Mapping surface conditions for crop residue assessment using PRISMA satellite imaging spectroscopy

Monitoring Non-Photosynthetically active Vegetation (NPV), as it plays an important role in the cycling of carbon, nutrients and water, is relevant to different studies including ecosystem dynamics, climate change, ecology, and hydrology, and hence is a topic of interest for remote sensing environmental applications. In croplands NPV represents a key information in the field of sustainable agriculture, given that the crop residue (CR) management affects the agri-ecological functions of soil. A proposed conservation agropractice is to leave CR in field and perform minimum tillage. In the perspective of monitoring CR presence and management from farm to regional scales, two main information are requested: i) recognition of spatial distribution of different land surface conditions (soil, vegetation and NPV both from CR and dead standing vegetation) at parcel level and ii) the characterisation of NPV classes in terms of abundance of carbon base constituent (CBC) on surface unit. Some preliminary studies with PRISMA proved that the lignin-cellulose absorption band centered at 2100nm is apparent in such data and it is reliable for the detection of NPV, besides it is promising for the characterisation of CR abundance by spectral modelling (Pepe et al. 2020). Given that the requirements for assessing crop residue cover and soil tillage activity (when) and intensity (which type) are: i) an accurate land use map (Daughtry et al. 2005); ii) the knowledge of surface conditions changes as related to the timing of tillage or planting (Zheng et al. 2012); a classification paradigm is proposed to map PRISMA data in terms of five different surface status categories: bare soil, crop residue, vegetation at emergence (including plant regrowing on crop residues), crop in vegetative stage (green vegetation) and senescence phase (dead (dying) standing vegetation). To this purpose, the method previously proposed by (Pepe et al. 2020) is improved by extending the analysis to spectral intervals other than that of lignin-cellulose, including those of leaf chlorophyll pigments (centered around 690 nm) and water content (centered around 1200 nm). Such absorption bands, representing diagnostic features to assess the presence of the different surface category, are modelled by the Exponential Gaussian Optimization method (Pompilio et al. 2009, 2010). Parameters extracted from PRISMA spectra, from a supervised training set, are used for inferring classification rules using a decision tree approach. The training set comes from a reference imagery for which information on ground conditions from an intensive field campaign are available for the study area corresponding to a large farm (3800ha) in Jolanda di Savoia, Northern Italy. The classification paradigm (spectral modelling and decision tree) is run at pixel level, afterwards the results are post-processed to obtain a final map at parcel level (which is the extent of interest). The mapping approach is applied to a set of images acquired during two crop seasons (2019-2020 and 2020-2021) over the study area as the site belongs to the network of the PRISMA mission cal/val project (PRISCAV). A total of 12 images (6 per crop season), were available for the experiment. The performance of the approach is quantitatively assessed by traditional statistics for the image where ground reference data exists and qualitatively evaluated in terms of crop conditions trajectories derived from time series analysis and compared to crop map information and management knowledge of the estate. Results proved the method to be viable and reliable for identifying practices related to land management able to recognise existence and periods of crop residue presence and confirmed that PRISMA hyperspectral data are promising for monitoring and verification actions on the implementation of conservation agriculture. Moreover, even if the mission is not intended to be operative, the revisit time and tasking characteristics of PRISMA, seems actually not optimal but sufficient, to provide a number of cloud-free images during planting season useful for monitoring CR. Such results are also important in the perspective of the new and forthcoming operational hyperspectral missions such as DLR-ENMAP and ESA-CHIME. Future studies will be devoted to evaluate the reliability and consistency of these spectroscopic approaches in the characterisation of CR abundance.