senSCOPE: Modeling radiative transfer and biochemical processes in mixed canopies combining green and senescent leaves with SCOPE

Semi-arid grasslands and other ecosystems combine green and senescent leaves featuring different biochemical and optical properties, as well as functional traits. Knowing how these properties vary is necessary to understand the functioning of these ecosystems. However, differences between green and senescent leaves are not considered in recent models representing radiative transfer, heat, water and CO2 exchange such as the Soil-Canopy Observation of Photosynthesis and Energy fluxes (SCOPE). Neglecting the contribution of senescent leaves to the optical and thermal signal of vegetation limits the possibilities to use remote sensing information for studying these ecosystems; as well as neglecting their lack of photosynthetic activity increases uncertainty in the representation of ecosystem fluxes. In this manuscript we present senSCOPE as a step towards a more realistic representation of mixed green and senescent canopies. senSCOPE is a modified version of SCOPE model that describes a canopy combining green and senescent leaves with different properties and function. The model relies on the same numerical solutions than SCOPE, but exploits the linear nature of the scattering coefficients to combine optical properties of both types of leaf. Photosynthesis and transpiration only take place in green leaves; and different green and senescent leaf temperatures are used to close the energy balance. Radiative transfer of sun-induced fluorescence (SIF) and absorptance changes induced by the xanthophyll cycle action are also simulated. senSCOPE is evaluated against SCOPE both using synthetic simulations, forward simulations based on observations in a Mediterranean tree-grass ecosystem, and inverting dataset of ground measurements of reflectance factors, SIF, thermal radiance and gross primary production on a heterogeneous and partly senescent Mediterranean grassland. Results show that senSCOPE outputs vary quite linearly with the fraction of green leaf area, whereas SCOPE does not respond linearly to the effective leaf properties, calculated as the weighted average of green and senescent leaf parameters. Inversion results and pattern-oriented model evaluation show that senSCOPE improves the estimation of some parameters, especially chlorophyll content, with respect SCOPE retrievals during the dry season. Nonetheless, inaccurate knowledge of the optical properties of senescent matter still complicates model inversion. senSCOPE brings new opportunities for the monitoring of canopies mixing green and senescent leaves, and for improving the characterization of the optical properties of senescent material.


51
Consistent monitoring of relevant vegetation properties is an essential step towards understanding the 52 response of vegetation function (e.g., photosynthesis, transpiration) to changes in environment. Among 53 others, photosynthetic performance and water use efficiencies are key elements to predict and 54 understand vegetation responses to the climate change scenarios (e.g., elevated atmospheric CO 2 55 concentration, higher temperatures and altered water regimes) (IPCC 2014). However, current Land 56 energy balance model is customized to account for the presence of leaves that neither photosynthesize 163 nor transpire. Since green and senescent leaves feature different radiative balances, a modified RTM t 164 model quantifies thermal emission of each of these two leaf types separately and combines them 165  (Verhoef 1984) to combine the 179 optical properties of green and senescent leaves in an "averaged" leaf. This is simple if leaf angle 180 distribution is assumed the same for both types of leaves. The main advantage of this approach is that it 181 allows representing physiological processes separately in each leaf type. This is important since 182 photosynthesis and transpiration are non-linearly related with radiation and leaf temperature, and 183 therefore might not be adequately represented by a model featuring a unique leaf type characterized by 184 the "averaged parameters" of photosynthetic and non-photosynthetic leaves. 185 senSCOPE requires a larger number of parameters than SCOPE, since two different leaves must be 186 described, as well as their respective area fractions. Alternatives to minimize the number of parameters 187 and simplify the application of the model in inverse problems are presented in Sect. 3.2.2 and discussed 188 later. 189

Radiation Fluxes 190
As SCOPE, senSCOPE relies on SAIL 4-stream theory that can be summarized 197 In this system, x represents the vertical relative height within the canopy (x = 0 for top, x = -1 for 198 bottom), and L represents the Leaf Area Index (also LAI). The remaining variables are the SAIL 199 coefficients defined for first time by Verhoef (1984). k and K are the extinction coefficients in the solar 200 and observation directions, respectively. They depend on the sun-view geometry, LAI and the leaf angle 201 distribution (LAD); and they are therefore independent of leaf optical properties. s, a, σ, s′, w, v and v′ 202 are the scattering coefficients depending on sun-view geometry, canopy structure and leaf optical 203 properties. These coefficients define the relationship between a given incident flux (E 1 ) and a given 204 scattered flux (E 2 ) in the canopy, and they are computed by integrating single-leaf scattering efficiency 205 factors (Q sc ) that represent the analogous relationship for individual leaves. The scattering coefficient 206 (b) corresponding to all the leaves of given zenith inclination angle (θ l ) can be defined as (Verhoef 207 1984): 208 (2) 209 where ′ is the LAI contained in a horizontal layer of the canopy of width dx and φ l is the leaf azimuth 210 angle. 