Wednesday, November 7, 2007 - 11:45 AM
300-9
Predicting the Distribution of Mollic Soils using a Solar Radiation Modelling Approach.
Aspect angle is commonly used by soil scientists and ecologists as a simplistic, qualitative proxy for landscape-scale variation in microclimate caused by the slope and orientation of relief. While aspect angle serves as a useful surrogate for the underlying processes which create these microclimates (temperature and ET values related to solar geometry), it cannot be directly used in the creation of physically-based models of landscape-scale soil development. Furthermore, special numerical methods are required to include raw aspect angle (a value with intrinsic periodicity) into statistical models. Transformed aspect values exhibit extreme bi-modality and selection of a transformation if either region-specific or arbitrary.
Our objective was to develop a quantitative method for describing the variation in soil properties along estimated microclimate gradients, with solar radiation modeling, for use in current and future Soil Survey efforts in upland regions. Solar radiation models can numerically describe the components of a solar radiation budget at each cell of a digital elevation model for any given time interval, optionally accounting for seasonal variation in atmospheric properties along with the influence of local shading by adjacent terrain.
We used the ESRA model to estimate a solar energy budget at each grid cell of a 10-meter resolution DEM from Pinnacles National Monument, Ca. Parametrization of the clear-sky version of this model was accomplished with daily estimates of the Linke turbidity factor, using local pyranometer measurements (11 year record). Our estimated daily irradiance values matched the local weather station data with an R2 of 0.965 (n=365, p <2.2*10-16). Solar radiation thresholds partioning mollic and non-mollic soils were identified by stepping through the range of possible values and performing a binomial test of independence at each step. Next, solar radiation values coupled with a local geologic map were used as predictor variables in a logistic regression model constructed to predict the spatial distribution of upland soils with mollic epipedons within the park. 199 (of 277 collected) field observations were used to build the model, which had an average relative operating characteristic (ROC) area of 0.85 to 0.90. 100-fold cross-validation (repeated re-fitting of the model with a subset of observations) indicated a mean classification error rate of 27%. Field validation sites (n=35) showed an 80% PCC (percent correctly classified) value, highly significant kappa (0.57 - 0.67) and tau (0.58 - 0.78) statistics, and an ROC rating of 0.77 - 0.86. Total carbon and nitrogen (by Carlo-Erba) analysis on the validation sites revealed a good correlation between predicted mollic probability and Box-Cox transformed concentration values (R2 = 0.6 - 0.7). The solar radation method developed for this study out-performed the classical aspect-based aprroach, as well as an energy threshold method developed for this study.
Our objective was to develop a quantitative method for describing the variation in soil properties along estimated microclimate gradients, with solar radiation modeling, for use in current and future Soil Survey efforts in upland regions. Solar radiation models can numerically describe the components of a solar radiation budget at each cell of a digital elevation model for any given time interval, optionally accounting for seasonal variation in atmospheric properties along with the influence of local shading by adjacent terrain.
We used the ESRA model to estimate a solar energy budget at each grid cell of a 10-meter resolution DEM from Pinnacles National Monument, Ca. Parametrization of the clear-sky version of this model was accomplished with daily estimates of the Linke turbidity factor, using local pyranometer measurements (11 year record). Our estimated daily irradiance values matched the local weather station data with an R2 of 0.965 (n=365, p <2.2*10-16). Solar radiation thresholds partioning mollic and non-mollic soils were identified by stepping through the range of possible values and performing a binomial test of independence at each step. Next, solar radiation values coupled with a local geologic map were used as predictor variables in a logistic regression model constructed to predict the spatial distribution of upland soils with mollic epipedons within the park. 199 (of 277 collected) field observations were used to build the model, which had an average relative operating characteristic (ROC) area of 0.85 to 0.90. 100-fold cross-validation (repeated re-fitting of the model with a subset of observations) indicated a mean classification error rate of 27%. Field validation sites (n=35) showed an 80% PCC (percent correctly classified) value, highly significant kappa (0.57 - 0.67) and tau (0.58 - 0.78) statistics, and an ROC rating of 0.77 - 0.86. Total carbon and nitrogen (by Carlo-Erba) analysis on the validation sites revealed a good correlation between predicted mollic probability and Box-Cox transformed concentration values (R2 = 0.6 - 0.7). The solar radation method developed for this study out-performed the classical aspect-based aprroach, as well as an energy threshold method developed for this study.
New Soil Survey products including an estimated probability map of soils having a mollic epipedon, surface carbon and surface nitrogen percentages were produced. These methods can possibly be used to fill-in the currently un-mapped locations of soil components, identified within a Soil Survey manuscript. A comparison was done between the solar radiation approach and the conventional aspect classification approach.
Audio File