Estadística de los cultivos y de la salinidad en un regadío mediante teledetección

Abstract

Crop acreage estimates are obtained for an irrigation district (26313 ha) by means of an area frame sampling and are corrected by regression with classified satellite data. The usefulness of spectral signatures for detection of salinity is also studied, together with the a na lisis of crop distribution within the large soil units of the a rea under study. Unlike the classical applications of this method, done at a regional or national level, this method concerns the statistics of a small area. Besides, every operation and the obtained results are conditioned by the small size of the plots and by the complex pattern of soil salinity affection. The ground survey was made by visiting every plot within the area frame sampling units (segments). The ground truth was obtained for 4% of the studied area. The segments were squares of about 500x500 meters, randomly selected within the irrigated district. The ground survey allowed calculating the percentage of the main land uses and their distribution within the segments. This survey allowed for both the estimation of the land use surfaces by direct expansion and for the selection of training fields for the supervised mono- and multi-temporal classification of two Landsat-5 TM images. The normality of the distribution of each crop sample was tested, what showed that the assumption of normality did not hold for rice and sunflower samples. This may be common for small areas, depending on the representation and spatial distribution of crops. Although the referred methodology could not be applied to these crops, the total surface of rice was estimated by multispectral classification.

The acreage precision obtained by direct expansion was globally comparad with those from the different regression estimations by means of weighed coefficients of variation (C. V.p). For the regression estimations the relative weighed efficiencies (E.R.p) were also calculated. The precision of the expansion estimation (C.V.p = 11.6%) was increased by the different regression estimations thanks to the use of remote sensing. The best estimator was obtained by regression with multitemporal manual classification data (C.V.p = 5.8% y E.R.p =4.2). The spectral signatures established by selectioh of training areas based on the ground truth and on a simplified soil map, were not able to clearly descriminate salinity affection. However, it was possible to establish a relationship between crop distribution within the different soil units and their tolerance to salinity, in spite of the influence of agronomic and economic factors.