Cloud-free Datasets Research Articles

Modelling high-resolution Evapotranspiration in fragmented croplands from the constellation of Sentinels

The Sentinels for Evapotranspiration (Sen-ET) developed by the European Space Agency (ESA) provides a synoptic representation of actual ET (ETa) and other land-surface energy fluxes at high spatial (20 m) and temporal (daily) resolutions by the fusion of Sentinel-2 and Sentinel-3 datasets. However, the non-availability of optical sensor enabled Sentinel-2 images during the periods of high cloud cover such as monsoon limits the perennial use of Sen-ET. Here, we propose an alternate method to estimate ETa fluxes using locally observed weather data and the polarimetric parameters derived from Sentinel-1 with an application to fragmented, heterogeneous croplands. At first, the Sen-ET modelling approach with cloud-free datasets is validated using eddy covariance (EC) flux datasets (R2 = 0.74). Sen-ET derived ETa fluxes were used to develop crop coefficient (Kc) curves by considering reference ET (ET0) obtained from the weather data (Monteith, 1965b). Functional relations were then established between Sentinel-1 derived polarimetric parameters and Kc considering cloud-free datasets, and further translated to the periods of high cloud cover. We observed a strong dependency of VH on Kc for sugarcane and vegetables, and radar vegetation index (RVI) on Kc for rice (Rveg2=0.55,Rs.c.2=0.61,Rrice.2=0.51). The performance of Sentinel-1 derived ETa estimates were further validated with EC fluxes and found to be high during non-monsoon period (R2 = 0.74, RMSE = 1.25) than in monsoon period (R2 = 0.63, RMSE = 0.82). Our results indicate the suitability of Sentinel-1 in estimating ETa fluxes during the periods of high cloud cover for effective irrigation management.

Exploitation of optical and SAR amplitude imagery for landslide identification: a case study from Sikkim, Northeast India

Detection and mapping of landslides is one of the most important techniques used for reducing the impact of natural disasters especially in the Himalaya, owing to its high amount of tectonic deformation, seismicity, and unfavorable climatic conditions. Moreover, the northeastern part of the Himalaya, severely affected by landslides every monsoon, is poorly studied. The information on the inventories is inhomogeneous and lacking. In this context, satellite-based earth observation data, which has significantly advanced in the last decade and often serves as a potential source for data collection, monitoring, and damage assessment for disasters in a short time span, has been implemented. Keeping in mind the above framework, this study aims to exploit the potentials of Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 optical imagery for identifying new landslides in vegetated and hilly areas of the northeastern part of India. In order to assess the potentials of our data and methodology, a landslide event which occurred on 13 August 2016 13:30 h (IST) in North Sikkim, India, triggered due to rainfall has been explored in detail. The landslide also resulted in the formation of a lake, 2.2 km in length and 290 m in width. Difficulty in procurement of cloud-free datasets immediately after the event led us to the use of Sentinel-1 SAR backscatter data, to assess its potential for this purpose. It is observed that the potential of SAR amplitude imagery is limited to different aspects as per the sensor look direction during the mode of acquisition. Furthermore, the present study also incorporates a change detection algorithm to evaluate the performance of the Sudden Landslide Identification Product (SLIP) model to identify new landslides using Sentinel-2 multispectral imagery. Overall, the results exhibit that integrated usage of both optical and SAR amplitude imagery may provide a plethora of information for identification and mapping of new landslides for damage assessment and early warning. All the above results combined together suggest this method for rapid identification of landslides in the Himalayan terrain with special emphasis on the northeastern part of the Himalaya. The automation of this method for future operational usage is also suggested.

An Automated Approach to Map Winter Cropped Area of Smallholder Farms across Large Scales Using MODIS Imagery

