Stubble burning is one of the major environmental hazards in almost all parts of the world. As agriculture became mechanized, combine harvesters leaves root-bound and scattered crop residues that are labor and cost intensive to remove, causing an increase in the recorded cases for burning of residues. The problem of stubble burning is more intense in the Indo-Gangetic Plain (IGP) of India due to highly mechanized farming practices, which leaves a huge amount of stubble in the field particularly after the harvest of rice. Traditionally, farmers collect crop residue to feed livestock manually. In this stubble burning study we explore the potential of the MODIS "MCD14DL-NRT" (Active Fire Data) product at a spatial resolution of 1000m available since November 2000, for the identification of fire prone areas in India. This study highlights the count of fire occurrences recorded from the MODIS Active Fire Data over a period of more than 20 years. During the months of October and November, which coincide with the paddy harvest season in the area an abrupt increase in fire activity is observed. The period of paddy harvesting, coupled with the onset of the winter season in the northern part of the country makes the region highly polluted and a breeding ground for numerous health problems for the citizens. In our study, we found that the state of Punjab records the majority of fires in India during this time period.

Accurate segmentation of cloud and shadow pixels in the image scenes obtained from optical satellite data is important for monitoring agriculture plots. At any given time instant, the images obtained from any multi-spectral satellite, especially during the wet season of the growing period, can have the reflectance value pixels corrupted either by cloud or shadow. Plot-level analysis of the derived spectral indices can have artifacts due to the clouds or cloud shadow resulting in wrong interpretation or non-rejection of tiles by automatic processing methods determining coverage extents. In this article, we proposed a new methodology to mask clouds and shadows from Sentinel−2 imagery. It consists of training a U-Net model on cloud and shadow masks generated by Support Vector Machines trained on the Hollstein dataset, a cloud mask from the s2cloudless model, and a projected shadow mask using sensor geometry information.

Land-use and land-cover (LULC) plays a crucial role in various agricultural applications, including crop identification, crop acreage estimation, and yield estimation. However, accurately mapping cropland extent can be challenging, particularly when grassland areas are misclassified as cropland due to similar temporal signatures during the rainy season. To address this issue, this study explores the use of interferometric coherence, a measure of correlation between two measurements of the same resolution cell, as a distinguishing feature between cropland and grassland. The main objective is to improve the classification accuracy of cropland and grassland areas. The methodology involves computation of coherence, and generation and selection of statistical features based on NDVI, VH, and coherence-VV. Seasonal statistics are computed for these features, and the separability index is used to evaluate their discriminatory power. The results demonstrate that coherence features, particularly coherence-VV have high separability indices, indicating their effectiveness in distinguishing between cropland and grassland. The proposed methodology achieves promising results with an F1-score of 0.95. The study highlights the potential of coherence features for improving the accuracy of land-use/land-cover classification and provides insights for agricultural applications in different geographical regions.

The study of soil moisture is crucial for understanding the hydrological cycle and its impact on energy and water exchanges at the land-atmosphere interface. Synthetic Aperture Radar (SAR) data, such as Sentinel-1, has shown potential for estimating soil moisture at high spatio-temporal resolutions. However, the sensitivity of SAR responses to soil moisture and the applicability of Sentinel-1 data for soil moisture estimation at different land covers require further investigation. In this paper, we evaluate the sensitivity of Sentinel-1 data for soil moisture estimation and compare the estimated soil moisture for different land covers. A change detection approach combined with a vegetation scattering model is employed to estimate soil moisture. The results demonstrate that while the change detection approach with vegetation correction improves soil moisture estimation, the accuracy is still not within an acceptable range for plot-level decision-making, such as irrigation management. However, the results show that the soil moisture information obtained from Sentinel-1 data can be suitable for regional-level monitoring applications and decision-making.

Forecasting crop phenology helps in crop production estimation, irrigation scheduling and crop identification. In this article’ we propose a methodology that uses the Normalized Difference Vegetation Index (NDVI), derived from the bands of an optical multispectral satellite (Sentinel-2) data for forecasting crop phenology stages using a Long Short-Term Memory (LSTM) network. Firstly, the LSTM network is built for forecasting the NDVI values of the potato crop, until the end of the season. Then the crop phenology estimation technique is applied to this forecast data to obtain the phenology points. The potato plots considered for the study were grown around the Gujarat state in India. The performance of the LSTM network on the test dataset is given by MAE≈ 0.19, MSE≈ 0.06 and RMSE≈ 0.23, and the phenology points are detected with a mean deviation of ±10 days.