211 As in SCOPE, senSCOPE solves the radiative transfer problem numerically, defining a discrete number 212 of canopy layers and leaf angles. Q sc (E 1 , E 2 ) are defined assuming that individual leaves are Lambertian 213 diffusors of known hemispherical reflectance (ρ) as and hemispherical transmittance (τ). ρ and τ are 214 predicted in SCOPE by Fluspect (Vilfan et al. 2016). For each pair of incident and scattered fluxes, 215 Q sc (E 1 , E 2 ) is defined as a linear combination of ρ and/or τ weighted by spectrally invariant factors 216 determined by the geometry of the leaf, or more specifically, the projection of the leaves with respect to 217 the incident flux (E 1 ) and the downward (-) or upward (+) scattered flux (E 2 ). As proposed by Bach  = green green + �1 − green � senes , (3) 223 = green green + �1 − green � senes , (4) 224 where subscripts "green" and "senes" indicate the type of leaf. Notice that the weighted average of ρ 225 and τ is not equivalent to the factors predicted for a weighted average of the leaf parameters. 226 This approach is suitable to represent the radiative transfer of a canopy of homogeneously mixed green 227 and senescent leaves. In order to represent physiological processes for each leaf type separately, it is 228 necessary differentiating the amount total radiation absorbed by each leaf type, and the 229 photosynthetically active radiation (PAR) absorbed by chlorophyll (E ap,Chl ). SCOPE quantifies E ap,Cab 230 (W m -2 ) using the relative absorption of this pigment respect to the remaining total absorption in the leaf 231 in each spectral band (k Chl,rel ). E ap,Chl is computed for the direct (E ap,Chl,dir ) and the diffuse irradiances 232 (E ap,Chl,dif ) as follows: 233 ap,Chl,dir = green ∫ Chl,rel,green ( ) sun ( )�1 − green ( ) − green ( )� =700 =400 , (5) 234 where λ is the wavelength and k Chl,rel,green is k Chl,rel in the green leaves. These quantities are calculated 236 from the upward and downward fluxes without modifying the transfer of radiation. Since senSCOPE 237 defines senescent leaves as containing no chlorophyll, k Chl,rel = 0 in senescent leaves and for this reason, 238 absorbed PAR used to simulate photosynthesis in sunlit (E ap,Chl,u ) and shaded leaves (E ap,Chl,h ) per total 239 leaf area of the mixed canopy scales with f green . Shaded leaves (subscript 'h') are only illuminated by 240 diffuse light (Eq. 7); whereas Eq. 5 and 6 must be combined to get E ap,Chl in the sunlit leaves (E ap,Chl,u , 241 where f s is a geometric factor accounting for the projection of each leaf towards the sun. 245 Total absorbed radiation is used to compute the radiation budget in the canopy and determines leaf 246 temperature, which has implications for photosynthesis and transpiration, and must therefore be 247 computed separately for each leaf type. Total absorbed radiation is computed by SCOPE similarly as in 248 Eq. 5-8, but integrating the fluxes in the full spectral domain (e.g., 400-50.000 nm): 249 251 a, ,h ( ) = a, ,dif ( ) (11) 252 a, ,u ( , l , l ) = | s ( , l , l )| a, ,dir + a, ,dif ( ) , Where subscript "i" now stands for either 'green' or 'senescent'. 254

Energy balance 255
As in SCOPE, energy balance is closed iteratively by modifying canopy and soil temperatures until the 256 following is met for the soil and for all leaf angles and layers separately: 257 where R n is net radiation, H is latent heat flux, λE is energy heat flux, G is soil heat flux and ε treshold is a 259 predefined threshold for the accepted energy balance closure error (ε treshold ), all in W m -2 . 260 senSCOPE addresses the energy balance separating the processes occurring in green and senescent 261 leaves, where only the first are assumed to photosynthesize and transpire. Therefore, ε ebal is separated 262 into the following elements (Eq. 14): 263 n,green − n,senes − n,soil − green − senes − soil − green − soil − = ebal , (14) 264 where the subscript "soil" refers to soil fluxes, and only green leaves and soil contribute to λE. 265 However, notice that similarly as in SCOPE, the energy balance is separately closed for soil and for all 266 leaf angles, layers and leaf types. 267 In order to compute R n , the contribution of thermal emission must be added to the absorbed radiation 268 calculated in Sect. 2.1. senSCOPE separately represents the temperatures of green and senescent leaves 269 (T c,green , T c,senes , respectively) since they absorb radiation cool down differently. Distinguishing these 270 temperatures has an impact on the calculation of photosynthesis, which is temperature dependent. 271 Consequently, black-body thermal emission (H c ) is different for each leaf type (H c,green , H c,senes ); and the 272 on-sided black-body thermal emission of all leaves is computed as a linear combination of the emission 273 of each leaf type in the canopy: 274 c = green green c,green � c,green � + �1 − green � senes c,senes � c,senes � , where ε is the emissivity, and equals absorptance (1-ρ-τ) according to Kirchhoff's Law. The propagation 276 of emitted radiation by leaves and soil through the canopy is calculated using the averaged layer 277 properties as in the original SCOPE. In order to quantify the net thermal radiation (emitted minus 278 absorbed) (R n,t ) senSCOPE calculates the amount of energy absorbed by each leaf type using their 279 respective emissivity: 280 ,t,green = � − + + − 2 green � green green , where and Eand E + are the diffuse emitted fluxes. R n,t of sunlit and shaded leaves is computed 283 separately. These are energy fluxes per total (senescent plus green) leaf surface area. Therefore, canopy 284 net radiation is computed as the addition of E a and R n,t ; and R n,t = R n,t,green + R n,t,senes without the need to 285 further weight by fraction. 286 Aerodynamic resistances are computed as in SCOPE for the whole mixed canopy, since they depend on 287 meteorology and canopy structure. Consequently water and heat fluxes (H green , H senes and λE green ) in 288 senSCOPE are computed with an identical representation of resistances as in SCOPE, but with leaf 289 temperatures differentiated per leaf type. These fluxes are defined per unit of leaf-type surface, and need 290 to be scaled to the fraction of LAI represented by each leaf type in the mixed canopy. Eventually, 291 senSCOPE iteratively resolves six temperatures: sunlit and shaded green leaves (T c,u,green , T c,h,senes ), 292 sunlit and shaded senescent leaves (T c,u,senes , T c,h,usenes ), and both sunlit and shaded soil (T s,u , T s,h ). 293