Fine-scale agricultural statistics are an important tool for understanding trends in food production and their associated drivers, yet these data are rarely collected in smallholder systems. These statistics are particularly important for smallholder systems given the large amount of fine-scale heterogeneity in production that occurs in these regions. To overcome the lack of ground data, satellite data are often used to map fine-scale agricultural statistics. However, doing so is challenging for smallholder systems because of (1) complex sub-pixel heterogeneity; (2) little to no available calibration data; and (3) high amounts of cloud cover as most smallholder systems occur in the tropics. We develop an automated method termed the MODIS Scaling Approach (MSA) to map smallholder cropped area across large spatial and temporal scales using MODIS Enhanced Vegetation Index (EVI) satellite data. We use this method to map winter cropped area, a key measure of cropping intensity, across the Indian subcontinent annually from 2000–2001 to 2015–2016. The MSA defines a pixel as cropped based on winter growing season phenology and scales the percent of cropped area within a single MODIS pixel based on observed EVI values at peak phenology. We validated the result with eleven high-resolution scenes (spatial scale of 5 × 5 m2 or finer) that we classified into cropped versus non-cropped maps using training data collected by visual inspection of the high-resolution imagery. The MSA had moderate to high accuracies when validated using these eleven scenes across India (R2 ranging between 0.19 and 0.89 with an overall R2 of 0.71 across all sites). This method requires no calibration data, making it easy to implement across large spatial and temporal scales, with 100% spatial coverage due to the compositing of EVI to generate cloud-free data sets. The accuracies found in this study are similar to those of other studies that map crop production using automated methods and use no calibration data. To aid research on agricultural production at fine spatial scales in India, we make our annual winter crop maps from 2000–2001 to 2015–2016 at 1 × 1 km2 produced in this study publically available through the NASA Socioeconomic Data and Applications Center (SEDAC) hosted by the Center for International Earth Science Information Network (CIESIN) at Columbia University. We also make our R script available since it is likely that this method can be used to map smallholder agriculture in other regions across the globe given that our method performed well in disparate agro-ecologies across India.

NDVI from Landsat 8 Vegetation Indices to Study Movement Dynamics of Capra Ibex in Mountain Areas

Abstract. In this study we analyse the correlation between the spatial positions of Capra ibex (mountain goat) on an hourly basis and the information obtained from vegetation indices extracted from Landsat 8 datasets. Eight individuals were tagged with a collar with a GNSS receiver and their position was recorded every hour since the beginning of 2013 till 2014 (still ongoing); a total of 16 Landsat 8 cloud-free datasets overlapped that area during that time period. All images were brought to a reference radiometric level and NDVI was calculated. To assess behaviour of animal movement, NDVI values were extracted at each position (i.e. every hour). A daily "area of influence" was calculated by spatially creating a convex hull perimeter around the 24 points relative to each day, and then applying a 120 m buffer (figure 4). In each buffer a set of 24 points was randomly chosen and NDVI values again extracted. Statistical analysis and significance testing supported the hypothesis of the pseudo-random NDVI values to be have, in average, lower values than the real NDVI values, with a p value of 0.129 for not paired t test and p value of < 0.001 for pairwise t test. This is still a first study which will go more in depth in near future by testing models to see if the animal movements in different periods of the year follow in some way the phenological stage of vegetation. Different aspects have to be accounted for, such as the behaviour of animals when not feeding (e.g. resting) and the statistical significance of daily distributions, which might be improved by analysing broader gaps of time.

An approach to monitoring mangrove extents through time-series comparison of JERS-1 SAR and ALOS PALSAR data

Between 2007 and 2010, Japan’s Advanced Land Observing Satellite (ALOS) Phased Arrayed L-band Synthetic Aperture Radar (PALSAR) captured dual polarization HH and HV data across the tropics and sub-tropics. A pan tropical dataset of Japanese Earth Resources Satellite (JERS-1) SAR (HH) data was also acquired between 1995 and 1998. The provision of these comparable cloud-free datasets provided an opportunity for observing changes in the extent of coastal mangroves over more than a decade. Focusing on nine sites distributed through the tropics, this paper demonstrates how these data can be used to backdate and update existing baseline maps of mangrove extent. The benefits of integrating dense time-series of Landsat sensor data for both validating assessments of change and determining the causes of change are outlined. The approach is evaluated for wider application across the geographical range of mangroves in order to advance the development of JAXA’s Global Mangrove Watch (GMW) program.

A Method for Determining the Sensor Degradation Rates of NOAA AVHRR Channels 1 and 2

Abstract A method is described to determine the degradation rates of NOAA Advanced Very High Resolution Radiometer (AVHRR) visible channels. Thirty-eight desert targets (each 20 km × 20 km) were selected over the northwest region of China after testing uniformity of the underlying surface. Two cloud-free datasets for the same satellite were sampled from two different periods several years apart. The functional relationships were established between the observed reflectances and the satellite observation angles. By allowing comparison of the functional relationships of the two different periods and computation of the sensor degradation rates using the stepwise method, the degradation rates were determined to be 5.8% and 4.6% per year for channels 1 and 2, respectively, based on the NOAA-9 AVHRR data for October 1985 and September 1988.