SD-GCM V2.1 Tutorial

Open PDF Download PDF Formulas → Back to SD-GCM

---

1. Software Overview

SD-GCM (Statistical Downscaling — General Circulation Model) is a desktop tool for climate scientists, hydrologists, and engineers who need to bring raw GCM output down to the local scale of observed weather stations. The core problem it solves: GCMs simulate the global climate on a coarse grid (typically 50–300 km), but impact studies (flood modelling, crop simulation, water resource planning) need station-scale data. SD-GCM applies statistical transfer functions that learn the relationship between GCM output and observed station data during a historical calibration period, then apply that relationship to the future GCM projection to produce a bias-corrected, downscaled time series.

1.1 The Typical Workflow

1. Load your observed station data (Station Data tab) — daily, monthly, or 3-hourly time series from one or more weather stations.

2. Load GCM data (Gridded Data tab) — a historical GCM run and a future SSP/RCP scenario run, either as NetCDF files, Excel/CSV, or downloaded live from Copernicus CDS.

3. Evaluate methods (Evaluation tab) — split the historical period into calibration and evaluation sub-periods, apply each method, and compare performance metrics to identify the best approach for your data.

4. Apply bias correction to the full future period (Bias Correction tab) — run the chosen method on the future GCM scenario to produce the final downscaled output.

5. Analyse outputs (Distributions and Data Counter tabs) — fit probability distributions, plot CDFs, count threshold events.

1.2 The Main Window Toolbar

The toolbar runs across the top of the main window and is always visible regardless of which tab is active.

1.3 The Six Main Tabs

Six tabs span the top of the main content area. They unlock progressively: Gridded Data becomes usable after station data is loaded, Evaluation and Bias Correction activate after both station and GCM data are loaded, and so on.

2. Setting Up Your Station List (GetData Window)

Before loading any data files, SD-GCM needs to know the names and geographic coordinates of your weather stations. This is done through the GetData window, which opens when you click "Fill List by File" in the Station Data tab's left panel. You can also populate the list manually by typing directly into the List Of Stations table in the Station Data tab.

2.1 The GetData Window

The GetData window reads a text or CSV file that defines your stations. Each row in the file represents one station with columns for name, latitude, and longitude.

3. Station Data Tab

The Station Data tab is always the first tab you work with. Its purpose is to load your observed historical weather data — the ground truth that the bias-correction methods use as their calibration target. This data typically spans 20–50 years of daily, monthly, or sub-daily records from one or more ground-based weather stations.

3.1 Left Panel — Fill Station List By File

This panel lets you load the station list (names and coordinates) from a delimited text file. Once the list is loaded here, it persists for the entire session and drives the column assignment in the data-loading panel on the right.

3.2 Left Panel — List Of Stations

After loading, this table shows all stations currently registered. Each row has three columns: Name, Latitude, and Longitude. You can scroll down if you have more stations than are visible.

3.3 Right Panel — Load Data From File

This is the main data-loading panel. It reads your observed time-series data file and maps its columns to the station names in the list.

3.4 The Station Data Grid

After selecting a file, a data grid appears in the lower-right area. The top row of the grid shows column assignment dropdowns — one per column. Use these to tell the tool what each column represents: Date, or the name of a specific station. Set unused columns to None((if you have selected before).

4. Gridded Data Tab

The Gridded Data tab is where you load the GCM output. You need two separate GCM datasets: one covering the historical calibration period (same era as your station observations) and one covering the future projection period under an SSP or RCP scenario. The tab also contains the Unit Converter, which handles the mismatch between GCM native units and the units of your station data. If you load only historical data so, you will have only the Evaluation tab activated.

4.1 Input Method

Three radio buttons at the top let you choose how GCM data enters the tool. Only one method is active at a time.

4.2 Load Historical Data

This section loads the GCM historical simulation — the model's representation of past climate, used to build the bias-correction transfer function. It must overlap with your observation period.

4.2.1 Settings Sub-Panel

4.2.2 Loading Sub-Panel

4.3 Load RCPs/SSPs Data

This section is identical in layout to Load Historical Data but loads the future scenario simulation. All the same fields apply. The variable name and calendar should match the historical file exactly. The time range should cover your intended projection period (e.g. 2020–2100 for a full SSP scenario).

4.4 Currently Loaded Status Bar

At the bottom of the main panel, two read-only status lines confirm what has been loaded:

4.5 Unit Converter (Right Panel)

GCM NetCDF files store variables in SI units that often differ from the units used in station observations. For example, temperature is stored in Kelvin (K) in CMIP6 but your station data is in degrees Celsius. Precipitation flux is stored in kg m⁻² s⁻¹ but your station data is in mm/day. The Unit Converter applies a mathematical transformation to every GCM value before it enters any bias-correction calculation.

