The description of biomass development in space and time remains an enigma because deterministic model input parameters can only be collected during limited times at a limited number of locations. Moreover, not all relevant processes during the development of an annual crop are captured in crop growth models. Furthermore, spatial patterns of crop yield are scarcely related to classical soil variables such as soil texture. Nevertheless, crop yield usually does not spatially fluctuate randomly around the average crop yield within a field but exhibits a distinct spatially structured pattern. Half a century ago, Kalman developed a filter algorithm that helped to separate noise from signal. This filtering technique first applied to time series data was later adapted to the stochastic description of hydrologic time series. In principle, the application of this filter provides the opportunity to quantify the variance around a predicted value. Additional measurement information provides the opportunity to reduce the prediction variance in an updating step in which observations are weighted against predictions. The more the measurement information integrates the processes affecting the main variable of interest, the better the prediction result for the remaining time series. In this study, we postulate a scenario where this filtering technique is applied in two different ways in order to reduce the prediction variance by combining observations and estimations in the spatial and temporal domain. Observations of biomass status during the vegetation season are taken in order to support a very simple scheme of biomass development in relation to air temperature and average degree of water saturation in the soil profile. The equation for the biomass development (based on a simple empirical sigmoidal curve suggested by Childs, 1975) allows for crop response to two environmental factors. On the one hand, crop growth is favored by optimum temperature and water supply, while on the other hand, it is delayed or reduced if air temperature or soil water prevail at suboptimal or excessive magnitudes. The equation is presented in its differentiated form with respect to time while coefficients in the equation represent simple responses to both air temperature and soil water. These coefficients are assumed to be known in initial scenarios, but can be estimated with increasing availability of data. Sensor observations of leaf area index and vegetation status are calibrated with respect to crop growth. Based on these observations, the biomass development is predicted and updated. The magnitude of the temporal prediction variance is further reduced by using the spatial estimation variance as the measurement variance in the updating step of the biomass propagation and its variance in time in the Kalman filter. The spatial estimation variance is obtained from kriging or cokriging based on variograms and covariograms or from an autoregressive state-space model. The effect of the combined filter scheme in space and time upon the estimation variance and yield prediction is presented and applied to various sampling scenarios.