Location: Location not imported yet.Title: Performance Metrics for Soil Moisture Retrievals and Applications Requirements) Author
Submitted to: Journal of Hydrometeorology
Publication Type: Peer reviewed journal
Publication Acceptance Date: 1/11/2010
Publication Date: 6/15/2010
Publication URL: http://handle.nal.usda.gov/10113/60040
Citation: Entekhabi, D., Reichle, R.H., Crow, W.T., Koster, R.D. 2010. Performance metrics for soil moisture retrievals and applications requirements. Journal of Hydrometeorology. 11:832-840. Interpretive Summary: Accurate information regarding the availability of soil water is valuable for a wide range of agricultural applications (including irrigation scheduling, fertilizer application optimization and crop yield forecasting). Such information can be obtained in a variety of ways and we currently lack appropriate metrics to summarize the value of various soil moisture estimates for various agricultural applications. This makes it difficult to determine whether a given set of observations are “good enough” in an absolute sense or “better” than an existing product in a relative sense. This article reviews existing skill metrics for soil moisture retrievals (such as simple root-mean-square-error accuracy and the correlation between predicted and actual soil moisture) and goes on to propose a new general error statistic which - for any pre-defined application - provides a more appropriate measure of the overall value for a given product. The creation of such a generalized metric is critical for current efforts to development and NASA Soil Moisture Active/Passive satellite mission.
Technical Abstract: Quadratic performance metrics such as root-mean-square error (RMSE) and time series correlation are often used to assess the accuracy of geophysical retrievals and true fields. These metrics are generally related; nevertheless each has advantages and disadvantages. In this study we explore the relation between these two metrics in the presence of biases in mean as well as amplitude of fluctuations (variance) between estimated and true fields. Such biases are typical, for example, for satellite retrievals of soil moisture. Finally we introduce levels of discrimination – a metric that is easier to pin-point for a given application - and relate it to the RMSE and correlation performance metrics.