Photosynthesis 294
In senSCOPE, only green leaves photosynthesize and transpire. Photosynthesis is driven by the PAR 295 absorbed by chlorophyll (APAR Chl ; which equals E ap,Chl transformed from W m -2 to μmol m -2 s -1 ). The 296 absorbed PAR by chlorophyll in green leaves per unit green leaf area is E ap,Chl,green = E ap,Chl / f green . Other 297 area-based inputs such as maximum carboxylation rate V cmax [μmol m -2 s -1 ], as well as model outputs 298 (e.g., internal CO 2 concentration, C i [μmol m -3 ]) refer to green leaves only. Assimilation (A c ) is 299 therefore initially computed per unit green leaf area. The stomatal conductance (r cw ) as output of the 300 leaf biochemical model is further used to calculate the transpiration of green leaves λE green , also per unit 301 green leaf area. Consequently, both fluxes first calculated per unit green leaf area, and later scaled with 302 f green . 303

Fluorescence 304
SCOPE computes leaf level fluorescence emission using three main elements: incident irradiance in the 305 excitation range 400-750 nm, excitation-fluorescence (E-F) matrices (M(λ e ,λ f ) and M'(λ e ,λ f ) for 306 backwards and forward fluorescence, respectively), and the amplification factors Φ' f which are 307 provided by the biochemical model for sunlit and shaded leaves. E-F matrices represent the excitation 308 of chlorophyll and the radiative transfer of incident and re-emitted radiation inside the leaf (Vilfan et al. 309 2016). In senSCOPE the leaf fluorescence emission is only calculated for green leaves, because for 310 senescent leaves, the E-F matrices equal zero. Then the emission is scaled with f green . 311 Changes in leaf optical properties are computed after photosynthesis, as a function of the rate 320 coefficient for non-photochemical quenching (K n ) provided by the biochemical module. This rate serves 321 as a scaling factor of leaf ρ and τ between two extreme cases of with completely activated and 322 completely deactivated xanthophyll cycle. In senSCOPE, senescent leaves show no carotenoids, no 323 xanthophyll cycle and no related changes in optical properties; for this reason, the extreme cases 324 calculated on the averaged ρ and τ simulate only variations induced by the green leaves. K n is a rate 325 defining the probability of the different fates of photons exciting chlorophyll, therefore, and similarly to 326 Φ' f , it does not require additional correction. Therefore, senSCOPE uses the same radiative transfer 327 functions than SCOPE for the propagation of signals related with the xanthophyll cycle. 328

343
Short wave incoming radiation (R in , W m -2 ), long wave incoming radiation (R li , W m -2 ), air temperature 344 (T a , ºC), atmospheric vapour pressure (e a , hPa), wind speed (u, m s -1 ), air pressure (p, hPa) and soil 345 moisture (SM p , % volume) were provided by a sub-canopy eddy covariance station at 1.6 m height 346 (detailed description of the system can be found in El-Madany et al, (2018) and Perez-Priego et al, 347 (2017)). Vapour pressure deficit (VPD, hPa) was calculated from T a and e a ; also, soil resistance for 348 evaporation from the pore space (r ss , s m -1 ) was estimated as a function of using SM p the model SCOPE  Table 1 shows the ranges of variation generated for each parameter 359 varying in each F veg simulation. Additionally, a smaller dataset was generated modifying only LAI or 360 C ab (and V cmax and C ca as a function of these) to illustrate an example of the response of models to these 361 parameters. Several model outputs and internal parameters were evaluated. Moreover, we also 362 compared the predicted underlying water use efficiency (uWUE, Eq. 19): 363

Study site and datasets 381
The study site is located in the experimental station of Majadas de Tiétar. It is a managed tree-grass 382 ecosystem combining sparse trees (Quercus ilex L. subsp. ballota [Desf.] Samp) and a highly diverse 383 herbaceous cover combining numerous species of three main functional plant forms: grasses, forbs and 384 legumes. The climate is continental Mediterranean so that the grassland shows a strong seasonality 385 initiated by greening phase around April, followed by a dry season that starts between May and June, a 386 second re-greening driven by autumn rains, and a dormant phase during winter (El-Madany et al. 2018). 387 The grassland phenology and functioning strongly responds to light and temperature in spring and to 388 water availability in late spring-summer and in autumn (Luo et al. 2018). Several species grow and 389 senesce at different times, usually, in early spring senescent material remnant from the winter is already 390 present, then new material is also generated during spring, where f green can already be already as low as

senSCOPE and SCOPE. Forward simulation and evaluation 407
