Small area estimation of crop production using spatial models

Abstract

Agricultural surveys for estimation of crop productions in India are designed to obtain reliable estimates at high geographical levels (districts). Due to shift in emphasis of planning from a macro to micro level, there is a strong need to estimate the agricultural production at small area levels such as tehsils (middle administrative unit), blocks (low administrative unit), and even villages. The main objective of this study is to show the potential of spatial models based on geo-statistical techniques of variogram and kriging for estimation of crop production at small area level (low administrative level) through generation of production surface. To achieve this objective, Indian Remote Sensing IRS-1D satellite data for 1997-1998 for Rohtak district of Haryana state have been used. The results of the study indicate that application of spatial models for estimation of production of major crops at small administrative area level has great potential. The predicted yield/production of wheat in the district/tehsil turns out to be comparable to the estimated production of wheat crop through large-scale survey data. The proposed technique can be further explored and refined to provide reliable estimates of crop production at small area levels. Traditionally, in India, agricultural statistics related to crop production is based on two parameters, viz., (i) area under a crop and (ii) yield of a crop. In general, the former is obtained through a complete enumeration using land records whereas the latter is estimated through sample surveys. Because of a phenomenal growth in computer science, sophisticated computer intensive tools are now available, including geographic information system (GIS), which is useful for efficient estimation of spatial parameters. The agricultural parameters related to crop production are geographical (spatial) in nature; therefore, GIS, in combination with spatial statistical techniques, might be a powerful tool for estimation of these parameters. Spatial statistics deals with realization of random variables arranged over a two- dimensional surface (23). Crop production parameters, such as productivity of crop, soil parameters and availability of ground water are geographical in nature, that is, changes in properties of these indicator variables are gradual and directional over space. Therefore, two adjacent fields are likely to be homogeneous in respect to crop production as compared to fields that are far away. It is expected that this spatial relationship may be used to improve estimation of production parameters. Several attempts have been made to develop suitable spatial models to describe effectively spatial relationship. Brun and Stein (3) demonstrated that variables computed from digital elevation models served as independent variables in the regression analysis to predict soil properties. Lark (11) proposed variogram models to specify correlations of errors