2.6.

Discussion on methods advantages and limitations

(Approx. 5 min reading incl. references)

This lecture demonstrates multiple LSP retrieval methods applied to original S2 and fused time series and highlights the potential to improve accuracy and address data gaps caused by e.g. cloud cover. However, it also notes the limitations of data fusion, including the potential for artefacts, redundancies, or noise that can challenge the effectiveness of certain approaches.

The exemplified use of the B-spline fitting function to S2- and fused VI time series highlights the dependency on weather-driven data availability and thus the potential biases of fitting and smoothing techniques. The selection of the LPS retrieval method and VI has a significant impact on the accuracy of the phenometrics in comparison to the references. As demonstrated in this lecture and in the practical part, increased data density does not necessarily lead to higher accuracy. It is therefore essential to select retrieval methods considering both the phenological stage and the characteristics of the fused data.

In general, following gains and limitations are formulated:

Advantages:

Improved accuracy: The use of data fusion, advanced remote sensing techniques, and ML algorithms can lead to more accurate LSP results.
Increased robustness: Methods that can handle variability in data and synthetic data generation can improve the reliability of results.
Better decision-making: Accurate and reliable phenometrics provide robust input for the crop model simulations and support better decision-making in agriculture.
Increased efficiency: The use of automated methods and data fusion can reduce the time and effort required for data analysis.

Limitations:

Data quality: The quality of the data used in a study can significantly impact the LSP accuracy.
Methodological limitations: The choice of method or algorithm can affect the results, and some methods may be more suitable for certain types of data or applications.
Scalability: Methods that are effective for small-scale applications may not be suitable for large-scale applications.
Accuracy assessment: Reference data availability is challenging. It emphasizes the need for investment in modernized and systematically maintained reference datasets.