Soil Moisture Estimation Using GOES-VISSR Infrared Data: A Case Study with a Simple Statistical Method

Abstract

Five days of clear sky observations over Kansas and Nebraska are used to examine the statistical relationship between soil moisture and infrared surface temperature observations taken from a geosynchronous satellite. The approach relies on numerical model results to identify important variables other than soil moisture which have a significant effect on the surface temperature, and to define linear relationships between these variables and surface temperature. Linear regression is used to relate soil moisture to surface temperature and other variables that represent wind speed, vegetation cover, and low-level temperature advection. Results show good agreement between estimated and observed soil moisture features on each of the 5 days. The average coefficient of determination for five pseudo-independent tests in which the test day is held out of the regression is 0.71. When advection is neglected in these tests the average value of r2 drops to 0.57. It is shown that a depiction coefficient of 0.92, when used to compute antecedent precipitation index (API), produces the best correlation between API and soil moisture as interred from GOES thermal infrared data. By averaging daily predicted values over the 5-day rain-free case study period, 92% of the variance of the morning surface temperature change is explained by a simple multiple linear regression with all independent variables, or, alternatively, 85% of the observed variance in API is explained. It is concluded that this approach can distinguish at least four classes of soil wetness, but the necessity for measurement of surface advection may limit its usefulness in remote areas.