[1.3.71] 2026-04-14

- OBJ loader now generates unique material names when loading multiple OBJ files with colliding MTL material names (e.g., Blender's default "Material.001"), instead of silently reusing the first registered material.
- Added `Context::updateTimeseriesData()` method to modify the value of an existing timeseries data point, identified by its (Date, Time) key. Throws a runtime error if the variable or specified timestamp does not exist.
- `helios::writeEXR()` now maps common channel name aliases (`red`/`green`/`blue`/`alpha`, case-insensitive) to the standard EXR channel names `R`/`G`/`B`/`A` so that images written with descriptive band labels are recognized as RGB by external EXR viewers.

- Added `writePlantStructureUSD()` method to export plant structures as USDA files with PhysX articulated rigid bodies for NVIDIA IsaacSim physics simulation. Each tube segment becomes a capsule-shaped rigid link connected by spherical joints whose local frames encode the rest-pose orientation (zero joint angle reproduces the authored pose), with spring/damper drives derived from beam bending stiffness. Visual meshes use smooth area-weighted faceVarying normals and indexed primvars, materials are nested under the default prim, and a single shared angular drive replaces the previous per-axis rotX/rotY drives.
- Added `registerGrowthFrame()`, `writePlantGrowthUSD()`, `clearGrowthFrames()`, and `getGrowthFrameCount()` methods to capture per-step plant geometry snapshots and export them as a time-sampled USDA animation for Blender. Organs that appear during growth are toggled visible via scale/visibility keyframes, with step-style hold keyframes preventing interpolation between growth frames.
- Fixed `readPlantStructureXML()` to use the plant instance's base position instead of the origin when reconstructing the first internode of the base shoot.
- Fixed operator-precedence bug in `random_float()` helper used by Cowpea phytomer generation, where `RAND_MAX / (max - min)` was being divided before the cast instead of after, producing skewed random ranges.
- Floral bud `age` is now tracked from the time of bud break and written as object data on inflorescence and peduncle objects (when the `age` output is enabled), so flower/fruit organs report a meaningful age rather than inheriting the parent phytomer's age.

- `LeafOptics::run()` now assigns spectra to newly added primitives within an existing object, instead of skipping them when the object's nitrogen content has not changed enough to trigger a bin reassignment.

- Reformulated TPU limitation in the Farquhar model to compare at the assimilation level (`A_TPU = 3*TPU - Rd`) rather than via `Wp = 3*TPU*Ci/(Ci - Gamma*)`, eliminating the singularity at `Ci = Gamma*` that produced non-physical assimilation values when intercellular CO2 approached the compensation point. Also fixed `evaluateCi_Farquhar()` to read `TPUflag` from the model coefficients struct instead of the (uninitialised) variables vector.

- Fixed `computeCapacitance()` to use the first-derivative finite-difference stencil `(f2 - f0)/(2h)` instead of the second-derivative stencil `(f0 - 2f1 + f2)/h^2`, since hydraulic capacitance is defined as `Wsat / (dpsi/dw)`.
- Fixed `setStemHydraulicConductanceTemperatureDependence()` to modify the stem conductance struct instead of the root conductance struct.
- Fixed `setLeafHydraulicCapacitanceFromLibrary()` parameters: `PistachioFemale` had a positive (physically impossible) osmotic potential at full turgor and used Walnut's RWC at turgor loss; the unknown-species fallback claimed to be Walnut but used Western Redbud values.
- Fixed per-primitive coefficient lookup in `updateLeafWaterPotentialsOfPlant()`, which indexed `modelcoeffs_map` by the loop counter instead of the primitive UUID, silently discarding any per-primitive coefficients set via `setModelCoefficients(coeffs, UUIDs)`.
- Fixed non-steady-state leaf RWC update to divide the net flux by `saturated_specific_water_content`, so that leaves with a larger water reservoir respond more slowly (previously the dimensionless RWC was incremented by raw `mol/m^2` flux).
- Fixed `computeConductance()` and `updateRootAndStemWaterPotentialsOfPlant()` convergence test to use `fabs()` instead of `abs()`, which on float arguments could resolve to the integer overload and truncate intermediate values (also caused `delta_psi` in the root/stem solver to be signed, allowing premature exit).
- Added `setLeafHydraulicCapacitance()` overload that accepts an explicit `saturated_specific_water_content` argument.

