I. What is Triple Collocation?

Triple Collocation (TC) is a robust statistical method for assessing errors among three independent data sources measuring the same variable. Its distinct advantage lies in not requiring knowledge of the true value, making it indispensable in applications such as remote sensing, hydrological observation, and meteorological data analysis. This method ensures unbiased error estimation and is particularly useful for validating measurement accuracy across diverse observation systems.

1.1 Working principle of the TC:

TC assumes that three independent datasets X1, X2, and X3 simultaneously measure the same variable (e.g., precipitation or river discharge). Each dataset contains inherent errors, which are assumed to be uncorrelated and unbiased. By analyzing the discrepancies among these datasets, TC estimates the error variance of each dataset without requiring knowledge of the true value or identifying the most accurate dataset. This makes TC particularly valuable in evaluating observational systems, such as reconciling satellite-based measurements with in-situ observations.

1.2 Application Scenarios:

TC is particularly useful when comparing multiple data sources. Typical applications include:

• Remote sensing vs. ground-based observations: TC is applied to reconcile differences between satellite observations, in-situ measurements, and model outputs. For instance, assessing the accuracy of precipitation estimates by comparing satellite data (e.g., GPM) with ground-based rain gauges and reanalysis products.
• Hydrological monitoring: In hydrology, TC helps evaluate error variances between model-simulated discharge, river gauge measurements, and independent datasets like those from the Global Runoff Data Centre (GRDC). This provides valuable insights into the reliability of hydrological models and observation networks.
• Meteorological observation: TC is widely used to compare atmospheric variables such as temperature and humidity from numerical weather prediction models, ground-based weather stations, and satellite data (e.g., MODIS or AIRS), ensuring a robust understanding of measurement uncertainty.

1.3 Data Requirements and Limitations:

While Triple Collocation (TC) is a powerful tool, it is subject to specific assumptions and data requirements that can limit its applicability:

• Independent errors: TC assumes that the errors of the three datasets are independent. Correlated errors can lead to inaccurate estimations.
• TC assumes that the errors in the three datasets are independent. When errors are correlated—such as when multiple datasets rely on the same underlying model or sensor—TC can produce biased error estimations. Pre-analysis to identify and minimize correlated errors is recommended.
• Sufficient data points: Accurate and stable error variance estimates require a large sample size. Insufficient data points can lead to statistical noise, reducing the reliability of results.
• Linear relationships: TC assumes linear relationships among the datasets, which may limit its effectiveness for highly nonlinear systems.

1.4 Interpretation of TC Results:

The primary output of Triple Collocation (TC) analysis is the error variance associated with each dataset, which reflects the reliability of the measurements. Key interpretations include:

• Large Error Variance: A dataset with a high error variance is less reliable, indicating significant measurement noise or inaccuracies.
• Similar Error Variances: Suggests comparable levels of accuracy across the sources.

1.5 Advantages of TC:

TC offers several distinct advantages, making it a powerful tool for error analysis across various fields:

• No need for true values: TC can assess the errors even when the actual “true value” is unknown.
• Unbiased error estimation: Estimates independently the error variance for each dataset, ensuring that the results are not influenced by biases present in other datasets.
• Robust to different data sources: Can handle datasets with various sources and error characteristics.

1.6 Disadvantages of TC:

While Triple Collocation (TC) is a valuable method for error estimation, it has certain limitations that should be considered:

• Assumption of independent errors: If the errors in the three datasets are correlated (e.g., multiple versions generated by the same model), the estimation will be biased and unreliable.
• Inability to Detect Systematic Bias: TC is designed to estimate random errors and does not account for systematic biases in the datasets. As a result, datasets with consistent over- or underestimations may appear to have acceptable error variances.