5. Download CMIP6 Window (Copernicus CDS)

The Download CMIP6 window provides direct access to CMIP6 data from the Copernicus Climate Data Store (CDS), operated by ECMWF on behalf of the European Union. It eliminates the need to manually search, download, and preprocess multi-gigabyte NetCDF files: the tool downloads only the nearest grid cell time series for each of your stations. Open this window by selecting Method 2 in the Gridded Data tab and clicking "Download CMIP6 (OPeNDAP)".

5.1 Copernicus CDS Tab

5.1.1 CDS API Key Section

The CDS API uses a Personal Access Token for authentication. You need a free ECMWF account to obtain one. The key is saved securely in the Windows Credential Manager so you only need to enter it once.

5.1.2 Request Parameters Section

5.1.3 Extraction Target Section

5.1.4 Status / Error Log

The large text area at the bottom provides a real-time log of everything the tool is doing. It shows timestamps, bounding box coordinates, job submission URLs, job status polling updates, download status, and any error messages. This is the primary diagnostic tool when something goes wrong.

5.2 Downloaded Data Preview Tab

After a successful download, switch to the Downloaded Data Preview tab to verify the extracted values before using them in analysis.

6. Evaluation Tab

The Evaluation tab is where you determine which bias-correction method performs best for your specific combination of GCM and station data. The key principle is cross-validation: you withhold part of the historical record (the evaluation period) from calibration, apply each method trained only on the calibration period, and compare the corrected output to the actual observations during the withheld period. This gives an honest estimate of how the method will perform when applied to the future.

6.1 Evaluation Downscaling GroupBox

6.1.1 Calibration & Evaluation Periods

These six dropdowns define three non-overlapping (or overlapping, but logically distinct) time periods drawn from the historical GCM data.

6.1.2 Downscaling Method Radio Buttons

Six radio buttons let you select which method to evaluate. Only one method runs per click of the Evaluate button.

6.1.3 Monthly Stratification Checkbox

Monthly Stratification is a wrapper that applies the selected method independently to each of the 12 calendar months. When checked, the method is run 12 separate times: once for January data, once for February data, etc. The resulting method name gets an _M suffix (e.g. Delta_M, QDM_M, SDM_M).

Use Monthly Stratification when the GCM bias varies by season — for example, a model that overestimates summer precipitation but underestimates winter precipitation. Without stratification, a single transfer function would partially correct one season while worsening the other.

6.1.4 Action Buttons and Progress

6.2 Plot Panel

The chart below the method selector shows a visual comparison of the corrected output (orange, labelled _D for "downscaled") against the observed station data (green, labelled _O for "observed") over the evaluation period.

6.3 Method Comparison Table

Each time you click "Add to Comparison", a new row is appended to this table showing the performance of that method run.

7. Bias Correction Tab

The Bias Correction tab applies the method you selected in the Evaluation tab to the full future SSP/RCP GCM dataset, producing your final downscaled output. Unlike the Evaluation tab (which tests on historical data), the Bias Correction tab trains on the complete historical calibration period and projects onto the future period.

7.1 Left Panel — Evaluation Downscaling

7.1.1 Downscaling Period

7.1.2 Downscaling Method

The same six method radio buttons as in the Evaluation tab. Select the method you determined was best in the Evaluation step.

7.1.3 Monthly Stratification Checkbox

Same as in Evaluation. When checked, the method runs separately for each calendar month. The output is labelled with the _M suffix in chart titles and saved file names.

7.1.4 Action Buttons

7.2 Right Panel — Chart and Controls

7.2.1 Station and Scale Selectors

7.2.2 Plot Type Radio Buttons

8. Distributions Tab

The Distributions tab provides two tools for probability distribution analysis. The upper section ("Find Distributions") opens the Distribution Fitting window to rank candidate distributions against your data using formal goodness-of-fit tests. The lower section ("CDF/PDF Plot") draws empirical and fitted CDF or PDF curves directly in the main window for visual comparison across stations and periods.

8.1 Find Distributions Section

8.2 CDF/PDF Plot Section

9. Distribution Fitting Window

The Distribution Fitting window opens when you click "Check 12 Distributions" in the Distributions tab. It automatically fits a comprehensive set of continuous probability distributions to your data and ranks them by the Anderson-Darling goodness-of-fit statistic — the most appropriate criterion for hydro-climate data because it gives greater weight to distribution tails, where extreme events occur. The window has two tabs.

