your browser allows file downloads. Check your browser's download settings and security restrictions. The exported file is a CSV format that can be opened in Excel, Google Sheets, or any text editor. If the download fails silently, try a different browser or clear temporary files.
Ground-based stations are automated atmospheric monitoring observatories that measure CO, CO₂, and CH₄ concentrations directly at the surface using high-precision instruments. Unlike satellites which measure column-averaged concentrations, ground stations provide point measurements at specific locations and serve as critical validation references for satellite and model data. This application includes four stations across Indonesia: Muaro Jambi (rainforest/agricultural), Sorong (coastal), Kemayoran (urban Jakarta), and Bukit Kototabang (high-altitude baseline station).
All ground stations employ Cavity Ringdown Spectroscopy (CRDS), an advanced optical technique that detects atmospheric trace gases with exceptional sensitivity and precision. The CRDS method uses laser light bouncing in a high-finesse optical cavity to measure gas concentrations by analyzing light absorption rates. This technique provides real-time, continuous measurements with minimal human intervention and excellent repeatability when calibrated against international standards. For detailed technical information, see the "Ground-based Data" tab in the Methods section.
Each station displays two types of patterns:
You can toggle between CO, CO₂, and CH₄ measurements for each pattern. All data includes statistical reference lines showing the normal range (±1 standard deviation) to help identify unusual concentrations.
Spikes in ground station data often represent REAL atmospheric phenomena, not measurement errors. Common causes include:
The system uses statistical analysis (IQR, Z-score methods) to identify high-confidence outliers for awareness, but retains all data to preserve scientific integrity.
CO: Measured in ppb (parts per billion). Background levels ~100-150 ppb; elevated during pollution events.
CO₂: Measured in ppm (parts per million). Background ~410-420 ppm globally; varies with season and location.
CH₄: Measured in ppb. Background ~1900 ppb; increases from wetlands, agriculture, and biomass burning.
Note: Bukit Kototabang data is already in ppb for CO and CH₄. Other stations (Muaro Jambi, Sorong, Kemayoran) data has been converted to ppb for consistency.
The mini map displays the station's geographic location with a blue pin marker. Click the mini map or use the zoom feature to center the main map on that station's location. This helps you explore regional satellite data around each ground measurement point.
Detailed CO₂ concentration trends and fire hotspot analysis
This chart shows CO₂ concentration trends over time. The blue line represents CO₂ levels, while the purple dashed line shows fire hotspot activity. Historical data is shown as a solid line, while prediction data appears as a dashed orange line.
This dataset combines satellite observations with machine learning predictions to provide both historical analysis and future predictions of atmospheric CO₂ concentrations across Indonesia. The methodology uses a Hybrid Statistical–Machine Learning Model, specifically a Random Forest-based Hybrid Time Series Predicting approach that combines the interpretability of linear models with the predictive power of Random Forest regression.
NASA OCO-2/OCO-3 Satellites: Level 2 (L2) total column CO₂ measurements from the Orbiting Carbon Observatory satellites. These instruments use spectrometers to measure reflected sunlight in near-infrared wavelengths, providing precise atmospheric CO₂ concentrations. The data is processed and downscaled to municipal (kabupaten/kota) level.
Monthly averaged reanalysis data from ECMWF's ERA5 dataset provides environmental context:
VIIRS/MODIS thermal anomaly detections aggregated to monthly municipal counts, indicating fire activity that contributes to CO₂ emissions.
The prediction pipeline uses a decomposition strategy:
For future predictions, environmental features are predicted using SARIMAX (Seasonal AutoRegressive Integrated Moving Average with eXogenous factors) models with order (1,1,1) and seasonal order (0,1,1,12). These predicted features are then combined with climatological spatial patterns to generate spatially-resolved predictions.
The model produces gridded outputs that are regridded from ERA5 resolution to match the TCO2 satellite grid. Municipal-level values are extracted by spatial aggregation over each kabupaten/kota polygon using Indonesia's 514-municipality administrative boundaries.
SHAP (SHapley Additive exPlanations) values quantify each feature's contribution to the model predictions. This provides interpretability for the Random Forest model.
Mean absolute SHAP values showing overall feature importance
SHAP summary plot showing feature effects on predictions
The model is validated by comparing hindcast predictions (model predictions for historical periods) against actual observations in Bukit Kototabang, West Sumatra. This provides metrics like RMSE and correlation coefficient to assess model skill.
