Mandena mine-climate interaction, 1993-2022: complete reproducibility package

This record provides the complete offline reproducibility package for the 1993–2022 mine-by-climate interaction analysis conducted for the Mandena basin near Taolagnaro, southeastern Madagascar. The study examines whether hydrological responses to prescribed mining-related land transformation remain constant across years or vary with hydroclimatic state, transformation severity and landscape-unit type. The analysis links two independently established source records without redefining their spatial or temporal domains. The first is the Mandena hydrological BACI and SWAT+ record, which provides the fixed basin geometry, hydrological response units (HRUs), mining and spatial-reference zones, baseline simulation, three prescribed mining-transformation scenarios and annual hydrological outputs. The second is the Mandena hydroclimate record, which provides a calendar-aligned 1993–2022 climate series derived from DGM-RFE precipitation, NASA POWER temperature and ERA5-Land atmospheric variables. The interaction analysis therefore uses exactly one annual record for each BACI year and preserves the same 85.8388 km² basin and 30-year inferential window. The controlled scenario experiment evaluates three prescribed transformation severities, reported in the manuscript as low, central and high. The machine-readable scenario label mid corresponds to the central scenario. Each scenario uses the same 158 selected mining HRUs, comprising 144 non-wetland and 14 wetland HRUs. Scenario responses are calculated as arithmetic means across the same selected HRUs and are compared with the matching baseline simulation. These scenario results are therefore mean model-unit responses rather than basin-area means. This estimand is distinct from the historical BACI analysis, in which mining-zone and spatial-reference-zone outputs are reconstructed using intersection-area weighting. The prescribed scenario parameterizations modify surface-flow roughness, leaf-area and biomass parameters, residue cover, soil available-water capacity and SCS Curve Number. Low, central and high severities use roughness multipliers of 0.90, 0.75 and 0.60; leaf-area and biomass multipliers of 0.70, 0.40 and 0.15; residue multipliers of 0.80, 0.60 and 0.40; and available-water-capacity multipliers of 0.90, 0.75 and 0.60. Curve-number increments for hydrologic soil groups A/B/C/D are 3/4/5/5, 6/8/9/9 and 9/11/12/12, subject to scenario-specific caps. These scenarios represent prescribed model states and are not interpreted as a measured chronology of mine development. Nine annual hydrological responses are evaluated: generated surface runoff, water yield, percolation, actual evapotranspiration, plant transpiration, soil evaporation, final soil water, mean soil water and annual soil-water change. The primary hydroclimatic covariate is joint_climate_exposure_z. Heat exposure is the rank-normal score of NASA POWER annual mean temperature. Dryness exposure is the rank-normal score of ERA5-Land FAO-56 reference evapotranspiration minus DGM-RFE precipitation. Their mean is standardized across the 30 calendar years. Higher values therefore denote relatively hotter and/or drier annual atmospheric conditions within the observed 1993–2022 period. The index is an empirical description of interannual hydroclimatic state, not a future-climate scenario. For each mining scenario, landscape block and hydrological response, annual scenario-minus-baseline differences are regressed on hydroclimatic exposure. Reported slopes represent the change in modeled scenario-minus-baseline response associated with one within-window standard deviation of exposure. The central scenario across all 158 HRUs is the principal process experiment. Additional analyses evaluate the low-to-high severity gradient, non-wetland versus wetland contrasts and separate marginal associations with heat, dryness, annual precipitation, annual mean temperature, FAO-56 reference evapotranspiration, climatic water balance, Rx1day, maximum consecutive dry days and ERA5-Land wind speed. The historical analysis is intentionally separate from the controlled scenario experiment. The annual mining-minus-reference BACI contrast is reconstructed and modeled as a function of period, hydroclimatic exposure and their interaction. The post-2008 interaction coefficient estimates whether the climate slope of the mining-minus-reference contrast changed after the intervention boundary. This design prevents simulated scenario susceptibility from being interpreted as evidence of a historically detected mine-by-climate amplification. Statistical inference is conducted at the annual level. HRUs and gridded climate cells are not treated as additional independent observations. Coefficients are estimated by ordinary least squares. Standard errors use a Newey–West covariance estimator with lag 1 and finite-sample correction. Confidence intervals use 2,000 circular moving-block residual-bootstrap repetitions with a three-year block length and a fixed archived seed. Holm adjustment is applied separately within each declared nine-response family. A forcing-aligned 1993–2020 analysis tests dependence on the final two BACI response years, for which the unchanged SWAT+ BACI chain used weather-generator rainfall rather than the DGM-based forcing used through 2020. The controlled central scenario produces substantial mean