Temporal And Spatial Up-scaling Methods For Vegetation Temperature Condition Index

Droughts are lasting,recurrent natural disaster that can severely threaten agricultural production and thus cause considerable economic costs.Therefore,it will be of great importance in monitoring drought dynamically and in real time.The effects of droughts are most commonly first apparent in agriculture by affecting the water balance of the crop growing cycle.Therefore,agricultural droughts could be monitored by drought-related parameters,which closely describe the spatiotemporal variations in crop water use.Remotely sensed data are extensively used in drought research because of their accuracy,objectivity and flexibility for monitoring the temporal and spatial evolution of droughts across wide and non-homogeneous areas.Different drought-related land surface parameters,such as the normalized difference vegetation index(NDVI),land surface temperature(LST)and a combination of these factors,have been employed for drought monitoring.However,the patterns and intensities of droughts are not easily discerned because droughts are slow,cumulative events.After considering both the NDVI changes and LST changes in a region,the vegetation temperature condition index(VTCI)was developed to detect droughts and characterize their onset,duration and intensity based on the assumption that the shape of the scatter plots for the LST and NDVI is triangular.VTCI was found to be closely correlated with the crop water status and perform very well in agricultural drought monitoring and prediction.The VTCI has been extensively used for drought monitoring,drought impact assessment and crop yield estimation.An increasing number of spatiotemporal images are being acquired from different types of sensors with the development of remote sensing technologies.These datasets provide adequate spatial information to study land surface processes.However,spectral radiances and surface parameters are different when interpreted at different resolutions,and their comparability is weak because of differences in sensors,changes in the surface heterogeneity,and the nonlinearity of algorithms for retrieving land surface parameters.Scale factors play important roles in practical investigations.Scale is the spacial scale and temporal frequency of some phenomenon or process.Scale can be viewed as measurements of space and time of earth observation by sensors.So scale is not only a natural attribute but also about observing and analyzing.Droughts are complex and changeable phenomenon,so surface drought features and processes at a spatiotemporal scale are different because surface pattern and process are often scale-dependent.The temporal resolution of drought information from remote sensing data ranges from a day to a few days,and the spatial resolution ranges from kilometers to meters:when the temporal resolution is higher,the spatial resolution is lower.Surface drought features and processes at a spatiotemporal scale may be similar or may need to be transformed to another scale for effectively showing droght information and using.A given spatiotemporal scale may not be the best scale to monitor droughts because the distribution features,patterns,processes,and variations in droughts should be accurately mapped and clearly visualized.Drought variables should be retrieved and modified to an optimal scale in applications.Drought information retrieved frome different spatiotemporal scale remote sensing images are not a simple average relationship,but related to underlying surface status and character of target parameters.Therefore,we must develop transforming approaches of multi spatial and temporal VTCI products to effectively retrieve drought information.The scaling process may alter statistical and spatial characteristics and induce losses of information.Therefore,understanding the scale effect and developing appropriate scaling methods that retain statistical characteristics or reveal new spatial patterns to accurately infer drought information across scales is essential.The Guanzhong Plain is the most important agricultural region in Shaanxi Province because of its excellent natural conditions.Continuous spells of poor rainfall alongside a temperature increase and scarcity of ground-water recharge increase the susceptibility to spring and early summer droughts in most areas of the Guanzhong Plain.The increasing frequency of droughts imposes huge burdens on both regional water supplies and agricultural production.In this study,the genetic algorithm(GA)and the trend surface analysis(TSA)were used to construct temporal and spatial up-scaling models based on the time series of drought monitoring results of Aqua MODIS-VTCI and Landsat 8 OLI/TIRS-VTCI.The accuracy of temporal and spatial up-scaled results were evaluated to analyze applicability and superiority of