9.1 Goodness-of-Fit Rankings Tab

9.1.1 The Info Bar

The orange/amber banner at the top explains the ranking criterion: "Ranked by Anderson-Darling statistic (lower = better fit). Anderson-Darling is preferred for hydro-climate data: it weights distribution tails more heavily than KS, making it sensitive to extremes (floods, droughts). Chi-Square depends on arbitrary binning and is least reliable for continuous data." This is read-only scientific context.

9.1.2 The Rankings Table

Click any row to automatically switch to the Best Fit Chart tab and display that distribution's chart, making it quick to visually inspect any distribution in the table.

9.2 Best Fit Chart Tab

This tab shows a visual representation of a single chosen distribution fitted to the data, either as a CDF comparison or a PDF + histogram overlay.

10. Data Counter Tab

The Data Counter tab counts how many time steps in any loaded dataset satisfy a threshold condition, broken down by year or month. Common uses: counting wet days (days with precipitation above a threshold), counting extreme heat days (days above 35°C), counting drought days (days with precipitation below 1 mm), or counting days within a range. Results are shown both as raw counts and as percentages.

10.1 Counter Panel (Left)

10.1.1 Station Selector

10.1.2 Data Source

The Year Range Filter (below) automatically updates its year range to match the selected data source — selecting Observed populates the filter with observation years; selecting Future SSP/RCP shows the projection years.

10.1.3 Condition

Select one condition and fill in the required value(s). Exactly one radio button must be active.

10.1.4 Time Scale

10.1.5 Year Range Filter

10.1.6 Buttons and Progress

10.2 Results Panel (Right)

11. Complete Worked Example

This section walks through a full analysis from scratch using the dataset visible in the screenshots: daily precipitation for two stations (station1 at 35.19°N, 113.53°E and station2 at 34.48°N, 111.12°E) from 1993 to 2035.

Step 1 — Prepare your files

• Observed station data: a CSV file with columns Date, station1, station2 containing daily precipitation in mm, 1993-01-01 to 2012-12-31 (20 years).

• Historical GCM file: a NetCDF file from MIROC6 historical experiment covering 1993–2012, variable pr (precipitation flux, kg m⁻² s⁻¹).

• Future GCM file: a NetCDF file from MIROC6 SSP2-4.5 experiment covering 2020–2035, variable pr.

Step 2 — Load the station list

6. Open the Station Data tab.

7. In the "Fill Station List By File" panel, enter comma (,) as the delimiter.

8. Click "Fill List by File" and select your station list CSV.

9. In the GetData window, confirm First Row Is Header is checked, set column assignments (Name of Station, Latitude, Longitude), and click Load Data.

10. Verify the List Of Stations shows station1 (35.19, 113.53) and station2 (34.48, 111.12).

Step 3 — Load observed data

11. In the "Load Data From File" panel, set delimiter to comma.

12. Set Variable to "Rainfall".

13. Select "Daily data".

14. Leave Date format as "AutoDetect" (the dates are in M/D/YYYY H:MM:SS AM format, which AutoDetect handles).

15. Click "Select Data File" and choose your observation CSV.

16. In the column dropdowns at the top of the grid, assign: column 1 (Date) → Date; column 3 → station1; column 4 → station2.

17. Click "Load Station Data". The grid fills with daily values.

Step 4 — Load GCM data

18. Switch to the Gridded Data tab.

19. Select Method 3 — NetCDF Files.

20. In Load Historical Data → Settings: click Select File and choose the historical NetCDF. Set variable names: Time="time", Latitude="lat", Longitude="lon". Set Time Units="days" and Since="1850-01-01". Set Calendar="360_day" (check the file's metadata with ncdump -h).

21. In Load Historical Data → Loading: type "MIROC6" in GCM Model and "historical" in Scenario. Select "pr" from the Variable dropdown.

22. Click Load Historical. Wait for the progress bar to complete.

23. Repeat the same steps for Load RCPs/SSPs Data using the SSP5-8.5 NetCDF file.

24. In the Unit Converter: select "Flux → mm/day" (converts kg m⁻² s⁻¹ to mm/day). Set Correction Factor to Multiplicative. Click Save Settings.

Step 5 — Evaluate methods

25. Switch to the Evaluation tab.

26. Set Observation Period to 1993–2012, Historical GCM Period to 1993–2002, Evaluation Period to 2003–2012.

27. For each of the six methods (and optionally with Monthly Stratification checked), click Evaluate then Add to Comparison.