Scatter plot comparing hindcast predictions to observations
Time series showing fitted historical data and future predictions
Ground-based measurements provide in-situ (direct) observations of atmospheric gas concentrations at surface level. These stations measure CO, CO₂, and CH₄ using automated instrumentation at four key locations across Indonesia and Southeast Asia:
CRDS is a high-precision optical absorption technique that measures atmospheric gas concentrations by analyzing light absorption in a stable optical cavity. The method provides exceptional sensitivity and accuracy for trace gas detection.
| Gas | Unit | Precision | Temporal Resolution |
|---|---|---|---|
| CO (Carbon Monoxide) | ppb | ±2-5% | Hourly |
| CO₂ (Carbon Dioxide) | ppm | ±0.1-0.3 ppm | Hourly |
| CH₄ (Methane) | ppb | ±3-5% | Hourly |
All ground station data undergoes rigorous quality control:
Diurnal Cycles: Concentration variations throughout the day reflect atmospheric boundary layer dynamics, local pollution sources, and meteorological conditions. Morning/evening peaks in CO are typical due to boundary layer collapse.
Biomass Burning Events: Regional fires (especially dry season, June-October) can increase CO by 5-10× background levels. These are REAL atmospheric phenomena, not measurement errors.
Urban vs. Baseline: Muaro Jambi, Sorong, and Kemayoran show elevated concentrations due to local/regional sources. Bukit Kototabang, elevated at 864 m, captures background tropical conditions representative of the free troposphere.
The Copernicus Atmosphere Monitoring Service (CAMS) provides global greenhouse gas forecasts and analyses with 3-hourly temporal resolution. CAMS forecasts use advanced data assimilation techniques to combine satellite observations with state-of-the-art numerical weather prediction models, delivering up-to-date estimates of atmospheric CO₂, CH₄, and other trace gases.
European Centre for Medium-Range Weather Forecasts (ECMWF) operates CAMS on behalf of the European Commission. CAMS integrates observations from multiple satellites, ground stations, and aircraft measurements to constrain atmospheric composition estimates.
Forecast Range: 5 days ahead of the latest available initial date with 3-hourly output intervals. This high temporal resolution captures diurnal variations in atmospheric composition driven by boundary layer dynamics, solar radiation, and chemical reactions.
Global coverage at 0.1° × 0.1° horizontal resolution (~11 km at the equator) with 60 vertical levels from surface to 0.1 hPa (stratosphere). Data for Indonesia is extracted and regridded to match the monitoring network's spatial domain.
CAMS uses a 4D-Var assimilation system (Integrated Forecast System - IFS) that ingests observations from:
CAMS forecasts are issued daily. Each forecast provides predictions for the next 5 days at 3-hourly intervals. New satellite observations are incorporated with a latency of approximately 6-12 hours after observation time, ensuring timely updates while awaiting data transmission and processing.
The OCO-2 Model Intercomparison Project (MIP) provides top-down CO₂ flux estimates derived from OCO-2 satellite observations combined with inverse modeling. These products represent carbon budget components at high spatial resolution, showing net ecosystem production and carbon losses by assimilating satellite CO₂ measurements into chemical transport models and optimization frameworks.
The OCO-2 MIP Top-down CO₂ Budget v10 is a global gridded product derived from multiple inverse models assimilating OCO-2 satellite CO₂ column measurements. Each model solves the inverse problem to estimate surface CO₂ fluxes that would produce the observed atmospheric CO₂ concentrations.
Multiple assimilation configurations are included to assess sensitivity to model constraints:
Annual aggregated values for 2015-2020 with monthly intermediate data available. The product captures inter-annual variations in CO₂ fluxes driven by climate variability, landuse changes, and ecosystem productivity.
Three complementary carbon budget variables are provided:
Net carbon loss from a region, typically from biomass burning, deforestation, and oxidation of organic matter. Positive values indicate net carbon release to the atmosphere. Units: grams of carbon per meter² per year (gC/m²/yr).
Net exchange of CO₂ between land surface and atmosphere, representing the balance of gross primary productivity (photosynthesis) minus respiration and disturbances. Negative NBE indicates net carbon uptake; positive indicates net release. Units: gC/m²/yr.
Combines biospheric fluxes (NBE) with anthropogenic emissions from fossil fuel combustion and industrial processes. This is the total anthropogenic + natural net carbon flux that explains observed atmospheric CO₂ changes. Units: gC/m²/yr.
The OCO-2 MIP employs inverse modeling to estimate surface CO₂ fluxes from satellite observations. Unlike bottom-up approaches that calculate emissions from activity data, top-down methods work backwards from atmospheric observations to solve for the surface fluxes that would produce those observed CO₂ concentrations.