temporal up-scaling algorithm of GA and spatial up-scaling algorithm of TSA.Multi-temporal VTCIs can cover more drought information related with crop yields,and the drought occurred at different crop growth stages and its degrees lead to diverse yield reduction rate.Therefore,it is of great significance to explore how to integrate useful information from multi-temporal remote sensing data for improving the precision of drought impact assessment.Temporal up-scaling model was taken through endowed with weights to VTCI of each different growth stages based on the information of drought influence on the crop output.The key was to determine the weight coefficient of the drought influence of different growth stages on the crop output.In this study,the modeling for temporal scale transformation of VTCIs at the main growth stages of winter wheat in the Guanzhong Plain was carried out by using the normalized combination of factor weight sorting method and entropy method(CAFE),the exhaustive attack method(EA)and the genetic algorithm(GA).The results showed that the weights of impact of droughts at the main growth stages on wheat yields determined by the CAFE had large differences with the optimal weights obtained by the EA,while the weights determined by the GA were in agreement with the optimal weights by the EA.GA was superior to the CAFE in the regression analyses between the weighted VTCIs and the yields,and greatly improved the efficiency and precision of the drought impact.Meanwhile,the GA had the same performance of the EA,but the computation time of the GA was significantly lower than that of the EA.These results indicated that the weight at each growth stage of winter wheat in the Guanzhong Plain determined by GA was quite reasonable,and could more accurately reflected the drought information of the stage,and the GA was more suitable for temporal up-scaling of drought impact assessment study.Trend surface analysis(TSA)is a multivariate statistical analysis method in which a surface is"best fitted" to a set of data by using the least-squares criterion.This method is quite effective at processing large amounts of drought data,whose distribution within a study area is continuous for geospatial analysis.This characteristic reinforces the possibility of using TSA for scale effects and scale transformation.This study we developed and evaluated the TSA method for up-scaling Landsat-VTCI data of Guanzhong Plain from a finer to a coarser spatial resolution.We validate the performance of TSA in up-scaling Landsat-VTCI data by standard deviation(SD),entropy and average gradient(AG),and evaluate the superiority of TSA compared to the traditional window average(WA)method by the spatial distribution and texture features,structural similarity(SSIM),correlation coefficients,peak signal noise ratio(PSNR),root mean square error analysis(RMSE)and absolute error(AE)between up-scaled images and corresponding MODIS-VTCI images.The results showed that up-scaled Landsat-VTCI images using TSA and WA were generally matched the MODIS-VTCI images with respect to the spatial distribution and texture characteristics of droughts,results of SSIM,correlation coefficients and PSNR between up-scaled Landsat-VTCI images using TSA and MODIS-VTCI images had been in good agreement,which all larger than that using WA.Results of RMSE and AE showed that up-scaled images using TSA were closer to the quantitative drought monitoring results of MODIS-VTCI images.TSA can better characterize the dominant distribution features,variability,patterns and processes of drought information and is more applicable to up-scaling Landsat-VTCI drought monitoring results than WA.The improved spayial up-scaling models was structured by integrating TSA and dominant class variability-weighted method(DCVW),median pixel variability-weighted method(MPVW),arithmetic average variability-weighted method(AAVW),point spread function(PSF),mixed pixel area weight(MPAW).TSA was used to simulate the spatial distribution and regional trends to obtain the trend surface,and then DCVW,MPVW,AAVW,PSF and MPAW were used to analyze spatial information of the trend surface and structure up-scaling models.The improvement was assessed by comparing texture features,SSIM,correlation,PSNR and RMSE between up-scaled images and MODIS-VTCI images.The results showed that the values of SSIM,correlation coefficient and PSNR between MODIS-VTCI images and up-scaled images using DCVW and MPAW were all larger than that using TSA,the values of RMSE all smaller than that using TSA.This means that up-scaling model which integrated TSA and DCVW,MPAW can improve the up-scaling results.MPAW and DCVW has the remarkable improvement,especially MPAW,the performance of up-scaling can be best improved,it is more applicable and more reliable to up-scale Landsat-VTCI images from a finer to a coarser scale.