Modeling and Prediction of Land Condition for Fort Riley Military Installation

Abstract

Listen

Translate article

<abstract><title><italic>Abstract.</italic></title> In the U.S., the Department of Defense manages more than 5500 military installations that occupy approximately 12 million ha of land. These lands are used for various military training programs. Training activities inevitably degrade the land condition, and the degraded land condition, in turn, limits the land’s military training carrying capacity. To sustain the military training land carrying capacity and the environment, land managers must monitor and predict changes to the land condition under various military training schemes. The objective of this study is to develop prediction models for land condition based on military training intensity and on independent variables that play a significant role in driving land condition changes at Fort Riley, Kansas. It is assumed that land condition can be quantified using soil erosion as a surrogate measure, which is mainly determined by a ground and vegetation cover factor, in which the larger the factor, the poorer the land condition. In addition to military training intensity, the independent variables used in these prediction models of land condition included distance from the location to roads, terrain slope (which affects military training access), ground cover, landscape fragmentation (an indirect measure of military training induced disturbance), and spatial variability of canopy cover and military training induced disturbance (as reflected in Landsat Thematic Mapper [TM] images). Various regression models were developed, and predictions made by linear and nonlinear models were compared with and without TM images, with and without stepwise regression procedures, and with and without historical land condition variables. Results showed that the absolute Pearson product moment correlation coefficients of ground cover with the cover factor were larger than 0.63; the correlation was greatest and significant at a risk level of 5%. Ground cover was thus involved in all the stepwise regression and nonlinear models. Although military training intensity was significantly correlated with the cover factor, training intensity was excluded from the best models mainly because both ground cover and landscape fragmentation that existed in the models also reflected the military training induced disturbance. Compared to models in which all the variables were involved, the stepwise regression models reduced the number of the independent variables from 11 or 15 to 3 or 6 (depending on analysis year) with no significant loss of accuracy. In most cases, adding the near and middle infrared TM images, which revealed the spatial variability of military training induced disturbance, improved the prediction of land condition. Based on the correlation coefficient and root mean square error (RMSE) between the predicted and observed values of the cover factor, the nonlinear models that used significant independent variables led to more accurate predictions than the linear regression models. This suggests that the combination of stepwise regression and nonlinear models could increase the accuracy of prediction. Moreover, adding the historical land condition variables, such as historical cover factor and ground cover, into the models could greatly decrease prediction errors.