Valutazione dei danni da grandine e vento forte su frumento (Triticum aestivum L.) e mais (Zea mays L.) attraverso tecniche di telerilevamento

Extreme weather events such as hailstorm and strong winds pose a serious threat to agriculture. Their frequency and intensity are forecasted to increase in the next decades, fueled by climate change. Yield decreases are expected, such that food security and financial stability of farmers are threatened. Maize (Zea mays L.) and winter wheat (Triticum aestivum L.) are two of the main cereals worldwide, and are particularly susceptible to hailstorm and strong winds, suffering potentially high yield losses when these extreme weather conditions occur. The crop insurance refund scheme depends on damage assessment. Traditionally this is done by field inspections, but these suffer from limited damage spatial variability account and labor-intensive field operations. The need for an integration of traditional inspections with remote sensing arose, potentially allowing a faster, more efficient approach, and offering the capability to address damage spatial variability even within the field level. In this thesis work, the potential to integrate remote sensing into hailstorm and strong winds damage assessment has been explored. A three-year experimentation on both maize and winter wheat has been conducted on two experimental fields located in Veneto Region (Northeastern Italy). Hail damage at different intensities and strong wind was simulated using a specifically designed prototype, accurately resembling real damage thanks to insurance field inspectors’ supervision. Experimental plots have been surveyed both with ground vegetation traits measurements and with hyperspectral and multispectral remote sensing using Unmanned Aerial Vehicles (UAV) and satellite borne sensors. In maize, this allowed to refine and calibrate a leaf area index (LAI) estimation algorithm suitable for estimation from both UAV and Sentinel-2 applications (chapter 2). LAI represent a suitable proxy for hailstorm damage mapping and can be a primer for damage estimation. It was noted how LAI acts as a suitable descriptor of canopy changes following damage, and how different ground spatial resolutions can lead to noticeable differences during the estimation, with the higher UAV resolution capable of more accurate estimations. Nonetheless, Sentinel-2 proved effective and allows for regional damage estimations beyond the field level. Such a canopy structure driven approach proved less effective in winter wheat (chapter 3), where LAI did not appear to be an effective proxy for damage. Here, a spectral unmixing approach was tested using hyperspectral UAV data. Damage was mapped on the experimental fields through a near infrared (NIR) absorption feature linked to brown pigment occurence. These pigments are related to plant tissue decay and necrosis and can be effectively linked to hailstorm damage occurrence and intensity. Nonetheless, hyperspectral approaches can prove difficult to adopt out of the research, therefore a multispectral index capturing the NIR feature was designed and proved to be and effective mapping tool for hail damage at different intensities in winter wheat. Concerning wind damages, a study on maize was conducted using supervised machine learning classification to map lodged areas. Different sensors were used at different spatial resolutions, both hyperspectral and multispectral; the former on UAV, the latter on satellites. The research showed that machine learning tools can effectively classify severely lodged areas. Moreover it was shown that, for such a task, multispectral satellites and coarser spatial resolutions can produce even better results than UAV hyperspectral counterparts. Overall, the research has shown that different crops require different approaches and methods given their specific response to damage both in terms of plant physiology and canopy structure. The physiological stage at which damage occurs and subsequent survey timing were also shown a critical factor for both crops.