|
 |
The purpose of the current work was the development of a model for the estimation of monthly and annual precipitation within the GVRD. This model facilitated the construction of continuous precipitation surfaces for monthly and annual precipitation covering the extent of the GVRD.
Topography is a critical factor in determining spatial pattern of precipitation. The combined effect of slope, aspect and elevation on precipitation is known as the orographic effect, which is the predominant factor in determining the patterns of rainfall observed in regions of complex terrain, such as the Greater Vancouver Regional District (GVRD). It is commonly known that precipitation levels vary substantially across the GVRD, largely in a north-south direction, but this topographic enhancement is inadequately represented in the official precipitation data. The official precipitation measurement for Vancouver is made at the airport in Richmond, which is also the location for which the standard forecast values are made.
Since there is a general trend between elevation and precipitation, it is logical to propose that the incorporation of elevation as a model input will improve the level of estimation error in the predicted surface. No interpolation method is universally superior – ‘optimal’ techniques are specific to a particular study area. To evaluate the relative importance of elevation and spatial distribution of station in the modeling of the precipitation surface, three interpolation models were tested – Inverse Distance Weighting (IDW), Ordinary Cokriging and Linear Regression. IDW is a local, exact interpolator that accounts for the spatial distribution of stations but not elevation. Ordinary Cokriging is another local, exact interpolator but it accounts for both the spatial distribution of stations and the effect of elevation. Linear Regression is an inexact, global interpolator that accounts for the effects of elevation but does not explicitly account for the spatial distribution of stations.
Elevation was filtered in order to better model the orographic effect. This resulted in a new variable that was a better expression of the orographic effect as well as being more highly correlated with precipitation. Increased correlation between precipitation and elevation will improve the utility of elevation as a model input for predicting precipitation.
The results of the three models were compared visually and statistically. An analysis of the root-mean-square errors (RMSE) for each interpolated surface revealed that the linear regression model performed poorly relative to the other two models. The other two models were relatively close in performance but inverse distance weighting on average performed slightly better according to such an analysis. The monthly and annual surfaces produced by this model are presented as a precipitation atlas of the Lower Mainland.
The web-based component of the project presented here will present these surfaces with the ability to return precise precipitation estimates at user-queried locations. Simple forecasting ability is also incorporated in this product.
 |
back to
top |
 |