Observed/estimated forcing variables and vegetation properties were used to predict fluxes and 408 reflectance factors ±1 day around each flight campaign in each EC tower during daytime. Since no field 409 observations of all the vegetation parameters were available, some of them had to be estimated. When 410 missing, C ab and C ca were estimated from their relationship with N mass observed in the site. Also V cmax 411 was estimated as a function of N mass in the green leaves (N mass,green ) following the relationship in Feng 412 and Dietze (2013), and assumed 45 μmol m -2 s -1 for tree leaves. A constant m parameter of 10 was 413 assumed, N and LAD were assumed 1.5 and spherical, respectively. C s was estimated from the 414 remaining leaf parameters inverting the statistical model described section 3.3.2 and in Appendix A. 415 Then, we evaluated the capability of both models to predict GPP, λE, R n , G, and H comparing SCOPE 419 and senSCOPE predictions with EC fluxes in the site. We also evaluated model performance and 420 structure using predicted fluxes and computing quantities that describe energy partitioning, the 421 evaporative fraction (Eq. 20) 422 where λE and H are the total latent heat sensible heat fluxes, respectively. 424 Emitted irradiance in the TIR (E t ) was compared with net radiometer measurements in the EC towers 425 (CNR4, Kipp and Zonen, Delft, Netherlands); also R were compared with those of the imagery at the 426 time of the overpass. 427

Comparison with SCOPE model. Inversion on observational datasets 428
In order to assess the impact of accounting for senescence material during the estimation of key 429 biophysical (e.g., LAI, C ab ) and functional (e.g., V cmax , m) vegetation parameters, we compared the 430 parameter estimates and posterior predictions resulting from the inversion of both models against real 431 observations in a Mediterranean grassland in the context of a nutrient manipulation experiment with N 432 and P, featuring f green between 0.05-1. In this work, we inverted SCOPE and senSCOPE using the 433 inversion method and approaches proposed in Pacheco- Labrador et al. (2019). 434

Study site and datasets 435
The inversion the models is tested using field observations from the understory grass layer of the site of 436

senSCOPE and SCOPE. Inversion and evaluation 453
We inverted senSCOPE and SCOPE using the same datasets and methodology described for the 454 inversion of SCOPE in Pacheco-Labrador et al. (2019). Observations of R and L t , F 760 and/or GPP were 455 used to estimate LAI, C ab , V cmax , m and other biophysical parameters (Table 2) using an innovative 456 methodology that combined biophysical and functional constraints in two different steps. Three 457 different sets of constraints (inversion schemes) were tested, each combined in the first step of the 458 inversion (Step#1), noon R with noon GPP (I GPP ), noon GPP and F 760 (I GPP-SIF ), or nothing else (I R ). 459 Uncertainties were estimated using a Bayesian approach (Omlin and Reichert 1999). Then, in a second 463 step (Step#2) the guess of V cmax was used as a prior and diel cycles of L t combined with diel GPP (I GPP ), 464 diel GPP and noon F 760 (I GPP-SIF ), or only diel L t (I R ) were used to estimate the functional parameters 465 V cmax and m. f qe was estimated in both steps in the schemes I SIF and I GPP-SIF . Also, pattern-oriented model 466 evaluation was used to assess the results of the different schemes. Unlike the previous work, this time 467 we increased the inversion bounds (Table 2)  notably C ab during the dry season. In all the cases senescent material also was suspected to induce 475 underestimation of LAI. 476 We used the same methodology to invert senSCOPE on the same datasets in order to compare the 477 results provided by both models and to understand the suitability of using senSCOPE in environments 478 featuring large fractions of senescent leaves. However, in the case of senSCOPE, 6 leaf parameters of 479 two different leaf types must be estimated (Table 2). In order limit the number of free parameters in the 480 inversion, we applied the following constraints: We assumed that green leaves presented no senescent 481 pigments (C s = 0) whereas senescent leaves only presented senescent pigments (C ab = C ca = 0). We also 482 assumed that the mesophyll parameter (N) and dry matter content (C dm ), were the same for both types of 483 leaves, whereas that water content (C w ) of green leaves was four times higher than senescent C w (Kidnie 484 et al. 2015). This allowed us reducing the degrees of freedom by 6. We assumed that average leaf 485 parameters (X) could be computed as a linear combination of the parameters of each leaf type (X green and 486 X senes ) as in Eq. 21: 487 is possible in all the cases since at least the value one of them together with f green are known: Either they 492 are equal, 0, or their ratio has been prescribed. senSCOPE includes the additional parameter f green ; in 493 order to reduce equifinality and as well as the number of parameters to estimate we prescribed f green by 494 modelling it as a function of the averaged leaf parameters X using a Neural Network (NN). The NN was 495 trained from a look up table of individual X green and X senes parameters averaged as a function of f green ; no 496 assumptions on N, C w and C dm were made (Appendix A). As a result, the same parameters were 497 estimated in the inversion of SCOPE and senSCOPE. 498 As in Pacheco-Labrador et al. (2019), we used pattern-oriented model evaluation approach to assess the 499 retrieval of functional parameters, which cannot be determined from individual leaf measurements in 500 the highly biodiverse grassland under study. To do so, we assessed the relationship of V cmax and C ab 501 against N mass in the green fraction of the canopy (N mass,green ), and in the case of V cmax it was compared 502 with the relationship published by Feng and Dietze (2013) for grasslands. We also evaluated model 503 performance and structure using not directly predicted fluxes, but variables derived from them such as 504 EF, which describes energy partitioning (Eq. 19). In addition, a more traditional evaluation was also 505 done assessing the goodness of the fit or prediction of model constraints (R, L t , F 760 , and/or GPP) and 506 observed parameters (LAI, f green ). 507