The core mathematical problem minimizes the difference between modeled and observed atmospheric CO₂ while maintaining flux estimates close to prior values:
Where: C = atmospheric CO₂ concentrations, F = surface fluxes, R = observation error covariance, B = prior uncertainty
Multiple participating models provide ensemble estimates, allowing quantification of model uncertainty. Sensitivity tests with different assimilation configurations (IS, LNLG, LNLGIS, LNLGOGIS) reveal how results change with different constraint assumptions. Posterior error covariance matrices provide pixel-level uncertainty estimates.
The inverse model adjusts prior flux estimates using a Bayesian optimization framework to match observed OCO-2 atmospheric CO₂ columns. The cost function minimizes differences between simulated and observed CO₂ while respecting prior uncertainty ranges:
Forward simulations use GEOS-Chem or TM5 chemical transport models to compute how surface CO₂ fluxes affect atmospheric concentrations observable by OCO-2. Meteorological fields from MERRA-2 or ERA5 drive atmospheric transport.
Global gridded data is extracted for the Indonesian region (6°N to 11°S, 95°E to 141°E), producing 3,255 grid cells covering land and surrounding oceans. Data is masked to land-only where land-focused analysis is preferred.
Grid cell values are aggregated to 514 Indonesian municipalities (kabupaten/kota) using nearest-neighbor assignment: each municipality centroid is matched to the geographically closest 0.1° grid cell. This preserves fine-scale spatial variability while providing policy-relevant regional estimates. Average fluxes per administrative unit are calculated for visualization and analysis.
The Indonesian archipelago is divided into 7 major regions for detailed analysis. The regional breakdown charts show Net Carbon Exchange (NCE) distribution by island, revealing which areas contribute most to Indonesia's total carbon emissions:
Figure: Four-panel regional analysis showing NCE distribution across 7 Indonesian islands (Sumatra, Java, Kalimantan, Sulawesi, Papua, Bali & Nusa Tenggara, Maluku)
To understand Indonesia's role in global carbon dynamics, OCO-2 MIP data is compared across major emitting regions worldwide. The global comparison charts contextualize Indonesia's carbon fluxes relative to other significant countries and regions:
Figure: Four-panel global comparison showing Indonesia (highlighted in red) among 9 countries and EU, all analyzed with OCO-2 MIP v10 satellite data
Muaro Jambi, Jambi
Located in Muaro Jambi on the lowland floodplain of the Batanghari River, an area characterized by alluvial soils, wetlands, and periodic flooding. The regional climate is tropical rainforest (Af), with annual rainfall of around 2,300 mm and consistently warm temperatures near 30–32 °C. Land cover around Muaro Jambi includes urban areas, agriculture, and remnants of peat and wetland ecosystems. The station also hosts a 100-m greenhouse-gas monitoring tower established by BMKG to support long-term climate observations.
Sorong, Papua Barat Daya
This station is situated in coastal Sorong, Papua Barat Daya. The region has an equatorial rainforest climate (Af) with 2,400–4,000 mm of rainfall annually and no true dry season. Surrounding ecosystems include extensive mangrove forests, which several studies identify as high-biomass, high-carbon-storage habitats. The station provides atmospheric composition measurements in a relatively clean, nature-dominated environment important for background observations.
Jakarta Pusat, Jakarta
Kemayoran sits within the highly urbanized core of Jakarta, a coastal megacity experiencing strong urban heat-island effects and rapid land-cover change. Jakarta's climate is tropical monsoon (Am), with a pronounced wet season from October to May and peak rainfall in January–February. Studies show that Kemayoran frequently records higher temperatures than suburban stations, reflecting urban warming. Nearby coastal zones, including Jakarta's mangrove areas, remain ecologically important but are under pressure from development and land conversion.
Agam, Sumatera Barat
Bukit Kototabang is located at 864 m elevation in West Sumatra, positioned on the equatorial belt where atmospheric conditions are influenced by monsoon circulation and inter-tropical convergence. The site experiences a tropical equatorial climate (Af) with high annual precipitation and minimal temperature variation. Topographically elevated above the coastal plains, the observatory captures atmospheric composition representative of background tropical conditions while occasionally recording transboundary air masses from regional biomass burning events (especially during the dry season). As a global baseline station, Bukit Kototabang provides critical reference measurements for atmospheric trace gases in Southeast Asia. It also operates the first 100-m greenhouse-gas monitoring tower in Indonesia, established by BMKG.