redistribution among hydrological pathways. Across 1993–2022, plant transpiration decreases by 312.95 mm relative to baseline, soil evaporation increases by 189.17 mm, actual evapotranspiration decreases by 136.37 mm, percolation increases by 105.56 mm, water yield increases by 44.01 mm, generated surface runoff increases by 43.69 mm and final soil water increases by 40.21 mm. Hydroclimatic state modifies several of these modeled responses. Per one standard deviation of joint heat–dryness exposure, generated-runoff and water-yield differences decrease by 21.65 and 21.64 mm, respectively, while plant-transpiration, actual-evapotranspiration and mean-soil-water differences increase by 31.18, 20.80 and 8.14 mm. Five of the nine central-scenario responses are detected after Holm adjustment. These results indicate that the modeled hydrological effect of the prescribed transformation is not constant across annual hydroclimatic states. The interaction also varies with imposed transformation severity. The high-minus-low climate slope is −27.48 mm for generated runoff, −27.42 mm for water yield, +17.57 mm for percolation and +10.41 mm for mean soil water. Landscape-unit contrasts likewise vary with severity: the non-wetland-minus-wetland runoff slope changes from −12.20 mm per exposure standard deviation under the low scenario to −32.23 mm under the central scenario and −52.47 mm under the high scenario. Separate one-indicator models associate the runoff and water-yield response most strongly with variables representing water availability. Higher dryness is associated with smaller scenario-related runoff and water-yield increments, whereas higher annual precipitation and climatic water balance show the opposite pattern. Reference evapotranspiration, rainfall extremes and dry-spell duration also contribute to the observed association structure. These marginal models are not mutually adjusted and are not interpreted as a causal decomposition. The forcing-aligned 1993–2020 analysis preserves the principal directions for generated runoff, water yield, plant transpiration, actual evapotranspiration and mean soil water, indicating that the central interaction pattern is not produced by the two terminal weather-generator-forced response years. In contrast, the historical BACI analysis detects no post-2008 change in climate slope for any of the nine hydrological responses after Holm adjustment. For 1993–2022, all adjusted p-values equal 1.00 and all block-bootstrap intervals include zero. The archive therefore supports a distinction between modeled process susceptibility under controlled mining transformations and historically detected change in the annual mining-minus-reference climate response. The former is detected; the latter is not. The package does not support claims that climate change caused the simulated hydrological responses, that mining caused the observed basin hydroclimate trends, that the historical mining effect has already intensified because of climate change, or that a non-detected historical interaction is exactly zero. It also does not constitute a future projection. The archive is designed for complete offline reconstruction. python run_all.py, run_all.ps1 or run_all.sh reconstructs the annual scenario and BACI chains from the archived SWAT+ HRU outputs, checks equality against the deposited BACI tables, joins the 30-row hydroclimate bridge by calendar year, fits all declared models, regenerates the result tables and figures, runs the automated test suite, rebuilds the SHA-256 inventory and authenticates the package. The statistical workflow requires CPython 3.10 or later and the Python standard library; local figure conversion additionally requires a supported SVG renderer. No network request is required during analysis. The package contains the raw and reference SWAT+ inputs required for the interaction analysis, the complete embedded hydroclimate reproducibility archive, fixed analysis configuration, exact scenario parameterization, source-provenance records, reconstructed annual tables, fifteen result tables, twenty numbered figures plus a graphical abstract, source code, automated tests, execution receipts, environment information, data dictionary, interpretation guidance, figure catalogue and a SHA-256 manifest. Each figure is supplied in SVG, vector PDF, 600-dpi PNG and 600-dpi TIFF formats. This record is derived from and should be interpreted together with the two source records: Mandena BACI hydrological and SWAT+ record: DOI 10.5281/zenodo.21633869Mandena hydroclimate record, 1993–2022: DOI 10.5281/zenodo.22113639 The present record does not replace either source archive. It preserves both source contracts and adds the annual mine-by-climate interaction analysis, associated statistical inference and article-specific reproducibility outputs. Author: Zo Rivomanana Rasoanaivo, Department of Earth and Evolutionary Sciences, University of Antananarivo, Madagascar.ORCID: 0009-0003-0725-3764Temporal coverage: 1993–2022Study area: Mandena basin, southeastern MadagascarLicensing:Original data, documentation and figures in this record are released under the Creative Commons Attribution 4.0 International license (CC BY 4.0); Original source code is released under the MIT License.Third-party materials remain subject to their original terms, as documented in the archive.Software/model framework: SWAT+ revision 61.0.2; CPython 3.10+ for the interaction workflow.