Comparison of results and performance with SCOPE model: Sensitivity analysis. 509
For the F meteo runs, green and senescent leaf properties were kept constant for the different combinations 510 of f green . Fig. 3a,b show the leaf optical properties simulated with senSCOPE and SCOPE, respectively. 511 Accordingly Fig. 3c,d shows the TOC Hemispherical-Directional Reflectance Factors (HDRF) 512 simulated with each model at midday of DoY 139, the timestamp used for F veg runs. As can be seen, 513 senSCOPE predicts spectroradiometric variables that vary more proportionally to f green , whereas SCOPE 514 simulates stronger absorptions, especially in the visible region. This results from allocating all the 515 absorptive substances to a single leaf type. The largest differences between models are found in the red 516 and blue regions, where senescent leaves in senSCOPE increase scattering. We also verified that when 517 f green equals 1 or 0, the output of both models is the same. 518  variables that are strongly controlled by radiative transfer in the optical domain (APAR Chl (Fig. 4c,d), 525 the Photochemical Reflectance Index (PRI, Gamon et al, (1992)), sensitive to activation of the 526 xanthophyll cycle (Fig. 4q,r) and F 760 (Fig. 4s,t)) present a stronger and more linear sensitivity to f green . 527 The same is observed for the water and energy fluxes (λE (Fig. 4e,f) and H (Fig. 4g,h)). Differences for 528 variables related with the radiative transfer of thermal radiance seem to be lower (R n (Fig. 4i,j) and T c 529 ( Fig. 4k,l)). senSCOPE leaves feature a higher absorption of PAR per unit green leaf area, which 530 produces a stronger NPQ activation (K n (Fig. 4m,n)), and a depletion of photosynthetic efficiency 531 around midday (Φ' f (Fig. 4o,p)) for low f green (unlike the other parameters, these are only representative 532 of green leaves). Notice that the example shown is only representative of the meteorological and 533 vegetation properties represented during DoY 139, and the differences shown should not be taken 534 generally. 535 536  the right columns, respectively. As can be seen, both under varying meteorological conditions and 544 varying plant properties, the two models predict the same fluxes when f green = 1, but not always when 545 f green = 0. For f green < 1 SCOPE predicts higher assimilation (A, Fig. 5a,b); but in the case of f green = 0, 546 where SCOPE predicts negative A due to photorespiration, and senSCOPE represents no photosynthetic 547 leaf area. SCOPE also predicts in most of the cases higher R n (Fig. 5c,d) and λE (Fig. 5e,f), and lower H 548 (Fig. 5g,h) and G (Fig. 5i,j).  Differences between predicted fluxes usually maximize when C ab and LAI increase (Fig. S1a-e and S2  556 a-e, respectively). G (and R n ) also present large differences for low values of these parameters and mid 557 f green . In the analysis of the forward runs, the differences observed from F veg simulations are often larger 558 than those F meteo simulations since the variability in the meteorological variables is -in relative terms-559 lower than the variability simulated for the vegetation properties. 560 For each f green level, Fig. 6 presents the distribution of the difference between variables related to leaf 561 function, as predicted by SCOPE and (minus) senSCOPE. Results of F meteo and F veg are shown on the 562 left and the right columns, respectively. Similar to the fluxes, these variables are integrated according to 563 LAI and the probability of each sunlit and shaded leaf angle. APAR Chl (Fig. 6a,b) is equal for both 564 models when the canopy is totally green or senescent. For the rest of the cases SCOPE predicts higher 565 APAR Chl , except some cases when C ab < 10 μg cm -2 (not shown). senSCOPE predicts higher canopy 566 temperature (T c , Fig. 6c,d) than SCOPE; the largest differences are found when C ab is high (Fig. S1g), 567 or when LAI is low (Fig. S2g). Simlarly, uWUE (Fig. 6e,f) is higher for senSCOPE, but unlike T c and 568 most of the variables compared, differences in uWUE are more strongly controlled by meteorological 569 conditions than by vegetation parameters. The largest differences in uWUE are found under cold 570 conditions with VPD < 5 hPa (not shown). senSCOPE presents also higher K n (Fig. 6g,h). Differences 571 between models predictions increase with LAI (Fig. S2i), and decrease with C ab (Fig. S1i). Φ' f (Fig. 5i,j) 572 is most often higher for senSCOPE than for SCOPE, especially if LAI is high and C ab is low (not 573 shown). On the other hand, SCOPE predicts higher Φ' f when LAI is low (Fig. S2j) or when C ab is low 574 and LAI is moderate (Fig. S1j). As expected, both models predict the same values for these variables 575 when f green = 1. 