spatiao – temporal evaluation and comparison of mm5 model using similarity algorithm

introduction temporal and spatial change of meteorological and environmental variables is very important. these changes can be predicted by numerical prediction models over time and in different locations and can be provided as spatial zoning maps with interpolation methods such as geostatistics (16, 6). but these maps are comparable to each other as visual, qualitative and univariate for a limited number of maps (15). to resolve this problem the similarity algorithm is used. this algorithm is a simultaneous comparison method to a large number of data (18). numerical prediction models such as mm5 were used in different studies (10, 22, and 23). but a little research is done to compare the spatio-temporal similarity of the models with real data quantitatively. the purpose of this paper is to integrate geostatistical techniques with similarity algorithm to study the spatial and temporal mm5 model predicted results with real data. materials and methods the study area is north east of iran. 55 to 61 degrees of longitude and latitude is 30 to 38 degrees. monthly and annual temperature and precipitation actual data for the period of 1990-2010 was received from the meteorological agency and department of energy. mm5 model data, with a spatial resolution 0.5 × 0.5 degree were downloaded from the nasa website (5). gs+ and arcgis software were used to produce each variable map. we used multivariate methods co-kriging and kriging with an external drift by applying topography and height as a secondary variable via implementing digital elevation model. (6,12,14). then the standardize and similarity algorithms (9,11) was applied by programming in matlab software to each map grid point. the spatial and temporal similarities between data collections and model results were obtained by f values. these values are between 0 and 0.5 where the value below 0.2 indicates good similarity and above 0.5 shows very poor similarity. the results were plotted on maps by matlab software. results discussion in this study the similarity and geostatistical algorithm were combined to compare and evaluate spatio-temporal of predicted temperature and precipitation data by mm5 model with actual data. the analysis of the similarity map is based on the f values, the area and also the uniformity of distribution over the area. the similarity between predicted and actual data is higher when f values are low and distributed more uniform. the temperature similarity maps showed that f values are between 0.0 - 0.2 in cold seasons. it was shown that the values had spatial continuity and uniform distribution. a large part of area (almost 80%) is covered by lowest f value (f˂0.1), which shows very high similarity among temperature datasets. the highest values (0.15 0.3) in the central region indicate errors in the model predictions data. but generally prediction of model in both seasons for the temperature was good. in annual time scale, f values are between 0.0 - 0.25. the area of good similarity value (0.0˂f˂0.1) is almost 65% of the whole region with spatial continuity and uniform distribution. accuracy of the model declined from temperature of the cold season to annual and then warm season respectively. the precipitation similarity maps showed that in cold season f values changes between 0.05 - 0.4. these values had less spatial continuity than temperature. in more than half of the area (60%) there was fairly good similarity where 0.05 0.3) in the central mountainous area and south part of region suggests the low similarity in the model predictions. similarity between the cold seasons is much higher than the warm seasons, which is due to the variability of precipitation during the seasons. in the annual time scale, f values are between 0.05 - 0.3. f values (0.0˂f˂0.1) are almost 40% of the whole region with uniform distribution. overall, the higher uniform distribution of annual similarity values showed that prediction of model for annual precipitation data is better than seasonal. the maximum f values identified the areas with modeling error for various reasons. in this study the central and the southern parts had maximum f values at different time steps. plotted mean monthly values of similarity indicated minimum and maximum temperature f values were occurred in january and july while for precipitation was taken place in january and september respectively. this shows that mm5 model prediction was good in january. conclusion: in this paper, the similarity algorithm discovered spatial and temporal similarities between the predicted and actual data for temperature and precipitation variables. according to the obtained f values, the model predicts temperature was better than precipitation. due to the upward movement of the convective zone and the effects of topography for both variables, the similarity between predicted and actual data is low in warm seasons. in small areas of the south and the central region of the study area, f values are between 2.0 and 4.0, respectively, which could be considered as a weak similarity. the area with high f values (f > 0.45) can be seen on every precipitation map, which suggests a large error values related to reporting of the station data.