Indications that water bodies, forests and wetlands improve agricultural resilience to drought

Extreme weather such as droughts are rising in frequency due to climate change and its impact over croplands increase in relevance to ensure food security and economy stability. It was hypothesized that adjacent ecosystems might present services that spill over into croplands mitigating drought effects. The purpose of this study is to assess if and in which extent nearby wetlands, forests and water bodies provide drought alleviation to crops adopting a modelling approach. Sentinel-2 imagery collected over summer months from 2018 to 2022 were used to derive the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Land Surface Water Index (LSWI) and Near-Infrared Reflectance of Terrestrial Vegetation (NIRv) to proxy vegetation health and calculated over 96 arable fields selected over the study area, located on rural areas close to Uppsala and lake Mälaren in Sweden. The Savitzky-Golay smoothing filter was applied to account for seasonal characteristics and 9 metrics, including phenological parameters, were calculated to be used for multiple linear regression models for 2020, that were defined to statistically analyze the roles of different land covers and climate factors on vegetation responses. The reference year of 2020 was chosen based on the absence of drought characteristics and good number of not cloud contaminated satellite images. Area in hectares of wetlands, forests and water bodies were calculated based on three distance radii, 500m, 1000m and 5000m, as accumulated values of precipitation and temperature were also derived accounting for 14 different predictors. The model fitting was done using the Least Absolute Shrinkage and Selection Operator (LASSO) and analysis of variance (ANOVA) was used to reduce the models when necessary. Root Mean Square Error (RMSE) and Akaike information criteria (AIC) were used to assess goodness-of fit and a trade-off with model complexity was made to select the final models and applied to the remaining years, plus a model of All-years was created for each vegetation index (VI), resulting in 24 final models. The estimated values generated from these models were retrieved and subtracted from respective VI original values, the difference can be interpreted as how well the models’ predictors explained the vegetation health of the sites during the period, especially patterns over the dry years of 2018 and 2022. The study pointed to key finding as (1) Water bodies and forests are the most relevant land covers to influence positively and negatively the health of nearby croplands. Although their influence was weak. (2) LSWI models overall had a superior performance when compared to other VIs with majority highest of coefficient of determination values. It was sensitive to wetland ecosystem services, even with negative correlations for all years. (3) There is base to affirm that forests, water bodies and wetlands have the potential to provide water stress alleviation to close croplands, which can be considered natural based solutions for drought mitigation in agriculture.