576 577 578   Fig. 7 shows the distribution of some TOC spectroradiometric variables predicted by SCOPE and 584 (minus) senSCOPE for each f green level. Results of the F meteo and the F veg simulations are presented in the 585 left and the right columns, respectively. F 687 (Fig. 7a,b) and F 760 (Fig. 7c,d) are larger for SCOPE in 586 most of the cases the cases; the largest differences are found for low f green and large C ab (Fig. S1k,l) and 587 LAI (Fig. S2k,l). Differences in PRI are negative for Fmeteo, but of both signs for Fveg (Fig. 7e,f). In 588 this case, the influence of vegetation parameters is more complex and less linear than in other variables; 589 since it depends on the combination of C ab and C ca , their ratio and LAI (not shown). A similar analysis 590 carried out on the PRI computed from reflectance factors where the effect of the xanthophyll cycle is 591 not simulated reveals that differences between models rather respond to biophysical properties than to 592 differences in function (not shown). Two more spectral indices responsive to pigments content and 593 canopy structure are also analysed. respectively. For these indices, the absolute difference between models increase as f green decreases, and 597 as C ab and LAI increase (Fig. S1n,o   comparison is done using Total Least Squares (Golub and Loan 1980). In general, senSCOPE achieves 610 higher coefficients of determination (R 2 ) n lower relative root mean squared errors (RRMSE). Both 611 models overestimate high R (Fig. 8a), and GPP (Fig. 8b); but senSCOPE is less deviated. SCOPE 612 overestimates λE and EF, and underestimates H more than senSCOPE. Both models predict R n quite 613 accurately and precisely; but senSCOPE predicts R n , E t and G with slightly larger errors and in some 614 cases lower R 2 than SCOPE. 615   constraints in the different schemes tested; notice that not all the constraints are used to optimize 623 parameters in all the schemes. The relative differences between the statistics of the fit are calculated as 624 (100 · (x senSCOPE -x SCOPE )/ x SCOPE ); where x is the statistic and the respective model is presented in the 625 subscript. R 2 is estimated using Total Least Squares (Golub and Loan 1980), and the relative root mean 626 squared error (RRMSE) and mean average error (MAE) result of the comparison of the 627 observed/predicted values. Posterior uncertainty (σ post ) is estimated according to Omlin and Reichter 628 (1999). The relative differences of R in the visible spectral region (R Vis , Fig. 9a-d) and the near infrared 629 (R NIR , Fig. 9e-h), GPP (Fig. 9i-l), F 760 (Fig. 9m-p) and L t (Fig. 9q-t)  Parameters are evaluated both using field observations and pattern-oriented model evaluation approach. 644 LAI (Fig. 10a-c) and f green (Fig. 10d-f) are compared against observations using Total Least Squares 645 (Golub and Loan 1980). As can be seen, senSCOPE predicts similar LAI values but show higher R 2 and 646 significance. senSCOPE is also capable of providing reasonable estimates of f green , these are often 647 overestimated but still within the bounds of the relationship C ab -f green observed in the site (Fig. S3). 648 N mass,green is used to evaluate V cmax (Fig. 10g-i) and to compare the relationship between both variables 649 with the one reported in the literature for grasslands (Feng and Dietze 2013). Notice that senSCOPE 650 V cmax is provided per unit green leaf area and is thus comparable with SCOPE estimates and the 651 literature data. Results are coherent with those presented in Pacheco-Labrador et al., (2019), I R fails to 652 constrain V cmax , whereas the schemes using GPP provide relationships with N mass,green which are closer 653 to those in the literature. V cmax estimates are very similar for both models in I GPP ; however, the use of 654 F 760 in I GPP-SIF seems to slightly deviate the adjusted logarithmic model from the one fit to the data in 655 Feng and Dietze (2013). Similarly, C ab (per total leaf area) is evaluated against N mass of the whole 656 canopy (Fig. 10j-l)  dual-leaf approach generates averaged "brighter" leaves since not all the absorbent species are located 693 in the same leaf (Fig. 3). This has relevant consequences for the canopy-RTM, especially in those 694 spectral regions where senescent and the rest of the pigments overlap, and therefore for APAR Chl . 695 senSCOPE produces reflectance factors and APAR Chl that close-to-linearly vary with f green ; whereas in 696 the case of SCOPE, these variables vary logarithmically with f green since leaves absorptivity saturate due 697 to the large presence of pigments. This saturation, combined with the fact that R NIR was overestimated 698 during senescence, led to unrealistically high C ab estimates during the dry period when a strong 699 functional constraint -GPP-was used (Pacheco- Labrador et al. 2019). Notice that only the constraint 700 GPP provided robust estimates of functional parameters. In the present work, we repeated the inversion 701 of SCOPE allowing higher C s than in Pacheco-Labrador (2019) since this allowed predicting low R NIR 702 values observed in the site (Martín et al. 2019). This approach improved the fit of R NIR for all inversion 703 schemes during SCOPE inversion (not shown), but did not solve the overestimation of C ab in the most 704 strongly constrained schemes (I GPP and I GPP-SIF , Fig. 10a,b). senSCOPE fitted less precisely the 705 inversion constraints, and in some cases posterior uncertainties increased due to the strong control that 706 f green has on most of the model outputs (Fig. 9). For I R senSCOPE improved the fit of R, but the opposite 707 occurred when aPAR was constrained by GPP, suggesting that the model might not still represent 708 accurately the observed grassland However, senSCOPE led to C ab values more soundly related with 709 N mass than SCOPE during the dry season (schemes I GPP and I GPP-SIF, Fig. 10j,k). 710 The fact that senSCOPE limits photosynthesis and transpiration to the green fraction results in a close-711 to-linear relation between f green on the one hand, and A and λE on the other hand ( Fig. 4 and 5). SCOPE 712 predicts higher assimilation and transpiration unless f green is very low (~0); in that case A is negative 713 while λE is still high. Contrarily, R n and G predictions are similar for both models; also, differences in H 714 are lower than for λE, but still in senSCOPE H varies more linearly with f green than in SCOPE (notice 715 that f green is not a SCOPE parameter, but is used to average leaf parameters). In the forward simulation at 716 ecosystem scale senSCOPE predicted most of the ecosystem fluxes better than SCOPE (Fig. 8). In this 717 case we assumed a fixed value for m, which might be not completely realistic; however additional 718 works at ecosystem scale have shown that senSCOPE can more robustly represent water use efficiency 719 than SCOPE (not shown). In the inversion at plot level, senSCOPE predicted GPP better than SCOPE 720 when used as constraint (Fig. 9k-o). In contrast, EF was predicted more poorly in all the schemes. 