- Added four new optimization algorithms: L-BFGS and BOBYQA (local, NLopt-backed), SLSQP (local constrained, NLopt-backed), and Adam/AdamW (noise-tolerant, no external dependency). The existing GA, BO, and CMA-ES algorithms are unchanged.
- Added a composable gradient architecture (`GradientSource`, `makeGradientFunction()`, `makeFDGradientSource()`, `makeFDGradient()`) that lets users mix analytic/AD gradients with centered finite differences on a per-parameter basis, plus a `ConstrainedSimulation` overload that returns the objective, constraints, and all sensitivities in a single cached forward pass.
- Added three new `ParameterOptimization::run()` overloads for gradient-based, constrained, and combined-simulation optimization, and an automatic default-algorithm selector chosen from parameter properties and the overload called.
- NLopt 2.10.1 is now bundled under `plugins/parameteroptimization/lib/nlopt-2.10.1/` and built as a static library when CMake is configured; a system-installed NLopt is used as a fallback. If NLopt is unavailable, L-BFGS/BOBYQA/SLSQP are disabled and the remaining algorithms still build. Credit to Kyle Rizzo for developing the kernel-extraction/AD-enabling extensions.

- Added Basler acA2500-20gc (color, GigE) and acA2500-20gm (monochrome, GigE, base + 730 nm and 850 nm bandpass variants) to the camera library.
- Fixed `RayTracingGeometry::validate()` to size shared per-primitive arrays (`primitive_types`, `transform_matrices`, `object_subdivisions`, `twosided_flags`, `solid_fractions`, `object_IDs`, `primitive_IDs`) by `primitive_count` only; bbox geometry is stored in its own `bboxes.UUIDs` / `bboxes.vertices` buffers, so the previous `primitive_count + bbox_count` expectations crashed validation in Debug builds with periodic boundaries enabled.
- Vulkan backend now destroys the `band_map_buffer` during teardown, fixing a Vulkan resource leak on RadiationModel destruction.
- `updateRadiativeProperties()` now writes reflectivity/transmissivity directly into the flat `material_data` buffers via index helpers, eliminating large nested `std::vector` temporaries (previously `O(Nsources * Nprimitives * Nbands * Ncameras)` of nested allocation followed by a redundant `flatten()` copy).
- Camera output now records the actually applied exposure gain and white balance factors on the `RadiationCamera` and writes them to the camera metadata JSON, so downstream tools can recover the linear-radiance scaling that was applied to a saved image.
- Fixed `blendSpectra()` weight-sum check to allow floating-point rounding (tolerance `1e-5`) instead of requiring exact equality to 1, which previously rejected mathematically valid weight combinations.

- Fixed `addPoint()` argument list, which was passing the point size into the wrong parameter slot of `GeometryHandler::addGeometry()` after a recent signature change.

- CMake now checks at configure time that the installed CUDA Toolkit version does not exceed the NVIDIA driver's supported CUDA version, preventing cryptic `OPTIX_ERROR_INVALID_INPUT` failures at runtime due to incompatible device code.

Co-authored-by: Kyle Rizzo <ktrizzo@users.noreply.github.com>

Author: bnbailey-psl

.gitignore

@@ -4,4 +4,5 @@
 .idea/*
 # Compiled outputs 
 samples/*/build/*
-projects/*/build/*
+projects/*/build/*
+plugins/*/build/*

core/include/Context.h

@@ -6753,6 +6753,17 @@ namespace helios {
          */
         void addTimeseriesData(const char *label, float value, const Date &date, const Time &time);
 
+        //! Update the value of an existing timeseries data point
+        /**
+         * \param[in] label Name of timeseries variable (e.g., temperature)
+         * \param[in] date Date vector of the point to update (must match an existing entry exactly, see \ref helios::Date "Date")
+         * \param[in] time Time vector of the point to update (must match an existing entry exactly, see \ref helios::Time "Time")
+         * \param[in] new_value Replacement value for the specified timeseries data point
+         * \note Throws a runtime error if `label' does not exist or if no point with the exact (date, time) is found. To append a new point, use \ref addTimeseriesData() instead.
+         * \ingroup timeseries
+         */
+        void updateTimeseriesData(const char *label, const Date &date, const Time &time, float new_value);
+
         //! Set the Context date and time by providing the index of a timeseries data point
         /**
          * \param[in] label Name of timeseries variable (e.g., temperature)