Testing the Fluorescence Advantage: Does Solar-Induced Fluorescence Signal Agricultural Drought Earlier than NDVI in Marathwada, India?

SAKSHI D. MASKE 

India uses rainfall deficits and vegetation indices such as NDVI to declare drought, and PMFBY crop-insurance payouts depend on these declarations. Solar-Induced Fluorescence (SIF) is linked to photosynthesis, so many expect it to show drought stress earlier than NDVI. I tested this in the eight districts of Marathwada, Maharashtra, over nine Kharif seasons (2015–2023). I compared cropland-weighted GOSIF SIF with MODIS NDVI, with each NDVI value dated to the day the pixel was observed. My main question was whether SIF drops below its normal-year range earlier than NDVI in a drought year. I compared each year with the normal years, and in most cases the two alarms came within one 8-day composite of each other. For the whole region, both crossed z < −3 on the same date in 2015, and NDVI was eight days earlier in 2018. There were 15 drought district-years (each district judged against its own rainfall record). SIF flagged 12 of them and NDVI 13. Where both flagged, SIF was earlier in 7, they were on the same date in 2, and NDVI was earlier in 3. All but one of these differences were one composite or less. At looser thresholds NDVI was earlier more often. NDVI comes only every 16 days, so the direction of a one-composite difference depends on how NDVI is placed in time. When I used only the NDVI already observed on each date, SIF came first in both regional drought years and in 8 of the 12 district-years. My second question was whether SIF's normal end-of-season decline starts before NDVI's. When each year is scaled from its seasonal minimum to its peak, SIF crossed the decline thresholds first in all nine years (8.0–34.2 days) and in all 72 district-years. But in this region NDVI stays above its pre-season level after the monsoon, and this scaling pushes NDVI's crossings later. When each curve is scaled from its peak to its own post-peak minimum, the lead becomes −5.5 to +13.4 days, SIF is first in only six years, and the differences between districts go away (Friedman p = 0.97). MODIS NIRv and EVI also declined before NDVI, by about as much as GOSIF SIF. Under the second scaling, GOSIF did not decline before NIRv in any of the nine years. GOSIF is built partly from MODIS reflectance, so this is partly expected. TROPOMI SIF (2018–2020) is a direct measurement. It declined before NDVI under both scalings and before NIRv in all three years, by 1–13 days. Only 2015 and 2018 were drought years by rainfall, and the lead was not larger in those years. I then ran the same methods, without changes, on 2001–2014, which has five more drought years. The results were similar. At district level the two alarms were again mostly on the same date or one composite apart, in both directions (at z < −1.5, SIF earlier in 5 of 22 drought district-years flagged by both, same date in 11, NDVI earlier in 6). For the region, SIF flagged two of the five drought years at z < −3 and NDVI one, and the seasonal lead again matched NIRv's. My approximation of the rainfall-based Trigger-1, with dry spells counted only during the monsoon, never fired for the region. It fired in 2 of 15 drought district-years, and checking each rainfall pixel gave the same result. In this data, SIF does not give a clearly earlier drought alarm than NDVI at the sampling available. Most of GOSIF's seasonal lead comes from the way NDVI's decline is measured. The small lead of TROPOMI SIF over NIRv needs more years to test.