28. Review the Method Comparison table. Look for the row with the best KGE and NSE on the 2003–2012 evaluation period.

29. In this example, SDM_M (SDM with Monthly Stratification) shows KGE=0.575, NSE=0.196, WDF=1.047 on the evaluation period — a good result for daily precipitation.

30. Click Save Results to export the comparison table.

Step 6 — Apply bias correction

31. Switch to the Bias Correction tab.

32. Set Station Period to 1993–2012, Historical Period to 1993–2012, Downscale Period to 2020–2035.

33. Select SDM, check Monthly Stratification.

34. Click Downscale. The chart shows the corrected projection (green) against the raw GCM (red dashed) and observed (blue).

35. Check the Time Series chart: the green corrected series should show more rainfall amplitude than the raw GCM near-zero values.

36. Click Save DownScaled Data to export the final corrected daily time series.

Step 7 — Analyse the output

37. Switch to the Distributions tab. Set Use Monthly Data and Use Downscaled Data. Click Check 12 Distributions for station1.

38. In the Distribution Fitting window, identify the best-fitting distribution from the Rankings tab (lowest AD statistic that is not 999).

39. Switch to the Data Counter tab. Select Bias-Corrected, Between condition with values 2 and 10, Count per Year, years 2020–2035.

40. Click Count. Review how the number of moderate-rainfall days changes across the projection period.

41. Click Save to Excel to record the results.

12. Common Issues and Tips

12.1 Units Mismatch

Symptom: NSE is very negative (−10 or worse), the corrected time series is orders of magnitude off from the observed.

Cause: the Unit Converter was not set or was set incorrectly.

Fix: return to the Gridded Data tab, check the raw GCM values in View DB (e.g. if daily precipitation shows values around 0.0001, it is in kg m⁻² s⁻¹ and needs ×86400). Set the correct conversion and click Save Settings, then re-run Evaluation.

12.2 Dates Not Parsing Correctly

Symptom: date column shows wrong years, or evaluation period dropdowns are populated with unexpected values.

Cause: AutoDetect selected the wrong date format, or your Excel file stores years as numeric integers (e.g. 1993.0).

Fix: in the Station Data tab, change the Date format dropdown from AutoDetect to the explicit format that matches your file (e.g. "yyyy" for year-only columns). The tool handles the integer-to-date conversion automatically when a format is explicitly specified.

12.3 CDS Download Fails with HTTP 403

Cause: you have not accepted the CMIP6 Terms of Use on the CDS website.

Fix: log into cds.climate.copernicus.eu, search for "CMIP6 climate projections", open the Download Data tab, scroll to the bottom, and accept the licence. Then retry.

12.4 CDS Download Fails with HTTP 401

Cause: the API key is wrong, expired, or in the old UID:APIKEY format.

Fix: go to cds.climate.copernicus.eu/profile and copy your current Personal Access Token. It is a plain UUID with no colon. Paste it into the API Key field and click Save key.

12.5 Distribution Fitting Shows All Rank 999

Cause: data contains NaN, Inf, negative values, or zeros that prevent distribution fitting.

Fix: in the Distributions tab, check "Remove Values Less Than" and set the threshold to 0 (or a small positive number like 0.1 for precipitation). Then click Check 12 Distributions again.

12.6 KGE and NSE Are Poor on Evaluation

This is a sign of overfitting — the transfer function has memorised the calibration data rather than learning a generalisable correction. It is especially common when Monthly Stratification is used with short calibration periods (fewer than 10 years per month = fewer than 1 year per month).

Fix: either use a longer calibration period, or use the non-stratified version of the same method (e.g. switch from QDM_M to QDM).

12.7 Scientific Guidance: Which Method to Choose?

• For temperature: EQM_M or QDM_M. EQM_M is sufficient for most temperature applications. QDM_M is preferred if the study focuses on extreme heat events, since it preserves the warming signal at every quantile of the distribution.

• For precipitation: SDM_M or QDM_M. SDM_M explicitly corrects wet-day frequency (drizzle bias) and is recommended as the default. QDM_M is preferred when the study needs to preserve projected changes in heavy precipitation extremes.

• For wind speed or humidity with small bias: Delta_M is often adequate and produces the most interpretable correction (a simple seasonal scaling factor).

• Rule of thumb: if your Evaluation tab results show that a method performs well on the calibration period (same period rows) but poorly on the evaluation period (separate period rows), prefer the method with the smallest degradation between periods, not the one with the highest same-period score.

https://agrimetsoft.com/sd-gcm | https://www.youtube.com/AgriMetSoft

For support, video tutorials, and additional documentation visit our website.

---

Open PDF Download PDF Formulas → Back to SD-GCM