Conference,
Remote Sensing for supporting water footprint analysis of cotton production chain
(2018)
Abstract
Remote Sensing (RS) has become a plausible tool for performing different spatio-temporal analysis in complex agricultural systems of the world. The applications of RS techniques have been undertaken in Rechna Doab, Pakistan under the project “InoCottonGrow” (www.inocottongrow.net) for the purpose of (i) land use and land cover mapping, (ii) actual evapotranspiration modelling, and (iii) cotton crop yield modelling. The mapping of major land use land cover classes have been conducted utilizing time series MODIS NDVI data at 250m spatial resolution using k-means unsupervised classification technique from 2005 to 2016. Further improvements in mapping were done using high-resolution spatial data from Sentinel-1 and Sentinel-2, and applying machine-learning algorithms. Validation of these results is performed by comparison with ground information. The other applications of RS include modelling of actual evapotranspiration through fully automated SEBAL (i.e. METRIC model). Different MODIS data types were utilized, which yielded results at 1x1 km spatial resolution during 2005 to 2016. Validation of RS based actual evapotranspiration was done by comparison with Advection-Aridity approach. In addition, the in-situ sensors have been installed at two locations including the main campus of the University of Agriculture, and the Post Agricultural Research Station (PARS) for the estimation of different components of heat balance equation for the year 2018. GIS analysis has been performed by integrating the results of actual evapotranspiration with crop data, and distance from the irrigation system for its efficiency evaluations. Moreover, the RS based cotton crop yield was estimated for its further utilization in performing crop water productivity analysis. The cotton yield results were validated with the state-owned cotton yield data. The results of crop mapping show that apart from the long-term crop inventory possible from MODIS data, the enhancement of quality of results is possible using Sentinel data particularly for water and rural settlements at 20x20 m resolution. Synergic use of Sentinel based optical and radar data show the selection of the dataset is more important than either use of random forest or support vector machine algorithms. Further, the detailed mapping is helpful for devising land use scenarios and for water demand assessments. Significantly different actual evapotranspiration pattern is observed between rice and the other areas. Moreover, higher spatial difference of actual evapotranspiration patterns (upstream downstream) occur at the level of irrigation subdivisions. Irrigation divisions followed by irrigation circles and the whole Rechna exhibit less heterogeneous picture. Crop specific water consumption results show that rice is a dominant crop followed by sugarcane, fodder and cotton. Cotton crop modelled yield indicates low variability in the results which demands in the adjustments of fraction of light absorbed (FPAR) by the plants. The results of official cotton yields are lower than the modelled yields which demand for the use of the local parameters for light use efficiency, and harvest index for better comparisons.