721 senSCOPE assumes no transpiration from senescent leaves; however evaporation from their surface 722 might be relevant when these are moisturized by dew or rainfall. Neither SCOPE nor senSCOPE 723 represent that process and their use after such situations might result uncertain. 724 In inversion, both SCOPE and senSCOPE underestimated LAI, while senSCOPE overestimated f green 725 ( Fig. 10a-h). As discussed in Pacheco-Labrador et al, (2019) and Melendo-Vega (2018), the optical 726 properties of dry standing material might not be accurately described by RTM, leading to an 727 overestimation of R NIR, which seems to be counter-weighted in inversion by reducing LAI. In fact, 728 inversion schemes using GPP (I GGP and I GPP-SIF ) improved the estimation of LAI since GPP demands 729 higher APAR Chl in exchange for increasing the fitting error of R NIR (Pacheco-Labrador et al., (2019), this 730 work). In senSCOPE, underestimation of LAI was also compensated also by overestimating f green . These 731 facts suggest that the optical properties of the senescent material and/or the death standing material of 732 this grassland (and likely other ecosystems) are not accurately represented, leading to biased estimates 733 of some of the parameters. In fact, it was necessary increasing the upper bound of C s to be able to 734 predict low R NIR in the dry season. We allowed C s up to 7.5; whereas values up to 5.0 are reported in 735 literature . Too high C s might have led to unrealistic representation of ρ 736 and τ of senescent leaves, very dark in the visible region but also with low R NIR . In some cases SCOPE 737 estimated C s = 7.5, whereas senSCOPE predicted C s < 5 in most of the cases (Fig. S4c). Apart from 738 LAI, C dm and C w , -which are weakly constrained because the spectroradiometric measurements did not 739 include the short wave infrared range (SWIR)-, might have been affected by this problem. SCOPE and 740 senSCOPE estimates of C dm often hit the upper bound stablished from observations in the field. High 741 C dm also serves to reduce R NIR . In contrast, senSCOPE C w estimates are less often saturated; C w has 742 little effect below 970 nm, but influences leaf optical properties in the SWIR. The relationship between 743 N, C dm and C w of green and senescent leaves assumed during inversion might have contributed to 744 increase the uncertainty of the parameter estimates; for example, it has been observed that leaf thickness 745 decreases during senescence (Castro and Sanchez-Azofeifa 2008); whereas other works assign high N 746 values to senescent leaves ). However, a balance between model error and 747 equifinality must be also observed. Site-specific relationships between the parameters of each leaf type 748 or relationships found in global databases could be used in the future to improve the representation of 749 semi-arid canopies. senSCOPE does not include improved calibrated absorption coefficients or 750 refractive indices to more realistically represent senescent leaves and death standing material, but it 751 offers a formally more correct representation of mixed canopies. The model improves the representation 752 of these canopies, which could be used in the future to calibrate or validate specific absorption spectra 753 of senescent material. senSCOPE can also be applied to other canopies, such as crops and forests, which 754 are characterized a senescent stage. Moreover, the approach adopted in senSCOPE could be similarly 755 used to represent other mixed canopies combining plants with different biophysical properties and 756 function, such as C3 and C4 species. An additional problem for the representation of mixed canopies 757 would be the vertical distribution of the senescent material. The impact on the observed R and fluxes is 758 unclear, and further research is needed in this direction. In such studies, senSCOPE could also be 759 extended to other versions of SCOPE, such as mSCOPE (Yang et al. 2017) to describe the vertical 760 distribution of senescent matter. 761 f green is a critical parameter in senSCOPE, it strongly controls RTM and fluxes and increases equifinality 762 of the inverse problem. Thus, the use of prior information about is this variable is strongly 763 recommended during inversion. For this reason, in this work f green was indirectly predicted from leaf 764 parameter estimates using a NN while the model was inverted. The design of this model was critical to 765 achieve acceptable results, and during training C ab (and C ca ) had to be limited to the ranges observed in 766 the study site (up to ~40 μg cm -2 ). During inversion higher C ab values were allowed, but still, C ab -f green 767 estimates stood within or very close to the bounds observed and used to train the NN (Fig. S3) 768 As a result of the combination of changes in RTM and photosynthesis, not only carbon and water 769 fluxes, but also photosynthetic efficiency and downregulation resulted modified (Fig. 6). On one side, 770 senSCOPE tends to predict higher canopy temperatures than SCOPE, especially when f green decreases. 