The aim of this study is to model a hydrogenotrophic denitrification process in a venturi-integrated submerged membrane bioreactor (MBR) system. The MBR was operated in batch mode using feed concentrations of 100 and 150 mg NO₃-N/L. In contrast to most of the denitrification process models that represent the mixed culture with one composite biomass parameter, the biomass was subdivided into two main categories in this modelling study: mainly nitrate-reducing biomass and mainly nitrite-reducing biomass. The determination coefficients (r²) in the range of 0.97–0.99 indicate that the model successfully simulates the concentrations of nitrate- and nitrite-nitrogen in the bioreactor. The maximum specific growth rate of nitrite-reducing biomass (0.06 h⁻¹) was found to be higher than that of nitrate-reducing biomass (0.0002 h⁻¹). Similarly, the growth yield coefficient of nitrite-reducing biomass was higher than that of nitrate-reducing biomass (0.44 vs. 0.31 g biomass/g substrate). The kinetic and stoichiometric coefficients obtained from this modelling study suggest that the limiting step determining the overall conversion rate of hydrogenotrophic denitrification process is the conversion of nitrite to nitrogen gas. GRAPHICAL ABSTRACT

Agricultural drainage water is a saline return flow whose reuse is constrained in intensively cultivated basins such as the Harran Plain in southeastern Türkiye. This exploratory, descriptive bench study examined microfiltration (MF) pretreatment and adjustment of the feed pH to approximately 7 on real drainage water (4.8 to 8.1 mS cm⁻¹) with flat-sheet NF270 and XLE membranes at 8 bar. Six treatment trains, four forming a 2 × 2 factorial, were each run twice in 6 h batch concentration mode, with the two runs of each train made on separate days. XLE rejected 93 to 98% of feed conductivity and delivered 0.10 to 0.63 mS cm⁻¹ permeate, whereas NF270 rejected 76 to 80%. The second run of every factorial cell rejected more than the first, so run order was blocked post hoc. The blocked model associates MF pretreatment with a gain of 2.0 percentage points (95% CI 0.03 to 3.97) and bounds the pH effect within two points of zero. No result reaches the 0.05 level under familywise adjustment. Normalizing flux to net driving pressure separates osmotic loss from loss of membrane permeability; one unpretreated run of eight lost permeability, an observation compatible with chance. Water recovery was not measured, and irrigation suitability was not assessed, as conductivity alone cannot establish it. These runs tested acute rejection and flux, not the scaling control that motivates acid dosing in practice; the ions that set scaling potential were not measured. The findings are preliminary and describe process configuration of these feeds rather than membrane-level mechanisms.

Reference evapotranspiration (ET₀) underpins irrigation scheduling and allocation, yet most agricultural land in Türkiye lies far from any station that measures the FAO-56 Penman-Monteith drivers. We tested whether ERA5-Land can replace ground observations, against a benchmark computed exclusively from measurements at 134 predominantly airport-sited GSOD stations over 2000 to 2025 (864,323 station-days), with no reanalysis information in the reference. Three candidates that use no ground data at prediction time were compared under station-, region- and time-withholding protocols: the ready-made potential evaporation field, as delivered and rescaled; FAO-56 Penman-Monteith from ERA5-Land drivers; and EvapoTurk-XFE, an explainable gradient-boosting correction on reanalysis, terrain and season features. The raw field, an open-water (pan) evaporation by definition, is unusable as ET₀ (nRMSE 97.6%, bias +2.18 mm d⁻¹); a monthly empirical pan coefficient (annual mean 0.602) removes 61 nRMSE points but leaves a structural residual. Driver-based Penman-Monteith reaches 34% nRMSE but underestimates by 19%, with reanalysis wind identified as the dominant error source. EvapoTurk-XFE reaches 21.5% daily and 15.3% monthly nRMSE with near-zero bias at stations and in years never seen in training, beating the physical baseline at 114 of 134 stations, and its SHAP attribution mirrors FAO-56 physics. Propagated through crop coefficients over 2,414 station-seasons, it cuts the systematic seasonal crop-water-requirement bias to at most 72 m³ ha⁻¹ for cotton, maize and winter wheat. An explainable, retrainable correction thus turns gridded ERA5-Land into a reference evapotranspiration product usable for agricultural water management where no station exists.

Machine-learning drought studies typically regress a smooth, autocorrelated index at single stations and report near-perfect fits. Temporal leakage is rarely controlled, and the skill the index already carries in its own history is rarely separated out. Here we present HydroSPI-ARX (Hydroclimatic, Standardised-Precipitation-Index, AutoRegressive, eXplainable), an autoregressive and explainable framework for drought early warning over Türkiye. It recasts the Standardised Precipitation Index as a leakage-controlled, lead-resolved classification target built from ERA5 and ERA5-Land hydroclimatic fields. Five drought targets are evaluated at leads of one, three, six and twelve months, with feature selection and scaling recomputed inside every training fold. The moderate-or-worse warning reaches a macro- F 1 of 0.853 one month ahead, 0.794 at three months and 0.709 at six months. At twelve months it falls to 0.486, where no configuration retains skill. Naive persistence is ahead at one month; against it the framework gains 0.038 macro- F 1 at three months and 0.066 at six, with year-block bootstrap intervals that exclude zero. Against a per-cell linear autoregression on the index’s own memory, the gain is measurable only at six months (0.043, 95% interval 0.010–0.078). Exogenous hydroclimatic fields add 0.007 at one month and nothing beyond. Skill is therefore carried by the multi-scale memory of the index, and the learned model earns its place at the six-month horizon. Scoring the same held-out forecasts against station-derived labels lowers agreement (Cohen’s kappa of 0.30, 0.26 and 0.21 at one, three and six months). The ranking of the forecasts survives, so the advantage is not an artefact of verifying against the same reanalysis. ERA5 and ERA5-Land are equivalent within a margin of 0.02 macro- F 1 set in advance, and a model trained without a climate zone still forecasts inside it. A Mann–Kendall analysis shows rising three-month drought frequency over 1981–2025.