INVESTIGATE KRIGING VARIANCE AS A PROXY FOR ALS DEM ACCURACY

Airborne laser scanning (ALS) is widely adopted to acquire digital elevation models (DEMs) of the Earth's surface due to its ability to penetrate tree canopies. DEMs are derived by interpolating ground points acquired from ALS data. In this research, we selected Kriging as the interpolation method, which is popular for its consideration of best linear unbiased estimator (BLUE) and spatial correlation in the data. Kriging assumes data in the space have a certain degree of correlation. The spatial correlation is associated with the spatial distribution and distance of the data and is used to calculate both Kriging estimation and Kriging variance. Kriging estimation represents the DEM, while Kriging variance serves to assess the accuracy of Kriging estimator. For example, in thick canopies areas with limited ground points, the accuracy of Kriging estimation may decrease. In these cases, Kriging variance quantifies the accuracy of Kriging estimator, where a larger value indicates weaker spatial correlation and lower estimation accuracy, and vice versa. The reason why we focused on Kriging variance is that though we can use root mean square error (RMSE) to assess the accuracy of DEM. However, we can only get one value for whole site. If we want to know the accuracy of each grid, kriging variance is the best proxy. Our research encompasses 60 regions, including mountainous and hilly terrains, specifically selected to test the accuracy of void areas. Initially, we created voids of different sizes on mountain peaks, valleys, and slopes to simulate dense canopies areas. Subsequently, we calculated spatial correlation of data through semi-variogram model, with an emphasis on selecting the appropriate theoretical model and adjusting the range, which controls the consideration of spatial correlation within a given distance while disregarding those further away. In our study, range was adjusted based on void sizes to achieve better fit for the semi-variogram model. And the lag distance we used is 0.5 meter because the nominal ground point density is 2 points per meter square. After determining the spatial correlation of the data, we calculated the Kriging estimation and Kriging variance to assess the accuracy of DEM affected by different sizes and locations of voids. Then through cross validation, we calculated RMSE to validate the selection of Kriging parameters and evaluate the performance of it.