Why is Interpolation Required?
DAS receives monthly weather observations from New Zealand weather stations. However, due to factors beyond the vendor’s control, such as equipment malfunctions, network outages, or extreme weather events; approximately 5% to 15% of observations may be missing or contain incorrect values, also known as outliers.
To ensure a consistent and accurate supply of data to customers, these missing or incorrect values must be spatially interpolated using observed data and a range of historical precipitation datasets.
How Does Spatial Interpolation Work?
DAS uses an advanced algorithm called Kriging with External Drift (KED), which is widely used for spatial interpolation of precipitation at regional and national scales. This algorithm leverages historical precipitation data to detect rainfall distribution patterns across New Zealand and applies these insights to interpolate missing observations.
Step-by-Step Process:
1. Historical Climate Patterns
The Southern Oscillation Index (SOI) (NOAA) is used to analyze long-term rainfall distribution patterns.
2. Reference Data for Weather Patterns
Average annual rainfall datasets (1972-2013) from the NZ Ministry for the Environment (source) are classified into three categories:
El Niño
La Niña
Neutral
3. Selection of the Most Relevant Climate Pattern
The current SOI value determines which of these climate patterns was active during the previous month.
The corresponding rainfall raster dataset is used as an External Drift Surface in the KED model.
4. Running the KED Algorithm
The algorithm takes the available observed values and generates an interpolation surface to estimate missing data points.
5. Final Data Adjustment
The interpolated values replace only the missing or incorrect observations, ensuring data reliability while maintaining scientific accuracy.
📌 More scientific details can be found here: Regression-Kriging.
Limitations of KED Interpolation
While spatial interpolation is a robust solution for filling gaps in weather data, it does have limitations:
Reduced Accuracy Compared to Direct Observations
The Root Mean Square Error (RMSE) averages 53 over five years of test data.
Tendency Toward Long-Term Averages
When rapid precipitation changes occur, interpolated values may appear smoother and closer to historical monthly averages.
Accuracy Depends on the Amount of Missing Data
If a large percentage of observations is missing in a given month, the accuracy of the interpolation may decrease.
Key Takeaways
✅ Interpolation ensures a consistent supply of weather data when raw observations are missing or incorrect.
✅ The process incorporates long-term climate patterns and the current SOI index for improved accuracy.
✅ While interpolated values may slightly differ from real-time data, they remain within scientifically acceptable accuracy ranges.
✅ DAS ensures reliable weather data delivery through scientifically validated spatial interpolation methods.
By applying validated interpolation techniques, DAS ensures customers receive the most accurate and complete weather data possible, even when direct observations are unavailable.