1. The preservation of global biodiversity has become challenging due to intensifying anthropogenic pressures. This study addresses the complex challenges associated with long-term monitoring data (i.e. missing years and gap filling) on the accuracy of temporal biodiversity trends. 2. Here, we analysed over 20 years of annual river macroinvertebrate data, simulating missing entries and applying imputation methods with linear and non-linear models to fill gaps. Our findings show that increasing numbers of gaps lead to increased trend variability, lower Akaike Information Criterion scores and higher standard deviation in model-explained deviance, thus suggesting that models fit more easily to datasets with more missing values due to fewer data constraints, while also introducing greater uncertainty and unexplained variability in the inferred trends. 3. When evaluating different gap-filling algorithms, we found that their performance varied considerably, contributing to increased uncertainty in trend estimates. Random Forest Imputations and Random Sample from Observed Values performed best, introducing less variation and aligning more closely with the original trends, whereas Predictive Mean Matching and its weighted variant amplified deviations, particularly with increasing gaps. Importantly, even a small number of missing or imputed values could, in some cases, reverse the trend direction, highlighting the risk of misinterpretation from seemingly minor data loss. 4. Synthesis and applications. In the current era of large-scale biodiversity monitoring, our study highlights the risks of missing data and the need for cautious imputation. We show that, in many cases, retaining gaps may lead to more accurate trend estimates than imputing data. When imputation is unavoidable, methods, such as Random Forest and Random Sampling from observed values performed relatively well in our macroinvertebrate richness case study. However, the choice to impute as well as the method used should be evaluated in light of the biodiversity metric and the type of trend being analysed.
