Downscaling Solar-Induced Chlorophyll Fluorescence Based on Convolutional Neural Network Method to Monitor Agricultural Drought

Downscaling Solar-Induced Chlorophyll Fluorescence Based on Convolutional Neural Network Method to Monitor Agricultural Drought

Abstract:

Agricultural drought is a frequent global phenomenon. Solar-induced chlorophyll fluorescence (SIF) is a by-product of photosynthesis that can be used to monitor vegetation growth and agricultural drought. The global 0.05° spatial resolution data set has been obtained using the data-driven algorithm method. However, the broken farmland is not conducive to regional agricultural drought monitoring. Hence, 0.05° SIF products should be downscaled. On this basis, a convolutional neural network (CNN) downscaled work was conducted in this article to obtain 0.008° spatial resolution SIF results. The downscaled SIF and land surface temperature (LST) data were used to establish the temperature fluorescence dryness index (TFDI). The new TFDI was subsequently used for monitoring agricultural drought in Henan province (China) during the corn-growing season (from June to October 2013-2017). Results showed that the downscaled SIF data exhibit a good correlation with gross primary productivity (GPP) from the Moderate Resolution Imaging Spectroradiometer (MODIS) than 0.05° SIF products. During the study period, the soil moisture fluctuation corresponded well with precipitation, and the value of TFDI had an opposite fluctuation with soil moisture. Meanwhile, the annual averaged TFDI had a high correlation with summer corn yield (R = -0.84). In conclusion, the SIF results through the CNN-based downscaled method were reliable, and the new TFDI was suitable for region agricultural drought monitoring.