Abstract
Accounting for secondary exhaustive variables (such as elevation) in modelling the spatial distribution of precipitation can improve their estimate accuracy. However, elevation and precipitation data are associated with different support sizes and it is necessary to define methods to combine such different spatial data. The paper was aimed to compare block ordinary cokriging and block kriging with an external drift in estimating the annual precipitation using elevation as covariate. Block ordinary kriging was used as reference of a univariate geostatistical approach. In addition, the different support sizes associated with precipitation and elevation data were also taken into account. The study area was the Calabria region (southern Italy), which has a spatially variable Mediterranean climate because of its high orographic variability. Block kriging with elevation as external drift, compared to block ordinary kriging and block ordinary cokriging, was the most accurate approach for modelling the spatial distribution of annual mean precipitation. The three measures of accuracy (MAE, mean absolute error; RMSEP, root-mean-squared error of prediction; MRE, mean relative error) have the lowest values (MAE = 112.80 mm; RMSEP = 144.89 mm, and MRE = 0.11), whereas the goodness of prediction (G) has the highest value (75.67). The results clearly indicated that the use of an exhaustive secondary variable always improves the precipitation estimate, but in the case of areas with elevations below 120 m, block cokriging makes better use of secondary information in precipitation estimation than block kriging with external drift. At higher elevations, the opposite is always true: block kriging with external drift performs better than block cokriging. This approach takes into account the support size associated with precipitation and elevation data. Accounting for elevation allowed to obtain more detailed maps than using block ordinary kriging. However, block kriging with external drift produced a map with more local details than that of block ordinary cokriging because of the local re-evaluation of the linear regression of precipitation on block estimates.
Highlights
IntroductionIntroduction iationsAssessing the spatial distribution of precipitation is crucial for water resource management and, in particular, for facing the challenges of agriculture and food production [1].The accurate modelling of precipitation is a well-known topic and it essentially consists in predicting the precipitation over more or less areas, depending on the objectives of such modelling, from a few sparse measuring stations with good confidence [2,3,4].The use of remote sensing data to indirectly obtain comprehensive precipitation can be a feasible alternative to direct measurements, but the accuracy and resolution of precipitation may not be adequate for the intended use [5].Precipitation varies more or less continuously in the geographical space and is suitable to be modelled as an intrinsic stationary process by using the methods of geostatistics [3,6,7]
The results of the study showed that block kriging with external drift, compared to block ordinary kriging and block ordinary cokriging, was the most accurate approach for modelling the spatial distribution of mean annual precipitation
The results clearly indicated that the use of an exhaustive secondary variable always improves the precipitation estimate, in the case of areas with elevations below 120 m, block cokriging makes better use of secondary information in precipitation estimation than block kriging with external drift
Summary
Introduction iationsAssessing the spatial distribution of precipitation is crucial for water resource management and, in particular, for facing the challenges of agriculture and food production [1].The accurate modelling of precipitation is a well-known topic and it essentially consists in predicting the precipitation over more or less areas, depending on the objectives of such modelling, from a few sparse measuring stations with good confidence [2,3,4].The use of remote sensing data to indirectly obtain comprehensive precipitation can be a feasible alternative to direct measurements, but the accuracy and resolution of precipitation may not be adequate for the intended use [5].Precipitation varies more or less continuously in the geographical space and is suitable to be modelled as an intrinsic stationary process by using the methods of geostatistics [3,6,7]. The accurate modelling of precipitation is a well-known topic and it essentially consists in predicting the precipitation over more or less areas, depending on the objectives of such modelling, from a few sparse measuring stations with good confidence [2,3,4]. The use of remote sensing data to indirectly obtain comprehensive precipitation can be a feasible alternative to direct measurements, but the accuracy and resolution of precipitation may not be adequate for the intended use [5]. Precipitation varies more or less continuously in the geographical space and is suitable to be modelled as an intrinsic stationary process by using the methods of geostatistics [3,6,7]. Many different geostatistical methods have been developed for Licensee MDPI, Basel, Switzerland.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have