771 Senescent leaves are warmer than green leaves, but senSCOPE green leaves are not necessarily cooler 772 than SCOPE leaves (not shown). Leaf temperature strongly influences photosynthetic efficiency and 773 together with APAR Chl on photosynthesis down-regulation. Fig. 4m,h show how senSCOPE diel cycles 774 of K n reach higher midday values than SCOPE. SCOPE predicts larger variability of K n as a function of 775 f green under conditions of low illumination, whereas senSCOPE K n varies more strongly with f green under 776 high temperature and irradiance conditions (not shown). Non-photochemical quenching has also 777 different effects on the predicted Φ' f . For example, Fig. 4o and their drivers is fundamental to mechanistically interpret these signals; but also the representation of 783 the spectral variables used to obtain information about these processes, such as fluorescence radiance or 784 PRI. Similarly as R, spectral indices vary more linearly with f green in senSCOPE than in SCOPE (e.g., 785 Fig. 4q,r). Unlike other spectroradiometric variables, PRI show no clear differences between models 786 (e.g. distributions of the difference centre around 0). PRI is known to result sensitive to pigments pool, 787 ratio and to LAI (Gamon and Berry 2012; Garbulsky et al. 2011); results of this work also show that this 788 index is also strongly sensitive to the presence of senescent material. The magnitude of SIF emissions is 789 also modified by senSCOPE, which tends to predict less SIF when f green decreases, (Fig. 7a-d). 790 In this study we compare the inversion of SCOPE and senSCOPE using the data and approaches of in 791 Pacheco-Labrador et al, (2019), but allowing for higher values for C s (as well as C dm and C w ). The 792 wider parameter bounds did not change significantly the results obtained with SCOPE, and differences 793 were mainly related to the use of senSCOPE; which improved the estimation of C ab in the dry season. 794 As with SCOPE, SIF (not shown) and R failed to constrain functional parameters (e.g., V cmax ) and LAI; 795 and only inversion schemes relying on GPP provided robust estimates. However with senSCOPE, the 796 schemes relying on SIF reduced their performance respect to SCOPE. I GPP-SIF fitted the inversion 797 constraints more poorly, and could not correct high C ab estimates during senescence as much as I GPP . 798 This might be result from the use of large C s , which suggests further work is needed to more accurately 799 characterize the optical properties of death standing and senescent material. Also for senSCOPE, 800 functional parameters resulted insensitive to I R constraints (partly due to inversion method, see 801 Pacheco-Labrador et al, (2019)). Bayat et al., (2018) inverted SCOPE using R and found troubles to 802 predict low GPP and λE in a grassland during senescence, which was corrected constraining the model 803 with R and TIR radiance to reduce V cmax during this period. Fig. 10i-l compares V cmax estimates of both 804 models; for senSCOPE V cmax is presented respect to green leaf area, whereas in SCOPE, it is presented 805 respect to total leaf area (all considered "green"). As can be seen, when adequately constrained 806 estimates of both models are comparable. In senSCOPE GPP scales with f green , and V cmax (in the green 807 leaves) does not need to decrease to predict low assimilation. 808 senSCOPE is computationally more demanding (around 10% slower) than SCOPE since more 809 processes and calculations are needed, and more iterations are required to close the energy balance 810 (Table S5). However, senSCOPE seems more robust and provides lower energy balance closure error. 811 Since performance of both models is similar for large f green , both models can be alternately used through 812 the season according the presence of senescent material. 813

814
The combination of advanced radiative transfer models with models representing exchanges of matter 815 and energy between vegetation, soil and atmosphere is bringing new opportunities to improve our 816 understanding of ecosystem function from remote observations. For example, the model SCOPE is 817 being used in the last years with this purpose. However, the accuracy with which these models represent 818 reality limits their application; and ecosystem-specific features can bias results and their interpretation. 819 In this context, we present the model senSCOPE; which adapts SCOPE radiative transfer, energy 820 balance, photosynthesis and transpiration in homogeneous canopies with mixed green and dry leaves. 821 The separated representation of green and senescent leaves significantly modifies the simulation of 822 fluxes and spectra signals respect to a model featuring a single leaf with "averaged" properties. 823 senSCOPE reflectance factors, carbon assimilation and water and energy fluxes linearly scale with f green ; 824 it also improves the prediction of these variables in forward simulations as well as the estimation of 825 vegetation parameters, notably C ab , during the dry season. This is significant for the remote sensing of 826 vegetation function of semi-arid ecosystems, and potentially for phenology monitoring. Despite the 827 improvements, results suggest that not only model structure needs to be corrected; a more accurate 828 characterization of the optical properties of senescent material in grasslands is still needed. The use of 829 SCOPE and derived models is growing in the remote sensing community; however, further assessment 830 of their performance to inform about plant function should be tested in different ecosystems. For 831 example, the role of vertical and horizontal heterogeneity is still unclear. Robust evaluation, e.g. 832 pattern-oriented model evaluation approach, would contribute to identify caveats and ecosystem-833 specific features that prevent accurate monitoring of their function; and that therefore, should be also 834 represented. 835